1 VERBAL MORPHOLOGY IN SECOND LANGUAGE SPANISH ACQUISITION: THE ROLES OF DECLARATIVE AND PROCEDURAL MEMORY SYSTEMS By JUAN PABLO RODRGUEZ PRIETO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009
2 2009 Juan Pablo Rodrguez Prieto
3 To my sister Estela, for teaching me to enjoy life and t o keep mo ving forward even in the most adverse situations, and to my grandmother Maura, as I will always remember the happy summers we spent together in Villovieco during my childhood
4 ACKNOWLEDGMENTS My sincere gratitude goes to Dr. Theresa Antes, Dr. Joaquim Camps, Dr. Gillian Lord, and Dr. Wind Cowles for their mentoring, helpful comments and constructive feedback in the preparation of this dissertation. I could not have elected a better team of peers to guide my work I want to thank Dr. Wind Cowl es for the time she spent writ ing the scripts for the psycholinguistic experiments for me. I would like to thank Alicia Rodrguez for her sweet voice, used in the recordings for the materials, and to lex Quintanilla, Herlinda Flores and some anonymous students for t heir feedback in the pilot testing of the materials used in the data collection process. I want to thank the College of Liberal Arts and Sciences from the University of Florida for awarding me a McLaughlin Dissertation Fellowship because its generous suppo rt helped me persuade 60 people to participate in this study and the extra time allowed me to start the data collection earlier. My special gratitude goes to all those learners who participated in the study and attended both data collection sessions. I tha nk my parents Mara Francisca and Ladislao as well as my siblings Estela, Raquel, Javier, and Jos Miguel for their loving encouragement, which motivated me to continue my study and which gave me strength in difficult moments. I would also like to thank th ose friends who always kept asking about the progress of my work and who supported me when I needed them the most: Alicia and Sonia Rodr guez Herlinda Flores, Sory Gallo and Jarod King, Rosa Mar a Castaeda, Barbara Williams, as well as Lizeth Quintanilla and Jos Manuel Rodrguez My most sincere appreciation goes to lex Quintanilla. No one helped me more directly and incessantly in writing this dissertation than him, sharing the burdens, anxieties and pleasures of this project. Without his unconditional love and deep affection, I could not have written this dissertation.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ........................................................................................................... 4 LIST OF TABLES ...................................................................................................................... 9 LIST OF FIGURES .................................................................................................................. 11 ABSTRACT ............................................................................................................................. 13 CHAPTER 1 INTRODUCTION AND THEORETICAL MODELS ........................................................ 15 1.1 Competing Models for Morphological Processing ........................................................ 19 1.1.1 Connectionism and Single Mechanism Models ................................................ 21 1.1.2 Words and Rules Theory and Dual Mechanism Models ................................... 28 1.2 The D eclarative/Procedural (D P ) Model ....................................................................... 33 1. 2.1 The Mental Lexicon and the Mental Grammar .................................................. 33 1.2.2 The Declarative and the Procedural Memory Systems ...................................... 36 1.2.3 The DP Mo del and Inflectional Morphology .................................................... 40 1.2.4 The Neural Basis of the DP Model ................................................................... 42 1.2.5 The DP Model and the Acquisition of Inflectio nal Morphology ........................ 45 1.3 Outline ......................................................................................................................... 48 2 REVIEW OF THE LITERATURE ..................................................................................... 51 2.1 Introduction .................................................................................................................. 51 2.2 Connectionism and S econd L anguage A cquisition (SLA) ............................................. 51 2.3 Words and Rules and SLA ............................................................................................ 57 2.4 The DP Model and SLA ............................................................................................... 63 2.4.1 Previous SLA Studies Using the DP Model ...................................................... 64 2.4.2 Second Language ( L2 ) Morphological Processing and the Brain U nder the DP Model ...................................................................................................... 68 184.108.40.206 E vent r elated brain potential studies of L2 morphological processing ....................................................................................... 69 220.127.116.11 Functional magnetic resonance i maging studies of L2 morphological processing ................................................................ 77 2.4.3 The DP Model and the Cognitive Approach to Language Learning .................. 84 2.5 Additional Work on the Acquisition of Spanish Regular/Irregular Morphology ............ 92 2.6 Summary ...................................................................................................................... 97 3 METHODOLOGY ............................................................................................................. 99 3.1 Introduction .................................................................................................................. 99 3.2 Research Questions .................................................................................................... 102
6 3.3 Participants ................................................................................................................. 104 3.3.1 Participant Selection ....................................................................................... 104 3.3.2 Background Information................................................................................. 105 3.3.3 Recruitment of Participants ............................................................................ 108 3.4 Design and Materials .................................................................................................. 109 3.4.1 Verb Types ..................................................................................................... 109 3.4.2 Verb Frequencies ........................................................................................... 111 3.4.3 Fillers and their Frequencies ........................................................................... 117 3.4.4 Audio Recordings ........................................................................................... 118 3.4.5 Design of the Psycholinguistic Experiments ................................................... 119 3.4.6 Design of the Elicited N arratives .................................................................... 121 3.4.7 Proficiency Exams .......................................................................................... 123 3.4.8 Other Materials .............................................................................................. 124 18.104.22.168 The informed consent form .............................................................. 124 22.214.171.124 Foreign language background form .................................................. 125 126.96.36.199 Instructions for experiments ............................................................. 125 188.8.131.52 Tags for the button box .................................................................... 125 184.108.40.206 Protocol checklist ............................................................................. 126 3.5 Data Collection ........................................................................................................... 127 3.5.1 First Session: Main Data and Background Information ................................... 127 220.127.116.11 Experiment 1 ................................................................................... 131 18.104.22.168 Experiment 2 ................................................................................... 133 22.214.171.124 Task 1 .............................................................................................. 134 126.96.36.199 Task 2 .............................................................................................. 135 3.5.2 Second Session: Proficiency Exams ............................................................... 137 3.6 Data Analysis ............................................................................................................. 138 4 RESULTS ........................................................................................................................ 140 4.1 Introduction ................................................................................................................ 140 4.2 Psycholinguistic Experiments ..................................................................................... 140 4.2.1 Results from Experiment 1 ............................................................................. 140 188.8.131.52 R eaction time (R T ) results ............................................................... 142 184.108.40.206 Accuracy results ............................................................................... 147 220.127.116.11 Error types and participants behaviors ............................................ 154 18.104.22.168.1 Errors associated with regular verb forms ...................................... 157 22.214.171.124.2 Errors associated with irregular verb forms ................................... 159 126.96.36.199.3 Errors associated with stem change verb forms ............................. 162 4. 2.2 Results from Experiment 2 ............................................................................. 165 188.8.131.52 Accuracy results ............................................................................... 165 184.108.40.206 RT results ........................................................................................ 168 4.3 Results from the Production Task ............................................................................... 172 4.3.1 Accuracy Results ............................................................................................ 173 4.3.2 Error Types and Participa nts Behaviors ......................................................... 180 220.127.116.11 Errors associated with regular verb forms ........................................ 180 18.104.22.168 Errors associated with irregular verb forms ...................................... 183 22.214.171.124 Errors associated with stem change verb forms ................................ 189
7 4.4 Methodological Triangulation of Results .................................................................... 191 5 DISCUSSION .................................................................................................................. 197 5.1 Regular Irregular Dissociations and Frequency Effects ............................................... 197 5.1.1 Frequency Effects in the Accuracy Scores ...................................................... 199 5.1.2 Frequency Effects in the RTs .......................................................................... 202 5.1.3 Regular Irregular Dissociations in the Error Types ......................................... 206 5.2 Developmental Patterns in the Acquisition of Verbal Morphology .............................. 213 5.2.1 Improvements in the Accuracy Score s ............................................................ 213 5.2.2 Improvements in the RTs ............................................................................... 219 5.2.3 Changes in the Rates and Types of Errors ....................................................... 224 5.3 The Role of Age of Initial Exposure and Amount of Practice in the L2 ....................... 226 5.3.1 Differences in the Accuracy Scores ................................................................ 227 5.3.2 Differences in the RTs .................................................................................... 228 5.3.3 Differences in Error Types and/or Error Rates ................................................ 229 5.4 Outline of Mai n Findings ............................................................................................ 232 6 CONCLUDING REMARKS ............................................................................................ 235 6.1 Limitations and Future Directions ............................................................................... 236 6.2 Pedagogical Implications and Suggested Classroom Activities ................................... 239 6.2.1 Concentration Game ....................................................................................... 240 6.2.2 Reading Passages with Input Enhancement .................................................... 241 6.2.3 Multiple Choice Activity with Stem Change Preterite Forms ......................... 243 6.2.4 Verb Tree Poster for Patterns of Changes ....................................................... 244 6.2.5 Quiz Show Game ........................................................................................... 245 6.2.6 Correct the Errors ........................................................................................... 246 6.3 Final Remarks ............................................................................................................ 247 APPENDIX A FOREIGN LANGUAGE BACKGROUND FORM .......................................................... 249 B PARTICIPANTS PAST TRAVEL EXPERIENCES ....................................................... 252 C FREQUENCIES FOR THE FILLER ITEMS ................................................................... 253 D INSTRUCTIONS FOR EXPERIMENTS 1 AND 2 .......................................................... 257 E PROTOCOL CHECKLIST .............................................................................................. 259 F PICTURE STORY SEQUENCE 1 ................................................................................... 263 G PICTURE STORY SEQUENCE 2 ................................................................................... 264 H SAMPLE ACTIVITIES FROM THE DELE PROFICIENCY EXAMS ............................ 265
8 I INFORMED CONSENT FORM ...................................................................................... 276 J TAXONOMY USED FOR CODING ERROR TYPES .................................................... 278 K LIST OF TEST ITEMS AND THEIR ELICITED PRETERITE FORM ........................... 281 LIST OF REFERENCES ........................................................................................................ 282 BIOGRAPHICAL SKETCH ................................................................................................... 290
9 LIST OF TABLES Table page 11 Outline of the functional and biological bases of the declarative and procedural memory systems. ........................................................................................................... 43 31 Token count, frequencies per million words from the three corpora, and averaged frequency for regular test items .................................................................................... 114 32 Token count, frequencies per million words from the three corpora, and averaged frequency for irregular test items ................................................................................. 115 33 Token count, frequencies per million words from the three corpora, and averaged frequency for vowel stem change test items ................................................................. 116 34 Practice items for the psycholinguistic experiments ..................................................... 120 41 Independent samples t test results for the reaction times ( RTs ) for the correct responses by all groups of learners in Experiment 1: Comparison between test items of high and low frequency ........................................................................................... 143 42 Accuracy rates on the different verb types of high and low frequency for Experiment 1 for all groups of participants (3,600 tokens) .............................................................. 147 43 Independent samples t test results for accuracy rates by all groups of learners in Experiment 1: Comparison between test items of high and low frequency ................... 149 44 Summary o f behaviors/errors used for the coding of oral data for Experiment 1 and Tasks 1 and 2 ............................................................................................................... 155 45 Summary of participants main behaviors and errors for all groups of learners (N = 60) and by ve rb type in Experiment 1 .......................................................................... 156 46 Independent samples t test results for accuracy rates (as measured by the button pressed) by all groups of learners in Experiment 2: Comparison between test it ems of high and low frequency................................................................................................ 166 47 Independent samples t test results for the RTs for the correct responses by all groups of learners in Experiment 2: Comparison between test items of high and low frequency ..................................................................................................................... 168 48 Accuracy rates on the different verb types of high and low frequency for the Production Task for all groups of participants (1,440 tokens) ....................................... 174 49 Independent samples t test results for accuracy rates by all groups of learners in the Production Task: Comparison between test items of high and low frequency ............... 175
10 410 Summary of participants main errors for all groups of learners (N = 60) and by verb type in Production Task ............................................................................................... 180 411 Raw token counts for the most frequent verb for ms of the high frequency irregular verbs used in Task 1, ranked according to the data from the Corpus del Espaol (CDE): Verb forms, raw token count and usage by the participants .............................. 188
11 LIST OF FIGURES Figure page 11 Basic structure of the parallel distributed processing (PDP) model ................................. 23 12 Generation of regular and irregularly inflected verb forms under the Words and Rules theory .................................................................................................................. 29 13 Types of lexical forms stored in the mental lexicon. ....................................................... 34 14 Types of rule governed behavior found in various aspects of language. ......................... 35 21 Main influences on noticing and components of working memory and long term memory for the Declarative/Procedural ( DP ) model from the Cognitive Approach ........ 91 31 Next tag on top of the green button in the ioLab Systems button box ........................ 126 32 Outline of the two data collection sessions ................................................................... 128 33 Language and Cognition Lab at the University of Florida ............................................ 130 34 Screenshot of PsyScope X ini tial windows .................................................................. 131 35 Same Stem and Different Stem tags on top of the purple and red buttons, respectively, in the ioLab Systems button box.............................................................. 133 41 Screenshot of Audacity software .................................................................................. 141 42 One way analysis of variance ( ANOVA ) with Tukey post hoc test on the reaction times ( RTs ) for high frequency regular verbs com paring the beginning, intermediate and advanced late learners for Experiment 1 ................................................................ 144 43 One way ANOVA with Tukey post hoc test on the RTs for low frequency regular verbs comparing the beginnin g, intermediate and advanced late learners for Experiment 1 ............................................................................................................... 145 44 One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency stem change verbs comparing the beginning, intermediate and advanced learners for Experiment 1 ............................................................................................................... 151 45 One way ANOVA with Tukey post hoc test on the accuracy rates for low frequency stem change verbs comparing the beginning, intermedi ate and advanced learners for Experiment 1 ............................................................................................................... 151 46 One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency regular verbs comparing the beginning, intermediate and adva nced learners for Experiment 1 ............................................................................................................... 152
12 47 One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency irregular verbs comparing the beginning, intermediate and advanced learn ers for Experiment 1 ............................................................................................................... 153 48 Independent samples t test results for accuracy on the different verb types for Experiment 1: Comparison between the advanced and bilingual groups (N = 30) ......... 154 49 Independent samples t test results for RTs on the different verb types for Experiment 2: Comparison between the advanced and bilingual groups (N = 30) ........................... 171 410 One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency irregular verbs comparing the beginning, intermediate and advanced learners for Production Task ........................................................................................................... 176 411 One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency stem change verbs comparing the beginning, intermediate and advanced learners for Production Task ........................................................................................................... 177 412 Independent samples t test results for accuracy on the different verb types for Production Task: Comparison between the advanced and bilingual groups (N = 30) .... 178 413 Outline of statistically significant frequency effects in RTs for all verb types in Experiments 1 and 2 .................................................................................................... 192 414 Outline of statistically significant frequency effects in accuracy rates fo r all verb types in Experiments 1 and 2 as well as in the Production Task ................................... 193 415 Outline of statistically significant developmental changes in RTs and accuracy scores between the three grou ps of late learners in Experiments 1 and 2 as well as in the Production Task ........................................................................................................... 194 416 Outline of statistically significant differences in RTs and accuracy scores between the group of advanced late learners and the bilingual group in Experiments 1 and 2 as well as in the Production Task ..................................................................................... 195 51 Interlanguage processes during second language acquisition ( SLA ) ............................. 222 61 Concentration game for irregular verb forms ............................................................... 241 62 Reading passage with input enhanced .......................................................................... 242 63 Verb tree poster and suggested flashcards .................................................................... 244 64 Samples of Quiz Show Game slides ............................................................................. 245
13 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy VERBAL MORPHOLOGY IN SECOND LANGUAGE SPANISH ACQUISITION: THE ROLES OF DECLARATIVE AND PROCEDURAL MEMORY SYSTEMS By Jua n Pablo Rodrguez Prieto August 2009 Chair: Theresa Antes Major : Romance Languages This study analyzed the oral production and recognition of complex verb forms in Spanish as a second language by 60 learners at varying proficiency levels and with varying ages of initial exposure to the language to investigate the mental mechanisms second language ( L2 ) learners employ when representing and accessing verbal morphology. Participants completed an oral elicitation task, an oral and visual lexical decision tas k, and produced an oral narrative in which the third person singular preterite verb form of an equal number of regular, irregular, and stem change verbs of high and low frequency was tested. Results indicated that f requency effects, when present, were limi ted to the irregular and stem change forms for the most part. The patterns of results in the accuracy scores and the reaction times to the test items generally support dual mechanism models for morphological processing in which regulars are initially memor ized before rules are identified. Then, as learners become more proficient, regulars are computed online by mental rules in the procedural memory, while the declarative memory continues to handle irregularities. The advanced and intermediate groups of lear ners produced regular verbs significantly faster than the beginning group, irrespective of the frequency of the verb forms. These results were not observed for irregular or stem change verbs. The accuracy scores for the irregular and stem change verb forms were
14 significantly higher in the advanced learners when compared to those by the beginning learners, especially for high frequency items. Finally, o verregularizations showed the incorrect attach ment of two mental rules based on verb class, not simply the most frequent regular rule encountered in the inpu t. This is the first study providing support in favor of a shift of reliance from the declarative to the procedural memory system by late L2 learners in the computation of regular verb forms only. Both a me ntal lexicon and a mental grammar seem to explain better the results found in the present study than purely memorization or associative memory effects.
15 CHAPTER 1 INTRODUCTION AND THE ORETICAL MODELS Learners o f Spanish as a second language ( L2 ) often str uggle when they have to provide a correct verb form in Spanish even at advanced proficiency levels This research project aims to uncover some of the mechanisms that L2 learners at varying proficiency levels employ in the production of Spanish preterite v erb forms, and to track developmental gains and patterns cross sectionally, i.e. from the beginning, through the intermediate, and up to the advanced proficien cy levels. The error types these English speaking learners of Spanish make when producing verb f orms incorrectly (e.g., overregularizations, no vowel stem changes, incorrect person and number s uffix, and so on ) will also be considered. This study investigates the frequency with which learners make such errors in order to shed light on the developmental patterns that different groups of L2 learners follow when acquiring verbal morphology in Spanish. Spanish was selected for the present study because it is a language that has more inflectional endings than English and multiple ways to indicate a past m eaning. The main goal of the tasks selected (an oral elicitation task, a cross modal lexical decision task, and an elicited narrative) is to test whether single mechanism or dual mechanism models for the processing of inflectional morphology can better account for the data from L2 learners. Montrul (2004) has claimed that inflectional morphology is one of the most fragile areas of ling uistic development in which L2 learners display considerable variability (p. 25). A quick contrastive analysis between the English and the Spanish verb inflectional systems suffices to illustrate the challenge for L2 learners of Spanish. In English, the verb to be is the most irregula r verb in the language. But the different forms this verb ta kes are limited to : to be am is are being, been was and were With those eight verb forms and the help of auxiliary and modal verbs, anyone can master the most irregular verb in the English language quickly at least from a
16 purely morphological account. T urning to Spanish, if we tak e a regular verb such as amar ( to love ), the full present tense paradigm coupled with the infinitive form are enough to match the number of forms of the English verb to be : (1) amar to love (the infinitive form) (2) amo I love (3) amas you informal love (4) am s you informal love (another version1(5) ama s/he loves / you formal love of you informal) (6) amamos we love (7) amais you informal plural love (8) aman they l ove / you formal plural love Unfortunately for L2 learners, these eight forms are simpl y the beginning of dozens of verb forms they will encounter for every Spanish verb. The list continues with future, preterite, imperfect, present subjunctive, as well as no nfinite and other verb forms. For a speaker of English as the first language (L1), t he learning challenge is evident. However, the difficulty does not stop here, as Spanish contains a number of other irregularities such as suppletive verb forms (e.g., fue s/he went for the verb ir to go), internal vowel changes (e.g., p i di s/he orde red for the verb pe dir to order), and spelling changes due to syllable structure (e.g., bus que that I search for the verb bus ca r to search), among the most frequent ones. The challenge for English is getting the verb f orm and the word order correct Consequently, the contrast between English and Spanish is syntactic encoding for the former and morphological encoding for the latter. Of course, all languages have regular and irregular morphology (Pinker, 1999). Inflected verbs in Spanish usually follo w the prototype: root + thematic vowel + inflectional suffixes. The thematic vowel is used to determine the class to which the verb belongs: 1st conjugation for infinitives ending in ar 2nd conjugation for those ending in er and 3rd conjugation for i r 1 This form is used in all regions of Central America and most of the South A merican Spanish -speaking countries.
17 verbs Each class has a set of regular morphology, e.g. for the third person singular preterite tense of verbs of the 2nd conjugation, f irst the root must be accessed ( beb for beber to drink) and then the suffix is added i A key question for the present study is whether the resulting verb forms, e.g. beb i ( s/he dra nk ), are produced by m ental rules in a similar way as just described, or if they are stored in memory as a new item for retrieval at a later time. In the case of irregular verb forms, given that for the most part they are not predictable by derivation from the root by any obvious process (e.g., puso is the third person singular preterite form of the verb poner to put ), most theoretical models agree that they must be memorized and sto red in memory, unless they follow a subpattern (Seidenberg & Gonnerman, 2000) In fact, it is posited that these verb forms are usually high in frequency because, otherwise, they would have disappeared from the language (Bybee, 1985; Pinker, 1999). The p revious issues are at the center of an ongoing theoretical debate on the nature of the processing and storage of regular and irregular morphology as a window to a better understanding of two language capacities, a mental lexicon and a mental grammar. T he m ain claims for two of the most important theoretical positions are described i n the following sections The main goals and questions explore d in this dissertation relate to the acquisition of a language other than the first one. T he processing and storage of Spanish morphology as an L2 was chosen because its rich inflectional system provides an excellent opportunity to test the main hypotheses concerning the acquisition of regular and irregular morphology while, at the same time, allowing us to extend the d ebate beyond the regular irregular dichotomy on which it has focused for the most part. Spanish vowel stem change verbs in the preterite tense suffer a vowel change in some verb forms, which is usually conceived of as an irregularity to be memorized. Howev er, they also attach the regular infl ectional suffixes. T hese verbs are included in this study as a separate group
18 to test whether L2 participants treat them like new irregularly inflected forms which are stored in memory, or as regular products with an ad di tional internal rule to master. If complex Spanish verb forms are processed by different mental systems with different neural substrates, then we should observe dissociations in behavioral measures such as accuracy scores or reaction times (RTs) based o n the degree of regularity of the verb forms as well as their frequency in the language. This cross sectional study aims at uncovering some of these potential dissociations in the developmental stages of adult late L2 learners. This study also include s a g roup of English/Spanish bilinguals to compare data from two groups of advanced speakers with varying amounts of practice in the L2, a variable which may affect the way they process and/or access complex verb forms. In t he following sections, the main tenet s for each theoretical framework are discussed by using specific examples from one model for each framework. T he Connectionist model (Rumelhart & McClelland, 1986) is briefly described in section 1.1.1. as an example of a single mechanism model and the Wo rds and Rules model (Pinker, 1999) in section 1.1.2. as an example of a dual mechanism model before introd ucing the Declarative/Procedural ( DP ) model (Ullman, 2001a, 2001b)2 2 Ullman (2001a) is the main reference for the Declarative/Procedural model in language acquisition in general, whereas Ullman (2001b) is the model in relation to L2 acquisition. in section 1. 2. The DP model is the theoretical model used to frame the research questions for the present investigation as it is hypothesized in section 1.2. that it is the strongest theoretical framework to account for the data from L2 learners although, as noted below, there is still a need for data from languages such as Spanish to further the evidence for this model. This dissertation attempts to provide such support. At the end of this chapter there is a brief outline for the remaining chapters of the present research study.
19 1.1. Competing Models for Morphological Processing In the study of language acquisition and representation in the brain there is a fundamental distinction between a mental lexicon and a mental grammar. The mental lexicon is a metaphor for the list of memorized arbitrary pairings between sounds and meanings, i .e. a list of the words, bound morphemes, and idiomatic phrases and everything we know about them. The mental grammar on the other hand, is usually referred to as a set of rules (operations and constraints) that specifies how words may be arranged into p ossible and meaningful combinations, whether we are dealing with complex words, phrases, or sentences. There are two primary competing frameworks that describe the issues surrounding the storage and processing of regular and irregular morphology O ne of t hese frameworks cal led the hybrid or dual mechanism framework argues that these two mental abilities, the lexicon and the grammar, are both implicate d in the processing of complex verb forms. The mental lexicon handles the storage of arbitrary pairings for verb forms whose internal structure is opaque and cannot be easily explained by rules. The mental grammar on the other hand, is implicated in the operations necessary to att ach multiple suffixes to roots. T here are other theoretical frameworks thoug h, which predict that all complex words are processed via associative learning mechanisms, irrespective of their internal decomposition or predictability. For these unitary or single mechanism frameworks the frequency of the word and its membership in a family of words with similar inflectional patt erns is of greater importance. Among the best known single mechanism models are the Network model (Bybee, 1995) and the Connectionist model (Rumelhart & McClelland, 1986), while two of the best known dual mech anism models are the Words and Rules model (Pinker, 1999) and the DP model (Ullman, 2001a). Over the last few decades, our unders tanding of how speakers/hearers process inflected forms in real time has increased considerably thanks to a large body of expe rimental research
20 which has used both innovative psycholinguistic techniques and neurolinguistic methods. However, the interpretation of results of morphological processing is still controversial. There is an ongoing debate about the mental representation and production of inflectional morphology using the two broad theoretical frameworks mentioned before i.e., the single mechanism models and the dual mechanism models. The debate s main focus has centered on the representation and processing of regular and irregular English past tense verbal morphology by native speakers of English3 3 A brief summary of the main claims, th eoretical argumentation, and empirical evidence provided by each framework can be found in the sixth v olume of the j ournal Trends in Cognitive Sciences (2002). although some studies occasionally refer to Dutch and German. As a consequence, there is a need to conduct new research to provide additional empirical evidence on the mental r epresentation and processing of complex verb forms to test the main claims of the differen t theoretical models i n this long standing debate, a debate which stems from the very roots of linguistic study: the role of the mental lexico n and grammar. Cognitive processing research within the domain of language has focused on inflectional morphology extensively, especially in the English past tense. But complex words in languages with richer morphology can be used to test whether they are the result of combinatio ns of meanings or simply new memorized lexical items irrespective of their internal structure. Data from L2 learners and from languages with a richer verbal inflectional mor phology system than English are necessary to extend this debate beyond the boundar ies to which it has traditionally been confined. T wo of the most widely used theoretical models in this debate are described in the following sections, along with an examination of how their hypotheses have been applied to the acquisition of L2s
21 1.1.1. Co nnectionism and Single Mechanism Models Connectionism and the use of artificial brains stem from the basic principles of how a human brain actually learns. These principles are outlined by Altmann (1997) in his introductory book to psycholinguistics entitl ed The Ascent of Babel Altmann states that an average adult human brain weighs around 1.3 kilograms and that it contains little else but neurons. Neurons are responsive cells in the nervous system that process and transmit information by electrochemical signaling Th eir range of connections varies from a few hundred to about 100,000 othe r nerve cells. If one neuron is sufficiently stimulated it will stimulate the other neurons to which it is connected. But the key faculty of the human brain is that the connections be tween the neurons are constantly changing, allow ing us to learn from experience. There are three principles at work that underlie neural transmission: (a) neurons send impulses to the other neurons to which they are connected; (b) the connections can be ex citatory or inhibitory, i.e. an impulse makes it more or less likely that another neuron will generate an impulse of its own; and (c) connections can change, i.e. new ones can grow, existing ones can die, and the sensitivity of each connection can be adj usted (Altmann, 1997, p. 2067). W hen it comes to language, there is nothing unique : The meanings which we evoke with the words of our language are simply patterns of neural activity. These patterns reflect the accumulated experience of the contexts in whi ch those words are used, and as such they have gradually changed with those experiences (Altmann, 1997, p. 205) Having these principles in mind, connectionist models explain mental processes in terms of networks of interconnected units which resemble bra in cells in a very rudimentary way. There is no implicit knowledge of rules; instead, Rumelhart & McClelland (1986) claim that lawful behavior and judgments may be produced by a mechanism in which there is no explicit representation of the rule ( p. 217). In fact, regularity is a key distinction between single-
22 mechanism and dual mechanism models, the former rejecting rules for regular morphology. The c onnectionists account of mental rules was summarized by Ellis (2003) in this way: () structural regulari ties of language emerge from learners lifetime analysis of the distributional characteristics of the language input and, thus, that the knowledge of a speaker/hearer cannot be understood as an innate grammar, but rather as a statistical ensemble of langua ge experiences that changes slightly every tim e a new utterance is processed ( p. 6364) The parallel distributed processing (PDP) model employed by Rumelhart & McClelland (1986) is introduced in the remaining paragraphs of this section to illustrate some of the main claims of single mechanism models for inflectional morphology, with an emphasi s on language learning and second language acquisition ( SLA ) In 1986, Rumelhart and McClelland devised a computer model capable of simulating the acquisition of th e past tense by children whose L1 is English. More specifically, they wanted to capture the three stages of acquisition which are usually observable in children learning English: (1) the use of a small set of high frequency verbs, mostly irregular, and no evidence of use of a rule, (2) a larger number of verbs in the past, mostly regular, and evidence of implicit knowledge of a regular rule so that they can produce past tense forms for invented words and incorrect verb forms, which were correct in stage one because of attachment of the implicit rule to either the root or the irregular past tense form, and (3) the coexistence of regular and irregular verb forms, used correctly most of the time. Before carrying out the simulations, the authors selected 506 English verbs from the verbs (out of which there were two regular and eight irregular verbs), 410 medium frequency verbs (334 regular and 76 irregular), and 86 low f requency verbs (72 regular and 14 irregular). The model they designed contained units and connections. The units represented both the input
23 pattern of the root of the verbs and the output pattern or past tense forms corresponding to each input root. The co nnections were modifiable links among units. The heart of the model was a simple pattern associator network which gradually learned the relationships between the input form and the past tense form. During the learning process, the model only received input and compared what it generated internally as the possible past tense form to the expected correct output. When the model generated a wrong answer, it adjusted the strength of the connections between the input and output units to increase the probability o f successful suppliance of the correct past tense form the next time the input pattern was presented. Figure 11 is a reduced version of the pattern associator network with units and connections. Pattern Associator with Modifiable Connections Phonological Phonological representation Input units Output units re presentation of root form (sounds in verb stem) (sounds in past form) of past tense Figure 11. Basic structure of the PDP model. [Adapted from McClelland, J., & Patterson, K. (2002). Rules or connections in past tense infle ctions: What does the evidence rule out? Trends in Cognitive Sciences, 6 (Page 466, Figure 1).] The phonological representation of the root form in the first column, e.g. /k t / to cut is connected to the input units column which contains 460 neuron lik e units (the second column) which can be either on or off. Each unit represents a small stretch of sound that might appear (on) or might not (off) in the root. Each input unit contains three segments (including word boundaries), e.g., the root /k t/ to cu t would be represented as #k k t, and t #. Input units have no idea which root they are representing and similar sounding verbs will obviously share
24 most of the input units that are turned on an d to which they are connected (e.g., English verbs drink an d shrink share most of their input units for the last sounds ) The third column has an identical set of units in that this time they represent the sounds of the past tense form. Each input unit is linked to each output unit for a total of 211,600 connectio ns (460 460). These connections were initially set to 0 and during the learning trials the pattern associator modified the strength of the interconnections via excitatory (an input unit tends to turn on an output unit), neutral (no effect) or inhibitory (an input unit tends to turn off an output unit) connections. When the phonological representation of a root is entered, it sends a signal to every one of the 460 input units turning them on or off. These input units trigger excitatory, neutral or inhibit ory signals to all 460 output units also turning them on or off. With the information of the activated output units and the help of a decoding/binding network ( another mathematical/statistical formula), the model generates a target output which is compared to the actual past tense form it should have generated. By adjusting the values on these thousands of connections a little bit at a time, the model learns to gradually generate the expected past tense form for every root being fed. T he model thus simulat es what happens in our brains, but obviously on an exponentially smaller and more rudimentary scale. In their simulation of the learning of the English past tense forms, Rumelhart & McClelland (1986) first fed their PDP model with their 10 high frequency verbs for 10 training presentations after which the authors claimed that the model produced the past tense forms at a quite good performance rate. Then they fed the network with their 410 medium frequency verbs during 190 training sessions after which th e model exhibited errorless performance on the 420 total verbs. During the last phase, the model was challenged with the 86 low frequency verbs and produced three quarters of the new regular verbs correctly and made overregularization
25 errors (e.g., catche d instead of caught ) for most of the new irregular verbs. Since the model associated properties of a word to properties of another word, it could automatically generalize by similarity. The most remarkable finding was that the model mimicked some of the be haviors observed in children when they are acquiring English, namely : it produced incorrect forms such as gived during some training sessions when the model had previously generated the correct past form gave (U shaped pattern); it produced irregular past forms correctly based on analogy to similar sounding irregular verb s previously learned, e.g. cling clung; it inhibited overregulari zation to a greater extent for irregular verb s from a large family ( feel ) than those from a small family ( blow ); and it i nhibited attaching ed to verbs which end ed in t or d A single computer operation, the pattern associator, was capable of learning and producing regular and irregular past tense forms after intensive training sessions. No rulelike operations were ever in troduced to the model, only repeated input item s Based on these observations McClelland & Patterson (2002) claimed that if an input output relat ionship is fully regular, connectionists computer network simulations can closely approximat e a categorical, symbolic rule. The main argument against the computer gen erated pattern associator model of Rumelhart & McClelland (1986), however, stems from what it did with regular verbs Pinker & Ullman (2002) claimed that Rumelhart & McClelland s simulation fail ed ve ry frequently for regular verbs generating errors which children never make, e.g. sept as the past for to sip or membled as the past for to mail This is the primary difference between singlemechanism and dual mechanism models. These examp les are the r esult of the model s associations and Pinker & Ullman (2002) posited that language cannot be treated as just a collection of regularities in the input that can be approximated by some mec hanism ( p. 474). Instead, regularities are products of the human mind which need to be explained in a different way and that is why Pinker &
26 Ullman disagree with the claims of single mechanism models in relation to regular morphology Additionally, Pinker & Prince (1988) mentioned that Rumelhart & McClellands connectio nist simulations were not capable of handling homonyms, which are words that share the same spelling and pronunciation but have different meanings and, more importantly for the current discussion, they also have different past tense forms. For example, the y cannot handle the differences between to lie (not to tell the truth) vs. to lie (to be in a resting position). Another argument is that t he pattern associators usually fail ed to exclude changes that do not exist in any human language, e.g. reversing all phonemes from a root input as the past tense output, like producing pit as the past tense form for to tip A more recent connectionist model, the Simple Recurrent Network (SRN) by Elman (1990, 1991) used recurrent links to provide networ ks with a dynami c memory. Elman carried out simulation s in whic h hidden units were also linked to context units, which were used to remember the previous internal state of the network so that the subsequent behavior can be shaped by previous responses (Elman, 1990, p. 1 82) That way, the hidden units mapped both an external input and the previous state of some desired output. The network was trained on simple sentences using a limited lexicon ; the task of the network was to predict the order of words in sentences. When E lman analyzed the ways in which the hidden unit representations clustered, he discovered that the network learned to represent words by categorizing them as nouns or verbs, with further subdivisions of nouns as human/nonhuman, animate/inanimate, and the l ike, even though t hese representations were not explicitly taught to the network. Interestingly, Elman (2004) mentioned that the different tokens of a given word type (e.g., the different grammatical cases of the same verb) occupied different regions of me mory space which corresponded to number, inflection, grammatical role (subject vs. object), and the like. For
27 example, John (subject) occupied the same spatial relationship to John (object) as Mary (subject) to Mary (object). This mechanism gave rise to th e discovery of categories based on specific behaviors of classes within those categories. In sum, Elmans (1990) SRN simulation demonstrated that distributed representations used a space which was richly structured, not only for categorical relationships b ut also for type/token distinctions. In a different simulation, Elman (1991) trained a network to discover the regularities which underlie the order of the words in sentences. Using a limited lexicon and grammar, a network was trained with simple sentences as well as with complex multi clausal sentences. Even though words were not marked for number, class, or grammatical role, the network achieved a high level of performan ce in the prediction of the words that followed in the sentence For example, after th e word boy the network predicted a singular verb or the relative pronoun who as the following word whereas it predicted a plural verb or the relative pronoun who after the word boys To p ut it differently, t he network found a mechanism for representing ag reement and exhibited behavior which was highly regular as it learned to respond to contexts which were defined by abstraction. In 1 998, Elman showed that SRNs also generalize d to novel inputs and to novel uses of inputs, by showing that when a network was trained in sentences in which a given noun appeared only in subject position, it could also deal appropriately with novel sentences in which that noun appeared in object position, i.e., in another syntactic context. In sum, not only does input frequency p lay a key role in connectionist simulations for morphological processing, but more recent models also address the notions of class, inflection, number, grammatical role, and the like. In the next section, the main claims on morphological processing by the Words and Rules theory are described The Words and Rules theory is introduced in some detail as an example of a dual mechanism model
28 1.1.2. Words and Rules Theory and Dual Mechanism Models In broad terms, d ual mechanism morphology refers to a family of psycholinguistic models which hold that morphologically complex word forms can be processed both associatively, i.e. through stored full form representations, and by rules that decompose or parse inflected or derived word forms into morphological constitu ents (Clahsen, 2006, p. 2). According to Pinkers (1999) Words and Rules theory, irregular verb forms are stored words, which have nothing in particular when compared to other words in the language, except for the hypothesis that in their lexical entries they contain a grammatical feature such as [past tense]. However, regular verb forms are handled differently, because they can be productively generated by the application of an abstract rule which attaches an inflectional suffix to a verb root; i.e., reg ular verbs can be generated in a similar way to which phrases and sentences are generated. In sum, irregular verbs are words stored in a lexicon, whereas regular verbs are generated by grammar rules (see Figure 1 2). There are some special cases in which r egular verbs are stored (Pinker & Prince, 1994); e.g., when children have not inferred the regular rule, they need to store fully inflected regular verbs to be used in past tense contexts, or when some rule products have a meaning other than the intended m eaning after decomposition (such as the plural version of drink i.e., drinks which means alcoholic beverages). When there is a need to satisfy a given grammatical feature such as [ past tense ] both systems (the lexicon and the grammar) try to compute it from the verb stem as shown in Figure 12. If the lexicon generates an irregularly inflected form, it blocks the rule application, i.e., the attachment of the inflectional suffix to the stem; e.g., puso blocks *poni ( s/he put ). This is usually called the blocking principle.
29 Spanish verb roots : cantar to sing (regular) & pone r to put (irregular) + Grammatical feature which must be expressed overtly: [3 sg. preterite] Lexicon Grammar V suffix Elsewh ere Princ. X cant ar [p reter. ] V V X suffix pone r puso[p reter. ] Blocking Princ. V V V + suffix puso cant Figure 12. Generation of regular and irregularly inflected verb forms under the Words and Rules theory [Adapted from Pinker, S., & Ullman, M. (2002). The past and future of the past tense. Trends in Cognitive Sciences, 6 (Page 457, Figure 1).] On the other hand, t he rule can be applied to any word which is identified as a verb stem or with the feature V which explains why most new verbs in a given language get regular i nflections e.g. fax ed This is usually referred to as the elsewhere principle. These two phenomena can satisfy the demand s for a syntactic/semantic representation when a feature such as [past tense] must be expressed overtly in the language (Pinker & Ullman, 2002) Regular verb forms are predictable and new ones are newly created: downloaded googl e d, and so on. Irregular verbs are idiosyncratic and form a closed list. Pinker (1999) pointed out that in modern English there are about 150 180 ir regular verbs and that there have been no recent additions. According to Pinker (1999), there are not regular vs. irregular verb forms across the board because regularity is perceived with respect to certain inflections. In fact, most present day English past tens e irregular verb forms were regular verb forms that suffered
30 phonological changes from one generation of speakers to the next, e.g., bend + de = bende > bend > bent Other irregular verbs suffered similar changes such as spend, build, send and so on. When at a given moment an irregular verb is not frequent, the link to the memorized irregular form may become weaker and people can easily compute the stem as regular and generate a rule product. This is what happened with doublets, i.e., two words in a langua ge which have the same etymological root but have entered the language through different routes, e.g. dived / dove learnt / learned and so on. It is posited that the regular form in these doublets is also stored in memory because, if not, the irregular form should block the rule application (Pinker & Prince, 1994). The Words and Rules theory claims that memory, from which irregular verbs are retrieved, is not simply a list of unrelated words but is associative. This concept is similar to the pattern associ ator proposed by Rumelhart & McClelland (1986) except that not only are words linked to words but also to substructures of words such as stems, vowels, features, suffixes, and the like. According to Pinker (1999), this explains why people find families of similar irregular verbs easier to store and recall, since their associations have been strengthened. This also accounts for the occasional generalizations of irregular patterns that people produce in response to new similar sounding verbs (Pinker, 1999). I n 1993, Prasada & Pinker carried out three experiments with 20, 22 and 24 native speakers of English, respectively, in which participants were asked the following: (1) to rate the naturalness of regular and irregular past tense forms, (2) to rate how li kel y it was that an item was a past tense form of a given verb, or (3) to produce their past tense form to fill in the blanks in test sentences. The test items were regular and irregular nonce verbs under three conditions: prototypical, intermediate or distan t pseudo verbs from real verb forms. Results indicated that participants were more likely to accept as correct
31 and to produce irregular past forms for novel verbs when they were more similar to the prototype of an irregular family, based on phonological distance to existing forms. But their acceptance of and attachment of regular suffixes to novel regular verbs were independent from the stems phonological structu re (Prasada & Pinker, 1993). The researchers also created three analogy based computer models similar to those developed by Rumelhart & McClelland (1986), but the computer models failed to generalize the regular pattern to nonce verbs for which they had not been trained before. Human participants always obtained higher scores for all types of ver bs, which, coupled with the previous results, lend support for the application of rules in the production of regular verb forms, thus differentiating single mechanism from dual mechanism models. Pinker (1991) predicted that regular and irregular forms sh ould be dissociable from virtually every point of view since regular verbs are handled by a computational component (rules + representations) whereas irregular verbs are handled by an associative memory system (with properties of connectionist models). If irregular verb forms are memorized items, they should be better remembered the more often they are encountered. As predicted, t hese frequency effects have been found for irregular verbs but not for regular verbs : When subjects see verb stems on a screen a nd must utter the past form as quickly as possible, they take significantly less time (16 to 29 msec difference) for irregular verbs with high past frequencies than irregular verbs with low past frequencies (stem frequencies equated), but show no such dif ference for regular verbs ( <2 msec difference) (Pinker, 1991, p. 532) Additionally, Pinker & Prince ( 1994) claimed that low frequency irregular verbs: are more likely to be the less preferred in doublets; are more likely to coexist with regular versions in doublets; are more likely to drop out of a language diachronically (Bybee, 1985); are more likely to be produced incorrectly by adults under time pressure; and are more likely to be overregularized by
32 children. Regulars and irregular verbs should diffe r in similarity effects too. The application of the regular past tense should apply as a default irrespective of similarity to existing regular verbs. In fact, English regular verbs can have any sound pattern: e.g. even though in English no verb ends in a neutral vowel, the following past tense form could be naturally generated: We rhumba d all night (Pinker, 1991). In addition, t he Rumelhart & McClelland (1986) model could not produce an output for many novel regular verbs which did not resemble other re gular verbs in the training set. However, s imilarity effects were found experimentally for irregular verbs. Bybee & Moder (1983) found that their subjects produced splung as the past form of the verb spling 44% of the time (based on phonological similarity to string sling etc.), shunk in response to shink 24% of the time but only 7% sud in response to sid with each previous example resembling existing irregular verbs of prototypical families less Based on their results, they claimed that speakers form categorizations of linguistic objects in the same way that they form categorizations of natural and cultural objects by properties of their form, in this case, by their phonological properties (Bybee & Moder, 1983) In 1999, Gordon & Alegre provided evidence in favor of a weaker version of Pinkers (1991) rule/associative model, in which regular inflections are never stored. Gordon & Alegres (1999) results characterize lexical access for regular verb forms as occurring via either a wholeword route (asso ciative) or a compositional route (rule). Thirty native speakers of English responded to visual stimuli (regular verbs) by deciding whether each stimulus was a word or not. The authors found frequency effects for regularly inflected words. However, when they divided the items in two groups, regular verbs below or above the median frequency of six instances per million words, only items with frequencies above six showed significant effects. The authors explain that the whole word route wins over the composit ional route when the frequency of the
33 regular form is high. The whole word access route is faster unless the item frequency of the complex word is very low (Gordon & Alegre, 1999), and in those cases the compositional route wins. That explains why, for the ir regular verb forms below the median frequency of six, they found no frequency effects. The Words and Rules theory posits that the parts of the brain that handle the storage of memorized words should be implicated in the retrieval of irregular forms while the parts of the brain that handle grammar rules should be implicated in the use of regular forms (Pinker, 1999). Michael Ullmans DP model (2001a) makes stronger pr edictions for regular/irregular forms in relation to different brain structures and its claims are examined in detail in the next section. After introducing the main claims of the DP model mainly in relation to morphological processing, the predictions for the acquisition of verbal morphology by L2 learners are describe d. 1.2. The DP Mo del 1. 2. 1. The Mental Lexicon and the Mental G rammar Ullmans (2001a) DP model begins with the simple premise that language can be defined as a relation between form and meaning. This concept is very similar to the notion of Saussures linguistic sign, whi ch Matthews (1997) outlined as a relation of mutual dependence between a concept that is signified and an acoustic image of the form that signifies it, i.e. the union of an invariant form with an invariant meaning. According to Ullman (2001a), some relations between a particular form and a particular meaning are arbitrary and idiosyncratic and, as a consequence, they need to be memorized. One example is the relation between the Spanish form gato ( cat ) and the meaning cat or, better, the meaning a small domesticated carnivorous mammal with soft fur, a short snout, and retractile claws, as defined by the online Compact Oxford English Dict ionary of Current English ( Soanes & Hawker, 2005). In fact, the set of
34 memorized form meaning pairings can be defined as the mental lexicon (Ullman, 2004), which is a repository of all idiosyncratic word specific information. According to Ullman (2001b, 2004), t his lexicon contains the following elements ( see Figure 13) : (a) all non compositional simple words whos e forms and meanings cannot be derived from each other, such as gato ( cat ); (b) any irregular word specific information, such as the argument structure of a verb (e.g. amar to love takes a direct object in Spanish); (c) any unpredictable forms that a word takes, such as the preterite form fue ( s/he went ) for the verb ir ( to go ); (d) bound morphemes which cannot stand as a word on their own, like the Spanish suffix ano in the noun americano ( American ), which comes from the Spanish noun Am rica ( America ) ; and (e) idiomatic phra ses or complex linguistic structures whose meanings cannot be derived from the meanings of the individual words, like hablar por los codos ( to talk a lot ; literally: to talk out the elbow s ). gato amar fue ano hablar por los codos V, [ ___ NP] or verb ir to go X ano = man V, [ ___ NP [+ past] who comes (a)] [ + sing] from X Figure 13. Types of lexical forms stored in the mental lexicon Besides this mental lexicon, language also consists of regularities which can be captured by rules of grammar (Ullman, 2001a, 2001b, 2004, 2005). These rules specify and/or con strain how lexical forms can combine to make complex representations whose meanings are predictable and can be easily interpreted, even if we have not heard or seen them before. Rule derived forms can be computed in real time and thus do not need to be memorized in the mental lexicon (Ullman, 2001b) because grammatical processing is fast as well as automatic (Ullman, 2004). In fact, this mental knowledge allows us to produce and comprehend an infinite number of complex forms whether they are sentences, phrases, or complex words (Chomsky, 1965) The
35 grammar rules specify (see some examples in Figure 14) : (a) the sequential order of lexical forms, e.g. the definite article el ( the, masc.) p recedes nouns in Spanish as in el gato ( the cat ); (b) the hierarchical relations between lexical forms, e.g. that a verb phrase can contain a noun phrase, as in comi la paella ( s/he ate the paella ); (c) the structure of phrases and sentences, i.e. s yntax; and (d) the internal composition of complex words, i.e. morphology, as in coma ( s/he used to eat ) = com (root of the verb comer to eat ) + a (suffix for the past progressive tense, 3rd person, singular) The learning and use of grammatical k nowledge are generally implicit (Ullman, 2004, 2005), i.e. not available to conscious awareness or access (Fodor, 1983). And it is often argued that aspects of the ability to learn, represent, and compute rules and constraints depend on innately specified mental constructs (Chomsky, 1965), so that the learner is innately equipped with a priori probabilities for all grammars (Pinker, 2004, p. 951). NP VP coma [[[com ] a ] ] DP N pro V [[[root them. vowel] TAM] PN] el N V NP gato comi DP N la N paella Figure 14. Types of rule governed behavior found in various aspects of language These two language capacities, the mental lexicon and the mental grammar, interact in a number of ways (Ullman, 2004). First, as shown in Figure 14, the mental grammar combines lexical forms into complex structures, such as the definite article el ( the, masc.) and the noun gato ( cat ) into a noun phrase (NP) Second, even though certain complex structures with idiosyncratic meanings may be stored in the lexicon, such as the idiom hablar por los codos ( to
36 talk a lot ), their structures generally follow the rules of grammar. In this example, the verb hablar is followed by the optional prepositional phrase por los codos (literally, through the elbows ) whose internal s tructure follows the rules of any prepositional phrase in Spanish, i.e. preposition + NP, as in por la maana ( in the morning ), en la mesa ( on the table ), and the like. Third, although regular complex representations could be computed every time they are used, especially if they are encountered for the first time, in principle they could also be stored in the mental lexicon after being encountered. For example, if a Spanish native speaker tells you: Ayer mi hermano google esa informacin ( Yesterday m y brother google d that information ), the meaning of the new verb form google should be extracted by detaching the regular preterite suffix from a potential verb root and interpreting the remaining material google as the root of a newly created verb i.e., googlear created in reference to the famous online search engine. Once this preterite form has been encountered and computed by the grammar rules for the first time, you could simply store it in the mental lexicon. F inally, a general pattern obser ved in languages is that idiosyncratic, memorized forms and meanings take precedence over general, derivable ones (Ullman, 2004). 1.2. 2. The Declarative and the P ro cedural Memory S ystems The DP model gets its name from the premise that important aspects o f the difference between the mental lexicon and the mental grammar are tied to the distinc tion between two memory systems implicated in other nonlanguage functions : the declarative and the procedural memory systems (Ullman, 2004). Various characteristics of these two memory systems provide important predictions about lang uage because, as we will see in the following paragraphs, they share a number of features w ith the two language capacities just described, i.e. the mental lexicon and the mental grammar (Ullman, 2004).
37 The first memory system is the declarative memory which is at least partially explicit and mainly refers to the conscious recollection of facts and events as Squire (1994) described: This is the kind of memory that is usually meant when the terms memory and remembering are used in ordinary language. Fact and event memory refers to memory for words, scenes, faces, and stories, and it is accessed by conventional tests of recall and recognition. This kind of memory was termed declarat ive to signify that it can be brought to mind and that its content can be declared () ( p. 203) Declarative memory is fast, accessible to conscious recollection, and flexible, i.e., availabl e to multiple response systems (Squire, 1994, p. 214) This m emory system is important in the learning of semantic knowledge or the learning of arbitrary relations (e.g., a watch is a portable device which tells me the time) and also of episodic knowledge or the recollection of personal and contextually bound events (e.g., yesterday I had pizza for dinner ) (Ullman, 2005; Walenski & Ullman, 2005). Pieces of information that are connected arbitrarily are handled by this memory system, which is able to learn new associations fast, e.g. arbitrarily related information c an be learned based on a single stimulus presentation (Ullman, 2004). Knowledge learned in declarative memory can be generally accessed and retrieved consciously (Walenski & Ullman, 2005). The main language function of the declarative memory is to back up knowledge about words, including their sounds and meanings, and other memorized information (Ullman, 2001b). It handles all idiosyncratic word specific information (Walenski & Ullman, 2005), i.e. the lexicon, including complex structured expressions with no transparent meanings such as idioms. Additionally, the result of word formation rules in morphology can also be memorized and stored in the lexicon, e.g. cant ( s/he sang ), alth ough it is ex pected t o be handled by grammar rules in th e procedural mem ory system, which results in the possibility for redundant representations for certain forms (Walenski & Ullman, 2005). Evidence suggests that the lexicon is capable of co mputing a limited amount of associate based information (Pinker & Ullman,
38 2002). For example, new forms such as condujo ( s/he drove ) for the verb conducir ( to drive ) can be computed by generalizing patterns from previously stored arbitrary associations or memory traces, such as produjo ( s/he produced ) for producir ( to produce ); red ujo ( s/he reduced ) for reducir ( to reduce ), and similar forms According to Walenski & Ullman (2005), the reliance on either memorized representations or on computed representations seems to depend on multiple factors such as item specific information (the frequency of use of a specific form), task specific factors (if the language was learned in childhood or after puberty), and subject specific factors (the individuals sex and even her/his current hormonal status). The second memory system is the pr ocedural memory which is involved in the learning of new, and in the control of long established, motor and cognitive skills and habits, especially those involving sequences (Ullman, 2005, p. 146). This memory system () embraces several kinds of memor y and depends on multiple brain systems (Squire, 1994, p. 205). Nondeclarative memory includes information that is acquired during skill learning (motor skills, perceptual skills, and cognitive skills), habit formation, simple classical conditioning (incl uding some kinds of emotional learning), priming, and other knowledge expressed through performance rather than recollection. Experience can cumulate in behavioral change but without affording conscious access to any previous learning epi sodes or to any me mory content (Squire, 1994, p. 205) Nondeclarative memory is nonconscious and less flexible, i.e., it provides limited access to response systems not inv olved in the original learning (Squire, 1994, p.214) Learning by this memory system is gradual an d incremental, not implicated in the learning of novel associations. Procedural knowledge is not available to conscious access and for that reason the system is often referred to as an implicit memory system (Ullman, 2005; Walenski & Ullman, 2005). Once kn owledge is learned in this memory system the procedures apply fast and automatically (Walenski & Ullman, 2005). In the domain of language, this underlying system aids in the acquisition, knowledge, and processing of aspects of grammar, particularly the ru le governed
39 combination of morphemes, words and phrases into more complex and hierarchically structured larger units (Ullman, 2004; Walenski & Ullman, 2005). These two mental systems interact in a number of ways, learning and processing information both cooperatively and competitively so that memory functions can be optimized (Ullman, 2005; Walenski & Ullman, 2005). On the one hand, both systems can work cooperatively in order to learn a given task ; e.g. the declarative system may acquire knowledge firs t due to its fast learning speed and then the procedural system can gradually learn the same or analogous knowledge. In fact, very young children memorize complex forms first in declarative memory before they gradually abstract the grammatical information of rule governed complex forms in procedural memory (Walenski & Ullman, 2005). On the other hand, the competitive interaction can be observed in the seesaw effect (Ullman, 2004) which states that learning in one system inhibits the functionality of the other, and that if there is a dysfunction in the learning of one system the other one will be enhanced. Evidence further suggests that the lexicon blocks grammar, e.g. the access of a memorized irregular past form ( vino s/he came) blocks the grammatic al computation of the erroneous corresponding overregularized past form (* ven i ). In the same way, the availability of a memorized regular past form ( camin s/he walked) blocks its grammatical computation ( camin ). What pushes children to abstract the grammatical information of rule governed verb forms and to move beyond the memorization stage of all verb forms is the economy principle. That way, children do not store each and every verb form when they can compute them online through the application of a mental rule, thus having a more compact memory system. What truly distinguishes the DP model from other traditional dual mechanism theories, such as Pinkers (1999) Words and Rules, as well as from the single mechanism theories, like
40 Rumelhart & McClellands (1986) Connectionism, is the notion of double dissociations (Ullman, 2005). This is the belief that there is one set of links among neurocognitive markers of stored linguistic representations, conceptual semantic knowledge, and declarative memory bra in structures, which is separate from another set of links among neurocognitive markers of grammar, motor and cognitive skills, and procedural memory brain structures. This belief diverges from the Words and Rules model which posits that domainspecific components underlie each language capacity (mental lexicon and grammar). According to the DP model, lexical and grammatical functions are associated with distinct computat ional and neural substrates which are not dedicated exclusively to language, but are d omain independent (Ullman, 2005). The concept of double dissociations also diverge s from the Connectionism model which holds that both language capacities are linked to a single computational mechanism. Instead, the two memory systems just described play analogous roles working both cooperatively and competitively in the computation of language. 1.2.3. The DP Model and Inflectional M orphology The main claims of the DP model focus on inflectional morphology. According to Ullman (2001a), overt morphological transformations are at least of two types. By morphological transformations the author refers to the mapping between the word stem or root and its inflected forms for a specific syntactic configuration. Affixation is the first type of morphological transf ormation, which involves the sequencing of segments, i.e. the sequencing of stems or roots and inflectional suffixes, as in cocin ( s/he cooked ). The second type of morphological transformation involves an arbitrary relation between the base and the in flected form, whether it is the whole base (or suppletion) as in fue ( s/he went ) from the infinitive ir ( to go ), or whether it is the modification of only particular segments, usually phonological stem changes, as in pid i ( s/he ordered ) from the ba se ped ir ( to order ). Consequently, the DP model does not posit
41 a different category of morphological transformations for stem change verb forms, as Ullman (2001a) stated: Thus, in the (presumed) absence of a third component, stem readjustment rules can only be instantiated in the associative memory system and are therefore not mental rules, but only descriptio ns of patterns in the language ( p. 43) As we can see from the previous examples, any given word can undergo one type of these overt changes (like suppletion in fue s/he went) or more (such as affixation and vowel stem change in pidi s/he ordered) A given English word can also undergo none of these transformations, as in hit (bare infinitive) hit (preterite form), but this is not possible in Spanish because finite or inflected verb forms are always different from their corresponding nonfinite base, which indicates the verb class ( ar er or ir ). The DP model proposes that regularly and irregularly inflected verb forms are handled by two di fferent mental systems, the procedural and the declarative systems, respectively. In English, regular and irregular verb forms are matched in complexity because they are all made up of a single word, they are matched in syntax because they are all tensed, and they are also matched in meaning since all of them make reference to a past event (Ullman et al., 1997). But their main difference is that regular verbs are predictable in their internal structure, i.e. verb stem + ed and new ones are being added con stantly, such as fax ed while irregular verbs are unpredictable, such as hit hit bringbrought or sing sang, and they constitute a fixed list. The DP model claims that t he rule system subserves those morphological transformations that involve sequencing operations, such as affixation and compounding, while the me mory system underlies those morphological transformations which do not involve sequencing, including vowe l stem changes (Ullman, 2001a).
42 For a given syntactic configuration, the two systems attem pt to compute an appropriate morphologically complex form from the base. The declarative memory attempts to access a form in associative memory while the procedural memory attempts to compute a rule product. If the declarative memory retrieves an irregular form (such as vino s/he came for the verb venir to come), then it will inhibit or block the computation of the rule product (*ven i incorrect s/he came). But when a memorized form is not successfully retrieved, the rule may apply by default, somet imes even if the end product is incorrect, resulting in overregularization errors such as *trai instead of trajo ( s/he brought ). 1.2.4. The Neural Basis of the DP M odel Ullman and his colleagues have provided converging evi dence that largely supports th e view of the DP model from a wide range of studies of different natures i.e., psycholinguistics, neurology, electrophysiology, developmental studies, and neuroimaging studies (Ullman, 2001a, 2001b, 2001c, 2004, 2005; Ullman et al., 1997). What follows is a summary of the brain structures which have been identified to play an active role either for the declarative or the procedural memory sys tems, as well as some examples. The brain is the human organ for computation. Despite rapid scientific progress, muc h about how the human brain work s remains a mystery. Fairly recent methods of observation such as event related brain potentials or functional brain imaging tell us that brain operations are highly organized, but they do not reveal the activity of individual neurons. So even the most fundamental principles of neural network computation may remain unanswered for some time until they are finally deciphered. Furthermore, our knowledge about how the brain processes language is still rudimentary. If words and rules are the ingredients of language (Pinker, 1999), then we should be able to tell them apart in the brain, i.e. there should be cues that the
43 computation of irregular verb forms depend s more on the system for word memory while the computation of regular verb forms depend s more on the system for rules. According to the DP model (Ullman, 2001a, 2004) there are specific brain regions implicated in particular language functions, as related to inflectional morphology. What follows is a summary of the brain st ructures for the processing of regular vs. irregular morphology. Table 11 gives a su mmary of the functional and biological bases of the declarative and procedural memory systems Table 11. Outline of the functional and biological bases of the declarativ e and procedural memory systems. DECLARATIVE MEMORY NONDECLARATIVE/PROCEDURAL MEMORY EXPLICIT IMPLICIT Specialized for the learning of Arbitrary associations Sequences Learning speed Fast Gradual, incremental Learned knowledge is consciously accessi ble? Typically, but not necessarily No Non language functions Learning and use of fact (semantic) and event (episodic) knowledge Learning and use of motor, cognitive and perceptual skills; habit formation; simple classical conditioning including some kind s of emotional learning; priming; knowledge expressed through performance rather than recollection Language functions Acquisition, representation and processing of lexical knowledge, including memorized complex forms; conceptual semantics Acquisition, rep resentation and processing of complex structures, across syntax, morphology, and possibly phonology and compositional semantics Inflectional morphology Irregular verb forms Regular verb forms Subcortical/Medial brain structures Hippocampus Basal ganglia Neocortical brain structures Mainly temporal lobe regions Mainly frontal lobe regions, especially premotor and associated cortex, including posterior Brocas area [ Adapted from Walenski, M., & Ullman, M. (2005). The science of language. The Linguistic Review, 22 (Page 334, Table 1).]
44 On the one hand, the declarative memory system is subserved by medial temporal lobe regions such as the hippocampus and related structures which are largely connected with temporal and parietal neocortical regions (Ullman, 20 01a, 2001b, 2005). Brodmans areas (BA) 45 and BA47 also play a role in the selection or retrieval of declarative memories while parts of the right cerebellum may underlie the searching for declarative knowledge (Ullman, 2005). On the other hand, the proce dural system is rooted in frontal/basal ganglia structures, with a possible role for inferior parietal regio ns (Ullman, 2001b). This system depends mainly on brain structures in the left hemisphere of the cerebrum (Ullman, 2005). In particular, the basal g anglia and the supplementary motor area seem to play an important role in the processing of sequences (Ullman, 2001b) I n the frontal cortex, in addition to the supplementary motor area, Brocas area ( or BA44) also forms part of the procedural system netwo rk (Ullman, 2005). The DP model is one of the dual mechanism theories which posits that the parts of the brain that handle the storage of memorized words should be implicated in the retrieval of irregular forms, while the parts of the brain that handle gra mmar rules should be implicated in the use of regular forms. This model predicts that it should be possible to find one neural system impair ed by injury/trauma while the oth er remains untouc hed or fully functional. And it makes strong predictions for regul ar/irregular forms in relation to different brain structures For example, a nomia is impairment in word finding, usually associated with damage to temporal/temporo parietal regions in the brain (Pinker & Ullman, 2002). When patients with anomia are asked t o produce past tense forms, they do worse with irregular verbs than with regular verbs Their production shows overregularization errors such as swimmed and the generation of novel regular forms like plammed at high frequencies, along with low frequency o f extending irregular patterns to novel verbs which are phonologically similar to existing irregular
45 verbs like spling. The opposite is true for patients with damage to brain regions associated with grammatical sequences, such as in the case of agrammatism These patients evidence more trouble inflecting regular forms than irregular forms and show great difficulty suffixing novel words. M ost of Ullmans (2001a, 2004) evidence stem s from cases of a variety of language impairments due to damage to one of the neural systems involved in handling either the mental grammar or the mental lexicon41.2. 5. The DP Model and the Acquisition of Inflectional Morphology Chapter 2 includes a s ection with key studies that employed neurolinguistic techniques to assess whether these two mental systems are implicated in the learning and proce ssing of an L2. It is worth mentioning here, though, that other researchers do not agree with Ullmans conclusions. T here is a large body of SLA literature suggesting that initial age of e xposure to the L2 affects final linguistic abilities. Penfield & Roberts (1959) were the first researchers to state that children ar e more efficient L2 learners than adults, based on biological and neurological differences such as cerebral flexibility. Len ne berg (1967) coined the term critical period for the biological predisposition for language acquisition du ring the time span between age two and puberty. The main claim of what is known today as the Crit ical Period Hypothesis was summarized by Birdsong (1 999)5Many researchers e.g., Ullman (2001b), believ e that when someone starts learning an L2 past late childhood or puberty, s/he will not generally acquire the langua ge as well as younge r as follows: () there is a limited developmental period during which it is possible to acquire a language, be it L1 or L2, to normal, nativelike levels. Once this window of opportunity is passed, however, the abil ity to learn language declines (p. 1) 4 For more specific examples, cf. Ullman, 2001a, 2004 5 Birdsong (1999) and the following sourc es are quoted in this section because they are the ones Ullman (2001b) used to talk about a critical period in SLA, even though it must be acknowledged that there are perhaps better resources for that topic, e.g., DeKeyser (2000).
46 learners However, the piece of evidence which is of interest to the claims of the DP model is that grammar is usually affected much more negatively by later ages of exposure than lexical items. Since the learning and use of grammar depends largely upon procedural memory, age of exposure sensitivity in the L2 will affect th e procedural memory system, rather than the declarative memory system. The main piece of evidence for this claim is the study conducted by Johnson & Newport (1989). Their 46 Korean and Chinese learners of English had lived in the United States for three years or more at the time of testing, but they differed in their age of arrival: late arrivals (after age 17) vs. early arrivals (before age 15). Participants listened to 276 English sentences presented on an audiotape, roughly half of which were ungrammatical, and had to provide grammaticality judgments on an answer sheet. Of great interest to the DP model is the resul t for the past tense morphological rule, one of the 12 rule types tested. Results indicated that there was a significant positive correlation between age of arrival and number of errors in this rule (r = .79, p < .01). More interestingly, the researchers found a consistent decline in performance as age increased for early arrivals but a lack of performance differences by age among late arrivals (Johnson & Newport, 1989, p. 79) Overall, these findings support the conclusion that a critical period for language acquisition exten ds its effects to SLA The earlier the initial age of exposure to the L2, the better the accuracy on rules which depend o n the procedural memory system. The performance by late L2 learners was affected to a greater extent, which is why significant differences in accuracy scores in the group of late arrivals were not seen, because their opportunity to achieve nativelike comp etence in grammatical computations was significantly affected or even lost. Again the use of lexical items or L2 vocabulary is less affected by age of initial exposure to the L 2 than grammatical computations In this respect, Singleton (1995) concluded th at :
47 () there is no evidence with regard to second language learning any more than with regard to first language learning that the capacity to acquire new vocabulary disappears at any particular maturational point, or that it necessarily becomes radi cally impaired even in old age (p. 20). Since grammatical computations which rely on procedural memory become relatively difficult to learn while the declarative memory function remains relatively strong, the DP model proposes that later learners of an L2, par ticularly those who begin to learn after late childhood or puberty, will tend to shift to declarative memory for the processing of linguistic forms that are computed grammatically by procedural memory in the L1 (Ullman, 2001b). To put it differently, later learners will depend to a greater extent upon declarative memory in the L2, as a function of age of exposure. However, practice with the L2, in addition to age of exposure, should affect both grammatical proficiency and the degree of dependence on procedu ral memory for grammatical computations. The later the initial age of exposure to the L2 and the less the practice with it, the weaker the dissociations should be (perhaps even to the point of absence) between the declarative/procedural memory systems for the computation of verb forms. According to Ullman (2001b), the greater dependence on the declarative system by later learners will occur in two forms. First of all, forms which are normally compositionally computed by the mental grammar in the L1 may sim ply be stored as memorized forms in the L2 lexicon. Thus regular verb forms like cant ( s/he sang ) or even phrases and sentences, especially those of high frequency, may be stored in the lexicon and accessed from the declarative memory system rather than being computed by the mental grammar. However, the representations for these forms might be structured ( cant s/he sang) rather than unanalyzed ( cant ). And since the declarative memory is associative in nature, it can generalize patterns from memoriz ed forms or linguistic signs to new ones. This ability to generalize patterns to new forms could lead to productivity in the L2. Secondly, some rules may be consciously/explicitly
48 learned (e.g., in a formal instructional context) and applied in the declarative memory, although they need not be. What these rules specify could differ radically from the implicitly learned grammatical rules of the L1 and, more importantly, they do not depend at all upon the procedura l computations. These learned explicit rules may be used by L2 learner s to construct linguistic forms and have an impact in the present study since there is usually an emphasis on stem affix rules in foreign language ( FL ) teaching. The previous predictions for L2 learners led Ullman (2001b, p. 1101) to the following specific empirical hypotheses for SLA with resp ect to morphological processing : (1) L2 speakers should show less separability between regular and irregular verb forms than L1 speakers since it is posited that only for L1 processing do r egular and irregular verb forms distinctively rely on either grammar or lexicon, respectively. (2) Both types of forms (regulars and irregulars) may be learned in and computed over associative memory and, as a consequence, both may show associative memory effects, i .e. frequency effects and phonological neighborhood effects. (3) Both types of forms should tend to pattern with facts and events, and neither with skills and habits. (4) Medial temporal lobe structures as well as neocortical temporal/temporo parietal regions (de clarative/associative memory system) are expected to underlie the use of both types of forms with a general decrease in left lateralization (procedural/grammatical system) in the L2. 1.3. Outline Before discussing the specific research q uestions that guide d this study described at the beginning of Chapter 3, it is necessary first to review what progress has been made so far in the study of the developmental patterns in the acquisition of verbal morphology by L2 learners,
49 which is done in the next chapter Chapter 2 is a review of previous studies dealing with the acquisition of verbal morphology in an L2 under the three theoretical models outlined in the previous sections. It is claimed that research findings are still inconclusive about the processing of i nflectional morphology by L2 learners and that there is a need to conduct new research to evaluate data from L2 learners and from languages with a richer verbal inflectional morphology system than English as a necessary next step to extend the debate beyond the boundaries to which it has traditionally been confined. At the end of the second chapter there is additional information from L2 Spanish studies that include the acquisition of regular/irregular verbal morphology in their design. The main focus of th e chapter is on SLA ; information from L1 acquisition is only included when it provides data that are relevant to the SLA process C hapter 3 covers the methodology of this investigation. Once the research questions are specified, information about the parti cipants, the design of the tasks they completed and the protocol s for data collection are included. Special attention is given to descr ibing the main variables of the present study and how they were measured. At th e end of the chapter, I describe the main statistical tests used to check whether differences between the groups and/or the independent variables reached statistical significance. In Chapter 4 the results from the different tasks are reported Results include, but are not limited to, significant d i fferences in the RTs to the test items based on the frequency and regularity of the verbs the accuracy rates for the test items by each proficiency group, and the main co gnitive strategies and/or error types attested by the participants At the end of th e chapter, the main findings from all tasks are summarized This methodological triangulation of data will be helpful for identifying the main dissociations of results based on frequency and/or regularity
50 of the verbs, which will be used later to support a nd/or reject the main hypotheses under consideration. T he significance of the main findings is discussed in C hapter 5, based on the clai ms of the different theoretical model s, outlined in this chapter, for the acquisition of inflectional morphology by L2 learners. Specific examples are provided from the results of this study to support the main findings. This is the section where explanations and interpretations of the main data are provided bas ed on the predictions of the single mechanism and dual mechani sm model s The sixth and final chapter includes some concluding remarks and contains the following closing sections: (a) the main limitation s of the study (b) future directions for the current state on the debate about inflectional morphology and languag e acquisition and representation by L2 learners at large and (c) some pedagogi cal implications based on the results along with suggested activities for implementing these recommendations in Spanish classes of various levels
51 CHAPTER 2 REVIEW OF THE LIT ERATURE 2.1. Introduction This chapter provides a review of previous studies that tested the main hypotheses from the theoretical models described in Chapter 1, specifically with respect to the acquisition of inflectional morphology by L2 learners. What fo llows is a selection of studies that applied the predictions from the Connectionism model, the Words and Rules model or the DP model to SLA introduced in that same order in the following sections. The main findings of these studies helped in the design of the present investigation as well as in the selection of hypotheses included at the beginning of Chapter 3. In addition, a final se ction of Spanish SLA studies is includ ed in which verb regularity was a variable in the methodology and/or the results, even if the researchers did not use any of the predictions from the three theoretical models considered in this investigation. Regularity is a key variable when evaluating results from L2 learners because the theoretical models have different claims for the acquisition of inflectional morphology based on the regularity of specific verb forms, as described in the previous chapter. 2.2. Connectionism and SLA Connectionism has been employed in a number of simulations for SLA although there has not been much conn ectionist work directly concerned with SLA (Ellis, 2003, p. 87) Previous work has dealt with the acquisition of L2 word order (Gasser, 1990), the constraints of critical periods in language acquisition (Sokolik & Smith, 1992), the effect of input frequenc y and the interplay between source/target languages in processes of SLA (Blackwell & Broeder, 1992), gender processing (Taraban & Kempe, 1999), or the acquisition of casemarking (Kempe & MacWhinney, 1998). In th e following paragraph, the main findings of a study that employed simulations of PDP models in the acquisition of L2 morphology are summarized
52 Ellis & Schmidt (1998) studied the acquisition of L2 morphosyntax using a miniature artificial language, named MAL, under laboratory conditions. MAL consis ted only of 20 words, and the authors controlled their frequency (amount of input delivered, 10 high and 10 low frequency plural markers) and their regularity (10 regular plural markers vs. 10 idiosyncratic affixes). Seven speakers of English first learned the stem forms in blocks of 20 trials until they orally produced all test items with 100% accuracy on two consecutive blocks. Then participants learned the plural of the words in blocks of 80 trials in random order, including singular versions only (similar to the first learning stage, once per item for a total of 20 items) as well as singular and plural versions in the same screen for plural acquisition. Frequency was controlled for so that high frequency plurals appeared very frequently ( five times for each regular plural and five times for each irregular plural for a total of 50 items) while low frequency plurals only appeared scarcely (one time for each regular or irregular plural for a total of 10 items). The presentation of blocks continued until af ter the participant achieved 100% accuracy on all plural forms. Interestingly, Ellis & Schmidt (1998) found that regularity played a role in the acquisition of this artificial language. The regular plurals were acquired better than the irregular verbs and, in fact, in the last four blocks of training, they found no frequency effects in the RTs for regular items whereas the effect of frequency was maintained for the irregular items. Ellis & Schmidt (1998) conducted a new experiment using four connectionist m odels of learning with varying computational capacity and trained them in a similar way to their human participants. Results indicated that frequency played a role in the acquisition of the test items (high frequency items were learned faster) as did regul arity (regular items were learned faster). The computer simulations performed somewhat better than human participants on the regular items and worse on the irregular items but their simulations explained close to 80% of the relevant human data
53 and demonstr ated effects which broadly paralleled those found in humans. These results were interpreted as a direct outcome of a basic associative mechanism or power law of learning, i.e. the idea that the probability of an item occurring at a particular time is a po wer function of its past frequency of occurrence. Ellis & Schmidt c laimed that there is no need for a dual mechanism model because they were able to demonstrate that some aspects of L2 mor phology acquisition were explic able using simple associative learnin g principles. The connectionist simulations duplicated the human rule like behavior, yet there were no rules in the connectionist networks. As a result, their main contribution was to show that rule like behavior does not necessarily imply rulegeneratio n or the existence of mental rules. In addition to this classical example for the acquisition of L2 morphology using a PDP model, and before turning to the dual mechanism models, s ome studies whose main findings are claimed to support the hypotheses put fo rth by single mechanism models are also included in this section (Albright & Hayes, 2003; Bybee & Moder, 1983; Murphy, 2004; Orsolini & Marslen Wilson, 1997). Their designs did not include PDP simulations but their results can be better accounted for by un itary models. They concentrate on languages other than the L1 and their main findings are reported in chronological order. Bybee & Moder (1983) gathered some evidence in favor of a model based solely on analogy for both regular and irregul ar verb forms. They asked 34 native English speakers to write the past tense form of 93 nonce verbs in sentence contexts, as well as for 16 real verbs, presented orally. The main results indicated that participants responses were greatly influenced by the phonological similarity of the nonce verb to a prototype from a particular verb family Regularity was never a factor affecting participants responses and, as a consequence, Bybee &
54 Moder claimed that participants may have processed the nonce roots as wh ole words, i.e. irrespective of their internal decomposition and/or the application of mental rules: The decisions which our subjects made about how to form the past tense of a particular nonce verb may have been based on how the whole resulting word wou ld sound. Evaluations of how the word sounds may be based on compari sons with whole existing words (p. 264) Orsolini & Marslen Wilson (1997) conducted an experiment with 24 native speakers of Italian in which nonce words were included in the study d esign In this experiment, the researchers created an elicitat ion task using nonce verbs that varied with respect to their similarity to actual regular and irregular Italian verbs: high similarity, root similarity or low similarity. Under dual mechanism mo dels, one would only expect phonological similarity effects for irregular verbs because they are the only ones being stored in associative memory. However, Orsolini & Marslen Wilson (1997) found phonological similarity effects for both regular and irregula r nonce verbs, i.e. the less similar the nonce verb was to an actual regular or irregular verb form, the less likely it was to be made regular or irregular. Results were more robust for irregular verbs, but they were still statistically significant for regular verbs too, providing evidence against dissociations between regular and irregular verb forms, and thus favoring a single mechanism model. In 2003, Albright & Hayes carrie d out a similar study to Bybee & Moder (1983) and Orsolini & Marslen Wilson (1997). They conducted two experiments with a list of nonce verbs. In the first experiment they presented the nonce verbs orally to 20 participants who had to rate the stem on how good or n atural it sounded. They also had to give the past tense form fo r the nonce test items. In the second experiment, 24 new participants had to produce the past tense form for the nonce verbs and then they had to listen and rate how good the inflected form, regular or irregular, of the test items sounded in sentence contexts. If regular verbs are gen erated
55 by a rule, the ratings for them should not differ significantly. However, if all irregular verbs are stored in a mental lexicon, nonce verb ratings should differ significantly based on their similarity to existing forms. However, Albright & H ayes found that the ratings for both nonce regular and irregular verbs differed significantly based on their similarity to existing forms, which lends support to a simple associative memory in the processing of inflectional morphology Murphy (2004) tested the claims of the dual system model of Pinker & Prince (1994) in the L2 learning context. The main research question was whether nonnative speakers (NNSs) of English from a variety of L1s (such as Spanish, Croatian, Italian, and other L1s) manifested the same kind of dissociations in performance as native speakers (NSs) of English: one system used as a symbolic default or rule (for regular verbs) that applies when retrieval from memory is missing, and an associative system ( for irregular verbs) that accounts for exemplar based linguistic features. Murphy administered to 20 NS children, 20 NS psychology undergraduates, and 20 adult beginning ESL students a sentence completion task with 30 nonce verbs accompanied by a figure pe rf orming an ambiguous action. Participants were asked to provide the past tense form. There were 10 prototypical nonce verbs that rhymed with real English regular and irregular verbs (e.g., slace plare ). Another 10 intermediate nonce verbs shared the CCV_ or _VCC sequences with real English regular and irregular verbs (e.g., ploab ning ). The last 10 were distant nonce verbs (e.g., ploamph trisp ). Results indicated that all groups of participants treated regular verbs and irregular verbs differently, i.e. they were more likely to suffix a nonce regular verb than a nonce irregular verb. Participants produced more suffixed irregular verbs the further away the nonce verb was from the prototypical English verb. The only exceptions were the intermediate nonce verbs in the NNS group since they were suffixed less often than the prototypical verbs. The effect of similarity was in the opposite direction from the regular nonce
56 verbs because there was a decrease in suffixation in all groups for those that were distan t. Finally, when participants produced a nonsuffixed past tense, most of the time it involved vowel changes rather than no changes at all. Murphy claimed that NNSs used similarity to make past tense generalizations for both regular and irregular verbs. She indicated that any difference between regular and irregular features could be a direct result of the increased frequency and amount of exposure that learners had with the regular features (Ellis & Schmidts power law of practice, 1998) and that there are a number of indications that a single set of associative learning mechanisms could account for the development of inflectional morphology: a single route associative explanation seems to be able to capture the phenomena manifested in both the L1 and L2 l earning contexts (Murphy, 2004, p. 456). As described, all these previous studies (Albright & Hayes, 2003; Bybee & Moder, 1983; Ellis & Schmidt, 1998; Murphy, 2004; Orsolini & Marslen Wilson, 1997) used nonce verbs or an artificial language to test the ma in claims of their research questions. One of the biggest challenges for these types of experiments is to create items that u nequivocally belong to a single verb type. For example, Murphy (2004, p. 442) chose brip as a prototypical regular verb whereas bli p was selected as a distant irregular. The question here is on what grounds do these nonce verbs really belong to different verb type s, especially as perceived by a NNS of English. The way regular verb forms are processed is what sets apart singlemechani sm from dual mechanism models and the results of these studies provided evidence against mental rules for regular forms. Instead, purely associative learning mechanisms influenced by features of the input such as frequency and phonological similarity to f amily resemblance seem to be the best predictor for the pattern of results accumulated in these studies.
57 In the following section, SLA studies under the Words and Rules theory are introduced. In a dual mechanism model, regularity plays a role in the p roc essing of complex verb forms, and the Words and Rules theory claim s that regular and irregular verb forms are processed differently because irregular verbs are stored as fully inflected words but regular verbs can be computed online by the application of grammar rules. 2.3. Words and Rules and SLA A ccording to the Words and Rules theory, overregularizations of irregular verbs in the acquisition of L1 by children around age three are not the result of children being overwhelmed by an increasing number of reg ular verbs, as connectionists would suggest. Instead, they accompany the appearance of regular tense marking (Pinker, 1991) ; i.e. the child starts using the regular suffix for items which were previously produced as bare infinitives (in past tense context s) and, in the case of irregular verbs, the occasional overregularization errors are the result of rule application to verbs in sentences which require a past tense form but for which the child failed to access the irregular form. This is a good example of U shaped behavior; i.e. if we track the accuracy rates for irregular verbs by children over time, the shape of the graph will look like a U, starting high, then decreasing for a period of time, and then increasing again to the starting point or even higher. The main argument as posited by Pinker (1998), is as follows: Before the first error, children leave regular verbs unmarked most of the time (e.g., Yesterday we walk); then there is a transitional phase in which the child begins to mark these verbs mo st of the time (e.g., Yesterday we walked ). It is in this transitional phase that the first overregularization of an irregular form is produced (p. 236) In fact, at the transitional stage at which children mak e their first overregularization of an irregu lar, irregular verbs are produced with an accuracy rate of about 95%. Clahsen et al. (2002) found similar results in the spontaneous speech of Spanish children. They generalized regular patterns to irregular items but the opposite was never the case. Addit ionally, children started to
58 overregularize after a period of correct performance; these errors decreased as children got older. These overregularization rates followed the U shaped developmental curve Pinker (1998, 1999) previously discussed. These result s indicate a sharp regular/irregular distinction in the inflectional errors made by children, which are compatible with the difference posited by this dual mechanism model between rulebased and memory based representations for morphologically complex word s (Clahsen et al., 2002, p. 618). This error pattern can also be accounted for by singlemechanism mod els, as described in section 1.1.1 However, t he Words and Rules model as described by Pi nker (1999) did not account for the linguistic systems of L2 spe akers. Yaden (2004) pointed out: The main question of Pinkers approach, whether native speakers use a dual mechanism or a unitary approach for processing and storing morphologically complex words, is unanswer ed for L2 speakers in his book (p. 52) In f act, t here is a limited number of studies testing the main claims of the dual mechanism models using languages with a richer verbal inflectional system than English. Clahsen (2006) noted t wo factors in the English language that favor dual mechanism models for inflectional morphology: the frequency of the regular rule is higher in English and there is only one suffix attached to regular verbs, i.e. the ed suffix. Yaden (2003) specifically mentioned the case for Romance languages, as well as t he dearth of w or k in the area of L2 speakers. As a consequence, her research focuses on the acquisition of Spanish regular/irregular morphology by L2 learners (Yaden, 2003, 2007). In the rem a inder of this section the main findings of key studies dealing with the acqui sition of L2 inflectional morphology are described (Beck, 1997; Lalleman et al., 1997; Leung, 2006; Wolfram, 1985; Yaden, 2003, 2007) particularly those results which are
59 compatible with the dual mechanism models and the Wor ds and Rules theory They are d iscussed in chronological order. Wolfram (1985) interviewed 16 Vietnamese participants, learners of English and residents in the US from one to seven years who had no kn owledge of English prior to their arrival in the US He found that past tense markin g was favored with irregular forms in the earlier stages of L2 learning, mainly because of the saliency of the irregular verb forms. He also found frequency effects for the irregular verbs, with frequently occurring irregular verb forms being marked for tense more often than their infrequent counterparts. These data seem to support the dual mechanism model outlined before and also what Wolfram ( 1985, p. 247) called the principle of saliency which states that the more distant phonetically the past tense irr egular form is from the bare infinitive, the more likely it w ill be marked for tense, as a result of the ease with which it is noticed in the input Beck (1997) measured the reaction times for producing regular and irregular verbal morphology by NSs and NN Ss of English. In the last of a series of experiments that Beck (1997) carried out, her 32 NNS participants were asked to orally produce the past tense form of a given verb. She used regular verbs as well as nonce verbs that did not resemble any other word in the language. The most interesting finding was that there were no frequency effects for the regular test items. Beck (1997) concluded that her NNSs had been successful at internalizing a rule for generating inflectional morphology but, given their grea ter difficulty in terms of the overall grammatical correctness of the items as compared to the NSs, she hypothesized that her NNSs had what she called a Type II deficit: a problem with feature checking or with the ability to allow for verb raising in the L 2. Dual mechanism models posit that there should not be frequency effects for regular verb forms because the computation of the regular suffix and its attachment to
60 the verb root can be learned by L2 learners, provided that they have a high prof iciency in the target language. T he participants in her study had either scored 530 on the TOEFL exam or passed a universitys English exam for faculty whose L1 is not English, in addition to showing a range of length of residence in an Englishspeaking country from 5 months to 17,6 years, further attesting t o their overall proficiency. Lalleman et al. (1997) addressed the question of whether advanced L2 learners of Dutch employ two distinct processes (lexical lookup vs. rule application) when producing regular and ir regular past verb forms. The researchers presented their participants with Dutch v erbs in the infinitive form and asked them to produce their past tense as qui ckly and accurately as possible. They had 64 test items, 32 regular and 32 irregular, further divided into even numbers of high and low frequency items per group. They found frequency effects on the produc tion time of irregu lar verbs only, which favors the Words and Rules model. An error analysis showed that the most frequent error type was overregularization of irregular verbs. Lalleman et al. (1997) concluded that their advanced L2 learners of Dutch had been able to acquire the Dutch past tense rule since they applied this rule during o nline language production. Yaden (2003) was interested in how NNS s of Spanish process regular and irregular morphology, to test the dual system models in contrast to the unitary accounts. She used 15 intermediate/advanced L2 Spanish learners who began acquiring Spanish beyond the critical period. She used a sample conta ining regular and irregular verbs from all three verb classes, of high and low frequency, appearing in their infinitive forms on the computer screen. Participants had to orally produce the 1st person singular form of the present tense. Her hypothesis was t hat there would be no difference in RTs between high and low frequency forms for regular verbs but that t here would be for irregular s. She found no significant differences for irregular verbs for
61 NNSs but she found a statistically significant difference fo r the regular verbs the opposite results of what she predicted. Participants regularized the low frequency irregular verbs over 54% of the time but with longer RTs than for the regular verbs and the irregular verb forms conjugated correctly. Yaden suggest ed that these results could mean that they had a sense that these verbs might be irregular, and so they searched their mental lexicon longer before applying the default rule. In contra st, there were only six cases out of 225 where a NNS applied the regular pattern to a high frequency irregular verb. There were several instances of overirregularizations, but they were few and far between (4/7 instances). In this study, frequency played a role for both regular and irregular verbs. However, it was considered t hat the results supported the dual mechanism models because if all irregular verbs were stored in memory but the form was so infrequent that it could not be accessed for production, then the default rule of adding o (1st sing ular present) to the root took place. Leung (2006) tested the representation of Spanish inflectional verbal morphology by 10 Chinese and nine English learners of Spanish under the dual mechanism model. This model suggests that morphologically regular verbs are rulebased products w hereas irregular verbs are listed individually in the mental lexicon. Consequently, during interlanguage ( IL ) development we may find regularization of irregular forms but not irregularization of regular forms. Leung (2006) gave the participants an elicite d written production task with 60 regular or irregular based nonce verbs for which they had to provide the corresponding past participle form in a sentence context. For regular based nonce verbs there were very low rates of irregularization while for irre gular based nonce verbs there were high rates of regularization. The regularization rates were claimed to be related to the fact that the target irregular forms had not yet been acquired and were thus nonexistent in the mental lexicon of the learne rs for retrieval or analogy.
62 In 2007, Yaden conducted another research study with 13 L2 learners of Spanish6In sum, there is cumulative evidence that the time to react to irregular verb forms by L2 learners is affected by the frequency of the irregular items, while these frequ ency effects were not usually observed in the group of regular verb forms. W hen retrieval of an irregular verb form fails especially due to its low frequency, and if L2 learners have internalized a mental rule in who had completed at least two years of study of Spanish. Participants were presented with the 3rd person singular present form on a computer screen and h ad to produce the 1st person singular present form orally as fast as possible. Test items included regular and irregular verb forms of both high and low frequency. Again, her data showed that L2 speakers were significantly faster at producing a high frequ ency than a low frequency regular verb. However, participants took about the same amount of time for irregular verbs of high and low frequency, even when she considered stem change verb forms separately. She hypothesized that frequency played a role for al l test items bec ause L2 participants searched their mental lexicon for a cor rect memorized form even if the unknown verb was regular. This initial check slows down the processing for unfamiliar items before applying the regular rule. The lack of frequency effects for irregular verb forms was explained as a result of the low accuracy rates for low frequency irregular verbs Again, the main error type was overregularization of irregular verbs, especially those of low frequency. There were only a few ins tances of overirregularization. These accuracy patterns support dual mechanism models in that when participants do not retrieve a fully inflected irregular form from memory, they can add the regular rule as a default, resulting in overregularization errors. Low frequency items are affected to a greater extent due to weaker memory traces for them. 6 She also included data from L1 speakers but in this section only the data for the L2 speakers are reported
63 their ILs, then overregularization errors occur. The online attachment of the regular suffix to the stimuli stem can also be explained as the way L2 learners can test their developing hypotheses for low frequency items since an error will probably be followed by correct suppliance from the interlo cutor. If the L2 learner notices the error, this may trigger a restructuring of the L2 learners IL grammar What is interesting is that even though language learning is ultimately viewed as a matter of change in an individuals internal mental state (Doug hty & Long, 2003), these SLA studies did not account for changes over a period of time, not even cross linguistically or among different proficiency groups of L2 learners as they were all limited to a single proficiency group. The following section descr ibes research results from L2 studies under the DP model, which is used to frame the research questions of the present study as it is the theoretical model hypothesized to have the strongest predictive power for the study of the processing of verbal morphology in SLA The above studies in support of the Words and Rules theory are also compatible with the main cla ims of the DP model and their main findings should therefore also be taken into account when discussing new SLA studies under t he DP model. 2.4. Th e DP Model and SLA This section review s key studies whose main find ings support the claims for SLA under the DP model. The main evidence in favor of the DP model for SLA (Ullman, 2001b) is that both types of verb forms are usually learned over associative memory at earlier stages of acquisition and that the greater the practice with the L2 the greater the probability for the acquisition of L2 rules and the greater the separability between regular and irregular verb forms in terms of frequency effects, nei ghborhood effects, and the like. For the most part, the following studies : Birdsong & Flege, 2001; Brovetto & Ullman, 2001, 2005; Flege et al., 1999; Rodrguez, in press, support these claims P revious work on L2 psycholinguistic and neurolinguistic eviden ce, i.e. L2 RT studies and L2 morphological processing and the brain is also discussed in some detail
64 because the DP model emphasizes that two mental systems are implicated in the processing of regular and irregular verb forms Finally, it is proposed th at the main findings for SLA studies using the DP model can contribute to the Cognitive Approach (Skehan, 1998) to language learning because both theoretical constructs share key features that allow for some comp arability of research findings. 2.4.1. Previ ous SLA Studies Using the DP M odel Flege et al. (1999) conducted a study to evaluat e the Critical Period Hypothesis in SLA. They looked at the end state of 240 Kore a n learners of L2 English in relation to phonology and morphosyntax. Participants arrived i n the US between the ages of one and 23 years and had lived in the country for at least eight years They were assigned to one of 10 groups based on their age of arrival (AOA) The researchers assessed participants knowledge of English morphosyntax using most of the sentences from Johnson & Newports test (1989 reviewed in Chapter 2 ). Participants listened to 144 test sentences and decide d if they were correct in English. From this grammaticality judgment test, Flege et al. isolated a subset of 44 items w hich represented rule based and irregular forms, e.g. the rule based set included sentences involving case or number assignment on nouns and person or tense markers on verbs, such as Sandy fill a jar with cookies last night (incorrect) or A snake bit her on the leg (correct). In the lexical set there were irregular verbs marked for past tense, plural noun s and lexical idiosyncrasies such as choice and placement of particles and prepositions, e.g. The horse jumped the fence over yesterday (incorrect) or Th e policeman was talking to a woman (correct). Accuracy rates indicated that late learners with AOA greater than 12 obtained significantly lower scores for the lexically based than rule based sentences. Flege et al. (1999) argued that if their Korean partic ipants were able to deduce a simple rule with widespread application such as add ed to form the past tense, they were able to identify it with greater accuracy than those items whose learning was based on
65 associative or probabilistic representations. Si nce late learners had r eceived less English input, the limited frequency of verb forms in the input contributed to the greater difficulty for irregular forms only. The participants accuracy figures revealed a clear dissociation in performance as a functio n of age, bu t in the unexpected direction based on the claims of the DP model. In 2001, Birdsong & Flege wanted to replicate the results from Flege et al.s (1999) study by reducing the test items to regular/irregular forms only. More specifically, they w anted to check if input frequency was a factor in the knowledge of irregular, but not regular forms (Beck, 1997). They recrui ted 30 Spanish and 30 Korean natives, L2 learners of English, distributed over three groups based on their AOA in the US. Partici pants completed a multiplechoice judgment task with 80 items of regular vs. irregular inflection, high vs. low stem frequency, and noun plural vs. verb past tense morphology. For example: Yesterday the little girl (a. swim / b. swam / c. swimmed / d. swim s / e. swammed ) for the first time Birdsong & Flege (2001) found frequency effects to be significantly higher for irregular items than for regular ones. They also found both a decrease in accuracy and an increase in RT with increasing AOA, which was more pronounced for irregular verbs than for regular verbs In addition, no significant age related declines were observed for the regular items. When they related their results to Ullman et al.s (1997) DP model, Birds ong & Flege claimed that they found eviden ce that the declarative system may be more susceptible to aging effects than the procedural system. Brovetto & Ullman (2001), as cited in Birdsong (2004), found frequency effects for both regular and irregular verb forms in a study of oral production of r egular and irregular English past tense forms by 32 Spanish and 32 Chinese natives with a minimum of three years of residence in the US and with AOA of 17 years or greater. These results (Birdsong & Flege, 2001; Brovetto & Ullman, 2001) taken together support Ullmans (2001b) DP model for L2
66 learners in that, for learners at age stages leading up to the end state, most target language forms are bits of idiosyncratic information stored in declarative memory A s a consequence, there are observed frequency eff ects for both regular and irregular verb forms (Brovetto & Ullman, 2001). Brovetto & Ullman (2005) conducted two experiments in which 33 adult native Spanish speakers were asked to provide inflected forms of verbs in appropriate sentence contexts in Spani sh There were 80 nonce verbs, 40 ending in ar and 40 ending in er 40 containing class I rhymes and 40 with class II rhymes. Rhyme was defined as the identity in sound of an accented vowel in a word and all consonantal and vowel sounds following it. In Spanish, certain rhymes are more present in class I verbs ( ar verbs) while others typically appear in class II and III verbs ( e r and i r verbs) ; e.g. rhyme I ant usually appears in verbs of the 1st conjugation class such as cantar to sing, plantar to plant, and the like. There were an equal number of novel verbs for these groups based on rhyme and class: Rhyme I Rhyme II Verb class I vantar corbar Verb class II balter tonder Results indicated that conjugational class was the best predictor of the choice of the suffix aba (suffix for the 3rd person singular imperfect tense for ar verbs) if verbs were presented as class I (ar nonce verbs). In addition, the highest irregularization rates were obtained among verbs of conjugational class II/III (er nonce verbs) wi th irregular rhymes (43%). V erbs across both classes containing irregular rhymes yielded neighborhood effects whi ch, for Brovetto & Ullman suggested that irregular verbs were memorized on the basis of phonological similari ty. They concluded that the interaction of two different mental systems was necessary to explain the representation and processing of Spanish verbal morphology: a system based on the
67 manipulation of abstract symbolic categories and a system for associative memory (Brovetto & Ullman, 2005). By following the predictions of the DP model for L2 late learners, Rodrguez (in press) analyzed from a purely morphological point of view, th e oral and written production of the Spanish copular verbs from nine beginning learners during their first year of formal instruction at the university level The distribution of ser and estar (both to be ) verb forms across regularity groups showed a heavy reliance on the declarative memory, e.g. 366 of the tokens were the verb f orm es ( s/he /it is ), which accounted for more than half of the dataset (57.1% of the 64 1 inflected tokens). I claimed that this form was favored because it is a suppletive irregular verb f orm that can be easily learned and computed over associative learn ing mechanisms and because the verb ser ( to be ) appears more frequently in the input than estar (to be ), and, as a consequence, it usually appears earlier and with higher accuracy in the production of L2 learners. Except for the 1st person singular, a ll present tense estar (to be) forms are regular if one considers morphology only. However, when present tense forms appear in the production of L2 learners, it does not always necessarily mean that they are good examples of rule learning mechanisms. In fact, from the 14 examples of the regular verb form ests (you are), all but two tokens were used for the greeting cmo ests? (how are you?) and it was argued that learners memorized this fixed expression as an unanalyzed chunk that they inserted into their speech, which may explain the perfect accuracy rate on this verb form. For the most part, preterite, imperfect and future tense ser and estar verb forms were consistently used by each participant either correctly or incorrectly all the time Intere stingly, verb forms such as estaba ( s/he /I used to be) should have been absent in the participants production for the first semester because these tenses are introduced later in the Spanish program and participants had already indicated
68 that they had n ot learned any Spanish prior to entering college. The occurrence of these tokens can be better understood if we think of them as unanalyzed vocabulary items in the declarative memory which were inserted in the production of these participants when they had to provide a word with a given semantic feature, e.g. [+ past]. I concluded that the initial stages for the acquisition of ser/ estar are also a morphological challenge in nature, not only a semantic challenge. Previous work on the acquisition of complex L2 verbal morphology in languages with an inflect ional system richer than that of English is inconclusive. T here is a need to conduct more longitudinal studies in SLA working under either single mechanism or dual mechanism models if our goal is to track change in the learners internal mental state. Most previous SLA studies contributing to the present debate did not account for changes over longer periods of time (especially longitudinal studies over several years with the same participant pool), or even among different proficiency groups of L2 learners. Most of the time, these limitations hinder comparability among research findings because we miss the broader picture of their results. This study is a step towards that goal, i.e. to provide data from dif ferent groups of L2 learners at varying proficiency levels and in a language with richer morphology than English. 2.4.2. L2 Morphological Processing and the Brain U nder the DP Model Language processing occurs at an extremely fast rate (Kaan, 2007) and cog nitive l inguists need a method capable of capturing not only the timing but also the different stages involved in language processing. Since the 1980s, various linguistic phenomena have been investigated during language comprehension and production thanks to the discovery of a number of event related brain potential (ERP) components. In addition to ERPs, there is another method capable of capturing the brain activity that occurs during human learning, functional magnetic resonance imaging (fMRI). Both metho ds have also been used in SLA research to test wheth er
69 L2 learners employ the same neural and cognitive mechanisms as native speakers for language learning (Kaan, 2007). Since what makes the DP model different from any other dual mechanism theory is its he avy reliance on two mental systems with both language and nonlanguage functions, it is necessary to review previous work on SLA using psycholinguistic and neurolinguistic methods to test the DP model for L2 processing, which is addressed in the following sections. 126.96.36.199. ERP s tudies of L2 morphological processing In this section, the previous work on L2 processing using ERPs is covered. An electroencephalogram ( EEG ) is an electrical brain response that is the result of either an internal stimulus, i.e., a thought, or of an external stimulus, i.e., a perception. By placing 16 to 128 electrodes simultaneously on a persons scalp, the electrical brain activity can be recorded while the participant is being exposed to stimuli or simply carrying out a certain task. At the onset of the event of interest, e.g., the presentation of an overregularized verb form, the brain potentials from a large number of trials of the same type are averaged (Kaan, 2007). The ERP waveforms, which can be positive or negative, usual ly change due to the experimental manipulation. When there is activity from large groups of neurons situated re latively close to the skull, ERP data can be collected. There are three ERP components of interest for the clai ms of the DP model as well as for language processing in general: the N400, the early left anterior negativity (or ELAN), and the P600 components. The first ERP component is the N400, which is typically elicited by unexpected linguistic stimuli. It is a negative going component, peaking a pproximately at 400 m s (300500 ms.) after the presentation of the stimulus, usually a word or a picture. It is largest over posterior areas of the scalp, and either bilateral or right lateralized. The N400 was discovered in 1980 by Kutas & Hillyard in the first study that used ERPs to examine language processing. This component is
70 often elicited by semantically inappropriate words in an otherwise acceptable sentential context, and has also been shown to occur in response to words at the end of a sentence when there was a problem earlier in the sentence. In general, the more difficult it is to "integrate" a word into the preceding context, the larger the N400 that word will elicit. The second component is ELAN, a negative going wave that peaks around 200 ms or less after the onset of a stimulus, and has been found in some studies to occur in response to linguistic stimuli that violate word category or phrase structure rules (e.g., *the on table instead of on the table ). While ELAN is frequently used in such areas of neurolinguistic experiments as sentence processing it is still controversial if this ERP component is a language specific phenomenon (Kaan, 2007) The third and last component is the P600, a positive voltage deflection which occurs at 600 ms. af ter stimulus onset. This component is sensitive to the degree to which a syntactic continuation is expected; i.e., ungrammatical continuations elicit larger P600s than those that are grammatical. There is no evidence yet that it is necessarily a language s pecific phenomenon. The DP model predicts that ERP components associated with grammatical processing in L1 should be absent or modified in L2 speakers; ERP effects associated with lexical conceptual processing should differ minimally between L1 and L2, an d L2 speakers may show ERP components associated with lexical conceptual processing for grammatical computations (Ullman, 2001b, p. 114). The following studies (Hahne, 2001; Hahne & Friederici, 2001; Hahne et al., 2006; McLaughlin et al., 2004; Osterhout & McLaughlin, 2000; Rossi et al., 2006; Tokowicz & MacWhinney, 2005; Weber Fox & Neville, 1996) are arranged chronologically and are evaluated to the extent to which they support these claims. Weber Fox & Neville (1996) conducted an ERP study on syntacti c and lexical conceptual violations in bilingual learners. Their study included 61 adult Chinese/English
71 bilinguals with varying ages of initial exposure to English: 13, 4 6, 7 10, 1113, and greater than 16. Participants were asked to read sentences in E nglish which had semantic anomalies or syntactic violations (either phrase structure violations, specificity constraints or subjacency constrain ts). Weber Fox & Neville found that later age of exposure to English was associated with worse behavioral perfor mance than native speakers when judging syntactically anomalous sentences. When compared to monolinguals, t hese L2 learners also showed a different pattern of ERP components to syntactic phrase structure violations, such as no early left negativity compone nt or symmetric bilateral negativity. In contrast, only the participant group with the latest exposure showed worse performance than monolinguals when judging the lexical conceptual anomalies. In fact, these violations yielded a typical N400 component whos e distribution and amplitude were not affected by age of initial exposure except for a slight increase in the latency of the N400 for participants in the 11 13 and greater than 16 age of exposure g roups. This study shows that L2 grammatical processing yiel ds different ERP components than evidenced by L1 speakers while L2 lexical conceptual processing yields similar components to those of L1 speakers. In 2000, Osterhout & McLaughlin, as cited in Ullman (2001b), provided evidence supporting the DP model for SLA. They presented sentences with semantic or syntactic anomalies to native French speakers as well as to adult L2 learners of French while they collected ERPs. Semantic anomalies yielded an N400 effect and syntactic anomalies produced a P600 effect for the group of native speakers. However, for the group of L2 learners, both semantic and syntactic anomalies yielded an N400 effect after four weeks of instruction, although sometimes syntactic anomalies did not yield any effect. Interestingly, Osterhout & M cLaughlin (2000) reported that after four months of instruction, i.e., after a longer period of
72 practice in the L2, the ir L2 learners showed a P600 component for syntactic anomalies while keeping the N400 effect for semantic anomalies, a result which corro borated that the P600 remains robust for L2 learners with greater practice while the ELAN usually evaporates in late learners ( Ullman, 2001b, p. 117). This study supports the DP model in that there was a shift of ERP components for syntactic anomalies tha t depend on grammatical computations a shift of reliance similar to the one from the declarative to the procedural memory systems posited by the DP model for regular verb forms, as a result of greater practice in the L2. Hahne (2001) was interested in whether L1 and L2 processing are based on the same neural substrate. She investigated the semantic and syntactic processes in spoken sentence comprehension with a group of 16 L2 learners of German, whose L1 was Russian, and a group of 16 native speake rs of German. Both groups were asked to judge the grammaticality of correct sentences (e.g., the door was being closed), semantically incorrect sentences (e.g., the ocean was being closed), syntactically incorrect sentences (e.g., the shop was being on closed ), and filler sentences (e.g., the store was being on Saturday closed)7 7 This sentence structure is the corresponding literal German translation, although it is odd in English. which were presented orally via headphones. The sentences from the experimental conditions always ended with a regular German participial verb form. The EEG was recorded using 19 t in electrodes and RTs and error rates were also noted The main results indicated that t he L2 learners made more errors and showed longer RTs than the L1 group. In the L2 group, the correct sentences elicited a larger early anterior negativity (100 250 ms. ) and also in the time range of the N400 effect. This N400 effect was interpreted as a reflection of semantic integration of the sentence final participle in both groups. It was suggested that the semantic integration of the sentence final word was more di fficult for the L2 learners than for the native speakers. For semantically incorrect sentences, the
73 N400 effect was longer (between 400 and 700 ms.) than that found in the correct sentence condition. It was more pronounced, as well as found earlier for th e L1 group than for the L2 learners, but overall, there were similar semantic integration problems in both groups. For syntactically incorrect sentences, there was an early anterior negativity in the L1 group only, reflecting first pass parsing processes. For the second time window (5001200 ms.), there was a reliable posterior P600 component in both groups, which was slightly delayed in the L2 group. In previous studies, the P600 was attributed to processes of syntactic integration, reanalysis, and repair among other things This P600 component seems to come into play the higher the proficiency of the L2 learners, as observed i n previous research studies. These results support the DP model in that syntactically incorrect sentences elicited a delayed P600 component without ELAN for L2 learners, while semantically incorrect sentences elicited the same behavioral and functional problems (N400 effect) for L1 and L2 participants. In 2001, Hahne & Friederici reported cumulative data in favor of the DP model si milar to the previous study (Hahne, 2001). Although they used similar stimuli to those employed by Hahne (2001), their participants were different: native German speakers and 12 Japanese speakers, learners of German as an L2. These L2 learners began to learn German at a mean age of 21 (after puberty) and they had lived in Germany for about two and a half years on average. ERPs were collected from the participants while they listened to correct, semantically incorrect, syntactically incorrect, or both semant ically and syntactically incorrect sentences. After each test item they had to decide if the sentences were grammatically correct or incorrect. This kept participants focused although the main interest of the researchers was to collect their ERPs. Semanti cally incorrect sentences elicited in both the L1 German speakers and the L2 learners the same ERP patterns: a centro parietal N400 component, which lasted a little bit longer for L2
74 learners. However, the ERP components present for L1 speakers in reaction to syntactic violations were absent in the L2 group: an ELAN and a P600. Late learners, as evidenced by these results, may not be able to achieve nativelike competence for syntactic processing (procedural system), while their semantic processing (declarative system) seems to be similar to that observed in L1 speakers. The DP model also stated that for regular verb forms L2 learners may never achieve native like competence in that L2 learners may never handle them procedurally, especially those L2 learners whose age of initial exposure to the target language w as after puberty. McLaughlin et al. (2004) conducted a longitudinal study on L2 word learning by English learners of French. Participants were presented with pairs of letter strings and they had to ind icate whether the second string was a real word or not in French. ERPs were collected after 14, 63 and 138 hours of classroom instruction on average. The main result was that behavioral performance on the first session, after 14 hours, was at chance level, but, more importantly, the ERPs elicited different responses to pseudowords vs. real words, with the former yielding larger N400 components. McLaughlin et al. (2004) concluded that even though behavioral results were not beyond chance level, the L2 partic ipants seemed to have learned so me aspects of French word forms at a more automatic stage of processing since the nonce words simply had one or two of their central letters replaced. Tokowicz & MacWhinney (2005) investigated explicit and implicit processes during L2 sentence comprehension. Nineteen L2 Spanish learners who were enrolled in one of the four semesters of beginning Spanish at the college level (five in the first semester, three in the second, nine in the third, and two in the fourth) participate d in their study. Participants had to judge whether 360 sentences were syntactically appropriate in Spanish, as the explicit measure, while
75 their electrical activity in the brain was being recorded, as the implicit measure. They responded to the test items by pressing buttons on a computer keyboard. Tokowicz & MacWhinney (2005) predicted that beginning L2 learners would only be sensitive to violations of particular types based on L1 L2 similarity. In fact, L2 learners were marginally sensitive (P600 compone nt) to auxiliary omissions (similar syntactic characteristic in the L1 and the L2) and to violations of determiner gender agreement (unique to the L2), but they were not sensitive to violations of determiner number agreement (different in the L1 and the L2 ). L2 learners responded less accurately to the unique construction in the L2 than to the other two types: 57.88% for determiner gender, 70.13% for auxiliary omission, and 70.38% for determiner number. These results demonstrated that beginning learners sho wed implicit online sensitivity to some violations of L2 grammar during sentence processing, even when their grammaticality judgments were at chance levels. This is one of the greatest strengths of ERPs, i.e., being sensitive enough to obtain data whe n behavioral methods are not. These data may even suggest that L2 le arners were developing effective implicit processing for Spanish in the absence of an ability to make use of explicit processes or grammaticality judgments. Rossi et al. (2006) investigated the role of L2 proficiency (high versus low) in late learners of German and Italian. Four experimental groups were considered (n = 69) on the basis of the L2 as well as participants proficiency levels, i.e., groups 1 & 2 (n = 16 and 19, respectively) were hi gh and low proficiency L2 learners of German, whereas groups 3 and 4 (n = 16 and 18, respectively) were high and low proficiency L2 learners of Italian. Additionally, a group of 30 native speakers of German were included. The study used ERP recordings fr om 23 scalp positions while participants listened to 420 L2 sentences (three experimental conditions, one control condition, and 180 fillers). Three types of violations were under investigation: (a)
76 category violation where a verb directly followed a prepo sition so that the expected noun was absent, (b) agreement violation where the verb took an incorrect affix in relation to person/number agreement, and (c) combined violation where the preceding agreement and category violations were united. For the morpho syntactic agreement violation, high proficiency L2 learners showed the same ERP components as native speakers: left anterior negativity or LAN (for the detection of the morphosyntactic agreement error) and P600 (for the reanalysis processes); low proficien cy L2 learners failed to show a LAN component, since morphosyntax seems to be acquired later than word category information. Rossi et al. claimed that their results illustrate that high proficiency L2 learners show native like brain processing mechanisms d uring sentence comprehension, even at comparable timings. Furthermore, these findings provide evidence that there are universal syntactic processing mechanisms within Indo European languages as observed in the German and Italian L2 groups. These results ar e interesting for the DP model in that only high proficiency L2 learners were sensitive to the morphosyntactic agreement error, showing that they were processing grammar in a similar way to native speakers of the target language. Hahne et al. (2006) invest igated how L2 learners process inflected words online using ERPs. They concentrated on two German inflectional systems, participle formation and noun plurals. They presented to a group of late L2 German learners, with Russian as their L1, sentences contain ing two types of violations: overregularizations and incorrect regular verbs After the EEG sessions, participants had to inflect the test items (produce the corresponding participle form for verbs or the appropriate plural for nouns) and results indicated that they were familiar with the test items (95% accuracy for the participles and 86% for the plurals). Interestingly, the L2 learners reacted differently to the violations of regular and irregular
77 inflection. The overapplications of the s plural rule pr oduced a P600 component while the overapplications of irregular patterns elicited an N400 component. Hahne et al. (2006) concluded that at least in those domains in which L2 learners are highly proficient, they can employ the same mechanisms for morphologi cal processing as L1 speakers. The clear dissociations between regular and irregular inflection for participle and plural formations indicate that the two processing routes posited by dual mechanism models of inflection are also employed by L2 learners, pr ovided that they are highly proficient in the knowledge of the morphological domains under study. In sum, adult L2 language processing (ERP studies) showed some strikin g differences to native speaker processing in the domain of syntax and mo rphology Most ERP/EEG studies offered evidence that L2 learners are guided by lexical semantic cues for language domains which are thought of as being handled by the declarative memory system, but they were less dependent on the procedural memory system for syntactic representations, unless they had achieved high proficiency in the L2 or after they had had extensive training and practice in the L2. This is an essential difference between L2 speakers and native speakers. In the next section, cumulative evidence for the DP model and SLA is provided, but from studies measuring neocortical brain activation with the help of a different neurolinguistic tool, fMRI scanners. This tool provides a means to see the internal activation of the brain to the smallest detail but at the e xpense of temporal precision, just the opposite of ERPs. Comparable findings on the same topic from both types of data collection are the best way to provide strong results 188.8.131.52. fMRI s tudies of L2 morphological processing One of the most recently deve loped forms of neuroimaging is fMRI. Since the early 1990s, fMRI has come to dominate the brain mapping field due to its low invasiveness, lack of radiation exposure, and relativel y wide availability. When nerve cells are active they consume
78 oxygen and the local response to this oxygen utilization is an increase in blood flow to regions of increased neural activity, occurring after a delay of approximately 15 seconds. FMRI scans measure the blood flow response related to neural activity in the brain. This method can inform neurolinguists about the brain regions which are active when participants carry out a specific task such as producing complex verb forms or reading ungrammatical sentences. What follows is the su mmary of recent studies (Opitz & Friederici, 2003, 2004; Rschemeyer et al., 2005; Sakai et al., 2004; Tatsuno & Sakai, 2005) that used an fMRI scanner to investigate brain activation patterns for L2 learners in relation to the acquisi tion of inflectional morphology. These studies are presented in chronological order. They are critically reviewed to check the extent to which they support the predictions of the DP model for L2 learners, which are outlined here: (1) The subcortical brain structures for the p rocessing of irregular verb forms should be in the hippocampus, medial temporal lobe regions, BA45, BA47, and parts of the right cerebellum. (2) The subcortical brain structures for the processing of regular verb forms should be in the basal ganglia, the left hemisphere of the cerebrum, the supplementary motor area, and BA44. (3) Finally, medial temporal lobe structures (declarative/associative memory system) are expected to underlie the use of both types of forms with a general decrease in left lateralization (procedural/grammatical system) for L2 learners. Opitz & Friederici (2003) investigated the neural basis of the learning of a miniature artificial grammar BROCANTO. Its grammar was composed of eight words and several rules; two of the rules are abse nt in German (the L1 of the 14 participants considered for the analyses), or in any other natural language known by the participants in order to control for potential
79 transfer effects. Participants were instructed in BROCANTO visually via a computer for 15 blocks, each one composed of 70 sentences for learning (sentences for which they were instructed to extract the underlying grammatical rules), 70 testing sentences (a grammaticality judgment task that included phrase structure and determiner noun agreement violations as well as word class repetitions), and 2 0 sensorimotor control stimuli. Behavioral results showed improved performance across the 15 training blocks from the chance level, 52% correct responses, on block 1, to 81% correct on block 15. The BRO CANTO grammar was successfully acquired during the training blocks, and participants obtained 74% accuracy for the new sentences presented to them several minutes after scanning. During the learning process, there was a decreased blood flow activation on t he left posterior hippocampal area as well as on the retrospenial cortex and left anterior and posterior temporal areas (Opitz & Friederici, 2003, p. 1734). The authors explained these results in light of similarity based learning during the initial learni ng state of this artificial grammar, a grammar which was no longer used once the grammatical rules had been acquired. The lateral temporal lobes are traditionally thought of as being engaged in lexical semantic processing. The rapid decrease in these regio ns in this study was interpreted as the brains natural reaction to the irrelevance of the lexical semantic aspects of BROCANTO, an artificial language. Participants also showed a pattern of increased activity during learning on the opercular part of the l eft inferior frontal gyrus (IFG), traditionally referred to as Brocas area, and interestingly restricted to BA44. The involvement of these regions had been found previously in relation to abstract nonlinguistic rule learning as well as in a variety of tas ks with high attentional demands. These results were interpreted by Opitz & Friederici (2003) as consistent with the DP model in that there is an initial involvement of the posterior
80 hippocampus (declarative memory) followed by a shift to left inferior fro ntal activity (procedural memory) which was interpreted as the pr oficient use of an artificial grammar. In 2004, Opitz & Friederici again set out to examine whether the acquisition of grammatical knowledge depended primarily on abstract rules or on item s pecific information. T his time, 21 participants were trained in the artificial language BROCANTO for 48 hours, on average, prior to the experiment, in which participants were assigned to one of two conditions. In the so called WORD condition, nouns that we re presented only in the sentence initial noun phrase during training would later on appear in the sentencefinal noun phrase, but the structure of the noun phrases were maintained In the so called RULE condition, a modifier was allowed to follow the noun to which it referred, although during training the modifier was always placed before the noun. Results indicated that participants in the WORD condition judged the new items as grammatical although they differed superficially from the training exemplars. Interestingly, there was observable but transitory increased brain activation in the left anterior hippocampus for the WORD condition. In contrast, the RULE condition triggered a gradual increased engagement on the left ventral premotor cortex caused by the introduction of the new phrase structure rule. Opitz & Friederici (2004) argued that their results pointed to a dissociation between similarity based learning versus rule based learning during language acquisition. Crucially, their results fully supporte d th e DP model for SLA in that there was a shift to a gradual engagement of the left inferior frontal activity in the RULE condition, which is associated with the proficient use of the artificial gr ammar of BROCANTO. Sakai et al. (2004) studied whether lea rning new knowledge, in this case the English past tense, for beginning L2 learners over a period of two months was associated with brain plasticity and cortical changes. Seven pairs of native Japanese speakers, all 13 years old and all twins,
81 received int ensive training in the English past tense formation for 64 regular and 64 irregular verbs via bingo games as part of their standard classroom education. They had 25 training sessions of about five minutes each. Participants completed two fMRI sessions, one before and one after the training period, in which they completed four tasks: an English and a Japanese verbmatching task (EM and JM, respectively) as well as an English and a Japanese past tense task (EP and JP, respectively). Once a symbol was displaye d representing the task to be performed over a computer screen, a bare infinitive was presented and participants read it silently. The next screen displayed two stimuli for one second and participants had to either select the correct bare infinitive or the correct past tense form by pushing one of two buttons (participants always saw the correct regular/irregular verb form and an incorrect regular/irregular counterpart, such as catched vs. caught ). Images of their brains were obtained in 16 horizontal slic es per test item and participant. In this study, the brain subcortical activations did not show dissociation between the regular and the irregular English past tense forms. For time 2, when the EP regular and irregular stimuli blocks were contrasted with t he EM blocks, similar cortical regions were activated: the dorsal region of the left IFG, BA46 and the right cerebellum (Sakai et al., 2004, p. 1236). The only observed differences were present in the left angular gyrus and supramarginal gyrus for the EP i rregular verbs as well as the wider activations on the left IFG for the irregular verbs than for the regular ones. When the cortical activations for the EP regular and irregular verbs we re summed up, they exactly match ed the regions activated for the Japan ese (L1) past tense task only in the left dorsal IFG, suggesting that these paralleled increases represent the acquisition of past tense knowledge. In addition, Sakai et al. (2004) found a significant positive correlation between the increases in examinati on scores and the activation increases in the left dorsal IFG (Sakai et al., 2004, p. 1237), suggesting that this cortical
82 activation might be a good indicator of individual improvement in the acquisition of L2 past tense formation. T he following year, T atsuno & Sakai (2005) conducted a similar study to Sakai et al. (2004) comparing the cortical activations in two groups of participants, taking age (13 vs. 19 year old age groups), proficiency (high H vs. low L), and language task demands (verb matchin g M vs. past tense verb identification P) into account during L1 Japanese and L2 English processing. The 14 participants in the 13 year old age group had participated in the previous study (Sakai et al., 2004) while the participants in the 19 year old age group were 15 undergraduate students who were new to this study. Each age group was further subdivided into the two proficiency subgroups, based on the median of their accuracy on the English irregular verbs. Tatsuno & Sakai (2005) used 24 regular and 24 irregular English verbs and 48 Japanese verbs matched in meaning for the data collection. Tasks and procedure were exactly the same as in the previous study (Sakai et al., 2004). Behavioral results showed that there were significant differences between the 19EL and 19EH subgroups in the accuracy and RTs to the irregular verbs in the EP tasks, even though the four subgroups of participants showed accuracy greater than 90% (Tatsuno & Sakai, 2005, p. 1639). The brain activation patterns were similar between age 13 and age 19 groups as well as between EPirregular EM and EPregular EM tasks because there were consistent activations in the left IFG. However, the main results stem from the different activation patterns found among groups. First, there were lower brain activations at the left F3t in the EPirregular EM and EPregular EM tasks for the age 19 group, suggesting that proficiency level plays a major role in this region during L2 acquisition (Tatsuno & Sakai, 2005, p. 1642). Previous studies have also repo rted that activation in the left F3t increases when participants first learn the grammatical rules of an L2. Second, there were greater activations in
83 the left F3t/F3O region in the EPirregular EM task for the low proficient age 19 group than for the high proficient age 19 group, which may suggest that the left F3t/F3O activation is related to the sensitivity to specific language task demands rather than associated with cognitive development in general. In fact, this region was activated significantly more for the age 13 group during the JP JM task than for the age 19 group, even though performance in this task was matched between the high proficiency age 13 group and the low proficienc y age 19 group. Tatsuno & Sakai (2005) concluded that the left IFG subser ves language specific functions and is specifically involved in the cortical plasticity for L2 acquisition. Rschemeyer et al. (2005) presented imaging data from advanced L2 learners when processing spoken sentences with either syntactic or semantic violations. Fourteen learner s of German, NSs of Russian, participated in the study. They had been living in Germany for an average of five years. In the syntactically incorrect sentences, an auxiliary verb was followed directly by an inflected preposition that was immediately followed by the participle of the main verb instead of the necessary noun. In the semantically incorrect sentences, the meaning of the final sentence participle could not be integrated into the preceding sentence context. This group of L2 l earners showed no significant differences in the processing of syntactically incorrect sentences when compared to the correct ones, although there was a reliable region of increased activation in comparison to the NSs in BA44 in response to correct and syntactically incorrect sentences This same response did not occur with respect to the semantically incorrect sentences. According to Rschemeyer et al. the increased BA44 activation reflected additional phonological processing for understanding speech signals in the target language. Results also indicated increased activation in the left anterior IFG for semantically incorrect sentences versus the correct ones, which was partially comparable to the brain activations by NSs in response to
84 the same stimuli. T his finding led the authors to speculate that processes underlying semantic processing seemed to be more similar in native and non native speakers of a given language than did those underlying syntactic processing (Rschemeyer et al., 2005, p. 283). What is clear from the studies reviewed here is as Sakai et al. (2004) suggested, that data have not been conclusive as to the regular/irregular verb distinction (p. 1238). I n general, though, tasks which involved grammatical processing showed greater tempo ral lobe involvement in both hemispheres of the brain (Ullman, 2001b) and the extent of temporal lobe involveme nt was greater for later than earlier L2 learners. For example, Tatsuno & Sakai (2005) found increased activity in the lower portion of the left IFG (BA 45/47) in the processing of irregular but not in the processing of regular past tense stimuli compared to the processing of similar items in a verbmatching task. There had not always been consistent brain activations across these studies but it is important to remember that experiment specific factors such as processing difficulty, small number of stimuli or phonological complexity are usually cited as being responsible for some of the activation differences found in imaging studies. Nonetheless, t here have been some areas of agreement for most studies, e.g., irregular verbs activated more brain areas than regular verbs, and there are common regions of brain activation in the production of regular and irregular past tense compared to baseline items. ERPs and fMRI scans are two powerful tools which are used in neurocognitive linguistics to test the claims of the main theoretical positions in the debate about the mental representation and productio n of inflectional morphology. 2.4. 3. The DP Model and t he Cognitive Approach to Language Learning According to Doughty & Long (2003) research on SLA is increasingly viewed as a branch of cognitive science because language learning is ultimately a matter of change in an individuals internal mental state. The DP model facilitates the integration of data from across
85 different disciplines, primarily theoretical linguistics, cognitive psychology and cognitive neuroscience (Walenski & Ullman, 2005), as we have just seen from the previous sections. W hat is interesting about the DP model is that it can be easily integrated into a broader SLA theory, the Cognitive Approach (S kehan, 1998), since both share key features with parallel hypotheses and predictions for L2 learners. These parallelisms are described in this section by providing specific examples from key research studies as well as components of working memory and long term memory all of which are full y compatible with the DP model. The Cognitive Approach outlined by Skehan (1998) is primarily an information pr ocessin g account on how IL develops. Two assumptions underlie the main claims of this theory. First, the language acquisition device, i.e. Universal Grammar, is no longer available for L2 learners after a critical period. Second, meaning takes priority fo r older learners. Although in this theory meaning is considered the driving force for ongoing communication, this claim forces the model to search for an explanation of how is it possible that the IL develops and changes over time, since form is of seconda ry importance. A main point in the Cognitive Approach is that normal communication is heavily based on the pressures of processing language in real time rather than on the use of complete and well formed sentences. During ongoing conversation, we reduce th e processing load by relying on time creating devices, prediction skills, elliptical language, and so on. Furthermore, it is claimed that much adult conversation is incomplete in surface form, enabling inferences about intended meaning, speaker attitudes, and the like. Furthermore, the older we get, the more adept we become at developing such an analytic knowledge system. But language use, in itself, does not lead to the development of an analytic knowledge system because meaning distracts attention from f orm. The traditional rule based approach (starting every sentence from scratch and filling
86 the gaps in a given syntactic structure with appropriate words) is thought of as an imposition of the linguist that facilitates the expression of new meanings and en ables the production of utterances which have never been created before (Skehan, 1998) Instead, the Cognitive Approach claims that most of the speech we produce is likely to have been produced before, at least in part, through the use of memorized chunks of information, even by the same speaker: Id like to, Im gonna try to, It doesnt matter how, I dont agree with... As a result, an average native speaker knows hundreds of thousands of such lexicalized sentence stems which may be used during ongoing conversation to achieve fluency. It is argued that there is a parallel knowledge system which is primarily lexical in nature and very important for successful rapid communication. In such a way, a dual representation system (one rulebased and one exemplar based) is used to satisfy a number of competing d emands: improvisation, fluency, processing demands, creativity, syntactic constraints, rule generation, and the like. The rulebased system implies compact storage and powerful generative rules to compute w ellformed sentences, while the exemplar based system occupies a large redundantly structured memory system. The rule based system and the exemplar based system is a similar notion to the dual mechanism models. Such a collection of fixed phrases can be ada pted to convey different meanings, and individuals have the capacity of expanding their repertoires over time. It is further suggested that these fixed phrases will reflect the meanings that any particular individual is most likely to express, i.e., lexica lized idiolects. Under this approach, the task of the L 2 learner is that of appropriate selection of lexicalized sentence stems. There are L2 learners who produce grammatical and fluent utterances but still sound foreign because the choices they are maki ng are not grammatically correct in the target language such as The project consists about or La tarea es por maana ( The
87 homework is for tomorrow, wrong selection of the Spanish preposition, which should have been para). Much of the time, learners are thought of as relying on large chunks of memorized language rather than starting every single sentence from scratch. For example, given a fixed phrase like why dont you cancel class today?, learners can easily adjust it to the communicative needs of o ngoing conversation by changing the final words without the need to create the whole sentence by grammar rules: Why dont you + VB (infinitive). Instead, its meaning is of primary importance. In fact, there is also the possibility that the learner still do es not know how to form the contraction of a negative adverb with an auxiliary verb at all but will produce the previous example wi th accuracy and fluency, as in why dont you come with me? These memorized chunks of language are handled by the associate sy stem (i.e. o ur declarative memory), which is considered to be organized for ease of use rather than for efficient compactness, even if that implies that the same word may be redundantly stored more than once. Syntax and the application of grammar rules are de emphasized in this model for L2 learners, which fits perfectly with some of the claims for beginning L2 le arners from the DP model, e.g.; adult L2 learners at earlier stages of acquisition rely primarily on associative memory because both regular verbs and irregular verbs will be stored initially as opposed to being handled by the two memory systems separately, which is how native speakers handle regular and irregular verb forms. The Cognitive Approach claims that humans are thought of as possessing a limited capacity memory s ystem, and this limitation place s a fundamental constraint on how input is handled. Working memory need s to extract what is relevant for ongoing comprehension. This notion has direct consequences on how morphology is processed. F or example, learners process more meaningful morphology (imperfect suffixes in Spanish) before less meaningful
88 morphology (redundant verbal agreement) because learners process input for meaning before they process it for form, one of the principles for inp ut processing proposed by VanPatten (1996). Input frequency and saliency are incorporated in this model to promote what Schmidt (1 990) calls noticing or degree of awareness, meaning that the learner has to perceive a gap in her/his current IL grammar. This process can happen either during comprehension or pr oduction. F or several reasons, though, the notion of noticing is more important during production because it force s learners to test their hypotheses about the L2 and to focus on form. As an example, Swain (1995, 1998) emphasized that one of the key functions of output is that of testing hypotheses about the target language, which may aid L2 learners to find discrepancies between their current IL state and the L2 linguistic system. There is evidence t hat the first stages in SLA are primarily lexical in nature (which explains why irregular verbs have an advantage at first) and there is a danger that learners will not progress beyond this stage. According to the Cognitive Approach, the internally generat ed pressure for syntactization (language which was available on a lexical basis becomes reorganized to be syntactically based, i.e., fixed expressions are understood now in terms of their constituent parts) will not come into play naturally once the critic al period is past. Again, this is a claim fully compatible with the DP model in that, as we have seen before, age of exposure sensitivity in the L2 affects the procedural memory system, since the learning and use of grammar depends largely upon procedural memory. Ultimately, the Cognitive Approach (Skehan, 1998) proposes a methodology that focus es on form while keeping meaning as its key component: taskbased instruction. According to Long (1991) this approach is the most effective means of directing L2 l earners attention towards L2 grammar. His focus on form approach proposes that the main focus of the lesson should be on meaning via plenty of comprehensible input and interaction in the L2, while form focus should
89 arise in context as needed either expli citly by short grammar explanations or implicitly via recasts or negotiation of meaning. Additionally, the Cognitive Approach makes strong assumptions related to individual learner differences: some learners may have a better talent specifically for langu age, and this natural talent is a stable and untrainable construct. For inflectional morphology, this simply means that some learners will have the capacity to infer word internal structure from a given language and make generalizations about how other linguistic material might be encoded, progressing more quickly; other learners will need more time, so that focused instruction will have greater importance for them. In fact, one of the roles of output is to develop automaticity, an essential notion for the acquisition of morphology (Skehan, 1998). Empirical evidence that the first stages of SLA are usually lexical in nature is provided by Housen (2002). His analysis of 29,000 verb phrases from 46 learners of English showed three general stages. The first is characterized by missing verbs or verbs that appeared in rotelearned formulaic expressions, e.g., I dont know The second phase is one in which productive verbs show up as morphologically invariant forms (V ), e.g., eat, want ; some inflected forms, part icularly highly frequent irregular past forms, e.g., got ; and progressive forms (V ing), e.g., dancing. The third phase is marked by diversification of verb forms As exemplified, stages 1 and 2 can be claimed to be driven by lexical learning. However, eve n greater evidence in favor of a contrast between rule learning mechanisms for regular morphology and associative learning mechanisms for irregular morphology under the Cognitive Approach is Elliss (1987) study. He investigated the effect of planning oppo rtunities on SLA performance among 17 post beginner or early intermediate EFL students from different language backgrounds. Regular and irregular past tense forms were elicited through three narrative tasks involving story telling. In Task 1,
90 participants had to write a story from a set of six pictures. In Task 2, the author recorded the second attempt of an oral version of the same story. In Task 3, participants had to narrate a story to accompany a new set of pictures. Each task provided them with less pl anning time and, consequently, with fewer opportunities to attend to form. Results indicated that accuracy on the past tense morphemes declined as a function of less planning time being available. But the most interesting finding was that performance on th e regular and irregular past tense forms differed significantly across tasks Accuracy on the regular past tense forms decreased from 77% in Task 1, to 57% in Task 2, to 43% in Task 3. On the other hand, accuracy on the irregular past tense forms remained relatively constant across tasks: 60%, 57% and 55%, respectively. This dissociation in accuracy scores between regular and irregular morphology is based on the belief that when the online processing capacity becomes less available, attention to form decreases. Rule application requires a considerable degree of online computation during language production and that is why regular past tense forms decreased in accuracy from Task 1 to Task 3. In contrast, irregular verbs maintained a constant level of accuracy across tasks because they do not require a high degree of processing. In fact, Ellis (1987) concluded that irregular verbs have to be learned as separate tokens, each past tense form constituting a distinct lexical item in the learners interlanguage (p 11). Skehan (1998) further suggested that Ellis (1987) study illustrates the ability with which learners can switch between rulebased and exemplar ba sed representational systems. As shown, the main advantage of the Cognitive Approach in providing an ex planation of how verbal morphology is acquired is the integration into a unified model of external (comprehensible input) and internal factors (aptitude) that affect SLA, coupled with psychological factors that determine how real time communication is carr ied out. Morphology is
91 no diff erent from ot her grammatical features ; the greater the learners aptitude, the faster s /he will notice a gap in her/his IL, which will push the IL grammar to move forward. The conceptualization of language learning by the Cogn itive Approach is fully compatible with the Words and Rules theory and the DP model for regular and irregular past tense verb forms (Pinker, 1991, 1998, 1999; Pinker & Prince, 1994; Pinker & Ullman, 2002; Ullman, 2001a, 2001b, 2004, 2005). And e ven though the Cognitive Approach is not very specific concerning how verb al morphology is acquired in L2 the model can accommodate many of the arguments prop osed here for the contrast between regular and irregular verb morphology, as shown in Figure 21. Input frequency, for example, i s a key argument for the regular/irregular diss ociation. Figure 21. Main influences on noticing and components of working memory and long term memory for the DP model from the Cognitive Approach. [ Adapted from Skehan, P. (1998). A cognitive approach to language learning (Page 57, Figure 36). New York: Oxford University Press.] Pinker (1991, 1998) proposed that most surviving irregular verbs in English are used at high frequencies, and that irregul ar forms have to be memorized repeatedly, generation after generation, in order to survive in a language. In fact, low frequency irregular verbs were Noticing Input qualities: Frequency Salience Focused input: Instruction Task demands on processing resources Long term memory: Rule based analytical system Memory based formulaic system Working memory: Language processing Recombination, transformation Rule based exemplar generation O U T P U T I N P U T
92 converted to regular verbs over the centuries. For S LA, the prediction that would follow is that those for ms that appear more frequently in the L2 input (irregular verbs), will be the first forms noticed and will be acquired with greater accuracy. That is generally observed in the L2 literature. Even though the model deals with larger chunks of information, at the sentence level, the dual mode system in long term memory can account for a rulebased analytical system that will be used for regularly inflected verb forms, and a memory based formulaic system that will be used for irregular verb forms. In the previo us sections, some evidence that there are brain structures differentially implicated in either one or both of these memory systems has been provided. However, the debate remains unresolved, and there is a need for continue d empirical evidence from differen t populations, linguistic systems, and task features In sum, the Cognitive Approach to L2 learning (Skehan, 1998) has many parallels to the DP model. T he Cognitive Approach is a theory which is interested in interpreting SLA as the result of both a rest ructuring of linguistic material (McLaughlin, 1990) by the operation of a mental rule based grammar or by other cognitive processes, as well as by the accumulation of useful chunks of language as formulaic items. This interplay between a rulebased analyt ical system and a memory based formulaic system i n long term memory (refer back to Figure 2 1) mirrors the procedural and declarative memory systems in the DP model, which are posited to interact cooperatively and competitively in the processing of com plex word forms. In Chapter 5 the extent to which the research results of this in vestigation corroborate the Cognitive Approach to language learning is also evaluated. 2. 5. Additional Work on the Acquisition of Spanish Regular/Irregular Morphology This final section of the review of previous studies is a summary of additional studies in Spanish SLA dealing with the acquisition of verbal morphology, in which re gularity was a variable in the analysis and/or interpretation o f the results (Bruhn de Garavito, 2003; Camps,
93 2005; Salaberry, 1999, 2000). These studies did not use the DP model or any dual mechanism model to account for the main findings; nevertheless, they can be easily accommodated into the DP model to some extent. This review covers additional resourc es in the study of regularity in the acquisition of Spanish verbal morphology by L2 learners and it shows the extent to which the DP model can contribute to this area of research interest Salaberry (1999) tested the effect of inherent lexical aspectual va lue of verbal predicates in the use of past tense morphological marking in the L2 Spanish of 20 adult learners. The lexical aspect hypothesis states that the learners selection of verbal morphology is related to the inherent lexical semantics of the verb phrase. Sixteen of his 20 participants were enrolled in four different levels of Spanish courses, four participants per proficiency level. His data included oral narratives from a silent movie collected at two different times two months apart during the sa me semester. He found evidence that the preterite forms (used mainly for completed actions in Spanish) were gradually used for telic events (achievements and accomplishments), and that the use of imperfect forms (for background information, details and ong oing actions in the past) increased consistently towards correspondence with lexical aspect, i.e., stative events first. The relevant finding in this study for the present investigation was that the imperfect morphology was initially used with a restricted set of verbs, which was considered indicative of lexical learning instead of rule application, supporting the acquisitional route the DP model predicts for the beginning stages of SLA. Salaberry (2000) examined the written and oral narratives from 14 L1 S panish speakers learning L2 English in a classroom setting in Uruguay. Again, he found that all learners relied heavily on the use of irregular morphology to mark past tense in both written and oral narratives. In contrast, the use of regularized forms of past tense lagged behind in the production of most
94 learners. He cited Pienemann (1987) and Schwartz (1993) in order to account for his results. Pienemann (1987) stated that rules which require a high degree of processing capacity are acqu ired late, while S chwartz (1993) asserted that inflectional endings are among the most difficult features of non nati ve languages for adult learners, showing the highest amount of variability and the lowest degree of success. Salaberry (2000) claimed that the effect of inherent lexical aspect may be independent of the effect of the cognitive saliency of irregular morphology in languages such as English or Spanish. As mentioned earlier, and as predicted by the DP model, regularized past tense forms appeared later in the produ ction of most learners because regular complex verb forms are expected to be handled by the procedural memory system, unless they are processed at the beginning stages of acquisition by the declarative memory system. In 2003, Bruhn de Garavito examined te nse distinctions in the initial state of L2 learners of Spanish. The main questions were whether a group of beginning learners provided evidence for knowledge of tense distinctions (present vs. preterite), and if they showed differences in accuracy between two types of verbs: regular and irregular. Fifty three university students and 10 Spanish native speakers (as the control group) participated in an oral grammaticality judgment task in which the grammaticality of the sentences depended on the correct use of tense, either present or past. This format was used in order to decrease the processing load which is associated with production. The grammaticality of the sentences depended on the use of the correct tense, e.g., Anoche no tengo tiempo de hacerla ( la tarea ) *Yesterday night I dont have time to do it (the homework) vs. Anoche no tuve tiempo de hacerla Yesterday night I didnt ha ve time to do it. There were three verbs from the 1st class, three verbs from the 2nd/3rd class, and three irregular verbs Each verb was used 12 times, with each person (first, second and third, singular or plural) appearing four times, together with eight distracter sentences for a total of 116 test
95 items. Participants used a 5 point rating scale. Both groups rated the ungr ammatical sentences in the present and preterite significantly lower than the grammatical ones, but the L2 learners did so to a much lower degree. For the regular verbs, and in the past tense only, the acceptance rate for the grammatical sentences was sign ificantly higher than for the ungrammatical sentences. A different picture emerged for irregular verbs, i.e., subjects clearly distinguished grammaticality in the case of the present tense and failed to do so in the past. Why would the present be a problem in the case of regular verbs and the past in the case of irregular verbs ? Two possible explanations were offered. First, if learners were relying on stress to recognize the past tense and not on the verb endings, its reasonable that they had problems wit h irregular verbs only, given that irregular past tense verbs do not differ in their stress pattern from the present, e.g., [ t had vs. [ t .go ] I have or [ ] I did vs. [ ] to do But the regular stress pattern in the preterite i s more salient when compared t o that of th e present tense, e.g., [ k .mo] I eat vs. [ko. m ] I ate Second, irregular verbs in the present tense are generally the most frequent verbs which may have helped learners to acqu ire them. Bruhn de Garavito (2003) reco gnized that the result s of this experiment were difficult to interpret and that the best explanation is that these learners used the following properties in the input to make their judgments: stress saliency, adverb of time rather than tense of the verb, acceptability of a present tense in a past context when interpreted as a recent past, and frequency of the verb forms tested the latter being a key variable in studies under t he DP model. Camps (2005) investigated the initial stages of imperfective morphol ogy in the written pro duction of 30 first year learners of Spanish in writing activities using the same topic twice: at the end of one semester and at the beginning of the next one in the beginning courses His data supported the predictions of the lexical aspect hypothesis (Andersen, 1991) in that when
96 imperfective forms appear in the production of L2 learners, they appear first with states, spreading later to activities, accomplishments, and finally to achievements. For example, in the second writing acti vity of this study, state verb types were appropriately marked in the imperfect with 85.1% accuracy, before any other aspectual class, followed by 68.1% accuracy for the activity aspectual class. Camps (2005) also found that irregular verbs were more frequ ent in the data overall, resulting in the highest percentage of both imperfect (47.5% were irregular verbs vs. 36.7% regular verbs ) and preterite forms (42.4% vs. 30%, respectively). In the case of state verbs, the preterite appeared more frequently with i rregular verbs than with regular verbs (which could be explained by associative memory learning), while regular and irregular verbs were more frequently marked with the imperfect than with the preterite (interpreted as a possible outcome of a rule learning mechanism in both cases). In sum, these additional resources corroborated the hypothesis that the initial stages of verbal morphology acquisition in L2 Spanish are generally lexical in nature, which can be accounted for by the predictions of the DP model and that the appearance of forms which depend upon the application of complex rules is delayed to some extent in the production of L2 learners. Coupled with the previous work on SLA under dual mechanism models ( including section 2.3. ), it is striking to observe that most studies did not track developmental changes over time. Instead, they generally studied the recognition/production of verbal morphology by L2 learners at a single point in time or at a single proficiency level. However, some of the attract iveness of the DP model for explaining L2 learners behavior is the hypothesized shift from declarative memory reliance to procedural memory reliance for forms which are compositionally computed in the mental grammar for L1 speakers, a shift which is usual ly delayed for L2 learners and which may even be absen t for some learners. T he use of lon gitudinal
97 studies or of cross sectional studies is necessary if we want to find where and how this shift may occur in SLA. T he present study aims to provide more infor mation about the mental representation and production of inflectional morphology by L2 le arners at varying proficiency levels. 2.6. Summary In this chapter it has been shown that the acquisition of inflectional morphology by L2 learners under singlemechan ism models could be explained by purely associative learning mechanisms without the need for ex plicit mental rules for regular verb forms. However, these studies typically used nonce words, i.e., not a real L2, or computer models, i.e., not human L2 partic ipants, to test the ir hypotheses. In contrast previous work under dual mechanism models showed, although not always consistently, that the time to react to irregular verb forms by L2 learners is affected by the frequency of the irregular items, while thes e frequency effects were not usually observed for regular verb forms. They also found that w hen retrieval of an irregular verb form fails overregularization errors occur. The DP model has not been widely tested in the study of SLA even though it posits cl ear and well defined hypotheses for L2 learners, especially in relation to psycholinguistic and neurolinguistic variables. For this reason previous work has yet to yield conclusive findings, and, unfortunately, the lack of longitudinal or cross linguistic studies to track developmental stages among varying levels in the participants proficiency leaves a greater laguna in our understanding T he current investigation is a step forward to fill that gap in the literature review presented in this chapter. Th e current cross sectional study is interested in the speeded oral production and recognition of complex verbal morphology in L2 Spanish by English speakers, and isolates four key variables under the DP model: (a) the regularity of specific verb forms; (b) their frequency of occurrence in the target language; (c) participants proficiency level; and (d) participants age of
98 initial exposure to the target language. This study uses an oral production task as well as two psycholinguistic experiments measuring p articipants RTs on the test items of interest. It considers participants behavioral responses at large by including accuracy scores, RTs and the nature as well as the internal structure of errors they made because little is known about specific processi ng strategies employed by L2 learners at varying degrees of proficiency levels when handling complex morphosyntactic transformations. The specific methodology and resear ch questions that guided the present investigation are introduced in the following chapter.
99 CHAPTER 3 METHODOLOGY 3.1. Introduction In this chapter, the main goals of my study are described, including the research questions which guided the investigation. The design of the tasks (participants, test items, materials, etc.) and the data col lection procedure are also described. And at the end of the chapter the primary statistical tests used for the data analysis are explained. The goal of this study is to investigate the acquisition of Spanish verbal morphology by L2 learners by looking at their processing and storage of complex verb forms in the L2 Behavioral data from adult late learners of Spanish as an L2 at three proficiency levels were collected to track qualitative and/or quantitative developmental changes in the mental representation of and access to inflectional morphology over time. Data from adult English/Spanish bilinguals were also collected to test whether there were differences between the se advanced speakers and the advanced late learners in the mental representation of and access to inflectional morphology as a result of earlier exposure to the L2. Two psycholinguistic experiments and a production task were used to add ress the research questions M e thodology in psycholinguistics takes the form of behavioral experiments in which participants are presented with some form of linguistic input and asked to perfor m a task, e.g. reproducing the stimulus, or press ing one of several buttons. RTs and the proportion of correct responses are overall the most frequently used measures of performance. To expand on the previous research, the types of errors produced by these groups of learners were also considered, because this information can provide additional evidence about the mental mechanisms at work when handling complex verb forms in an L2. Each experimental task is described in detail below; the paragraph that follows outlines the goals of each.
100 The first behavioral experiment was an oral elicitation task (Experiment 1) used to : investigate participants recognition of the accurac y of preterite verb forms after they saw their written infinitive as a stimulus ; to analyze the nature of errors with Spanish regular, irregular, and vowel stem change verbs; and to measure the elapsed time between the presentation of the written stimulus and the oral response. The second behavioral experiment was an oral and visual lexical decision task (Experiment 2) used to : investigate the accuracy of stem recognition of the infinitive verb form after the oral presentation of a preterite form with the s ame or a different stem; and to measure the elaps ed time between the oral stimulus and the button press for regular, irregular, and vowel stem change verbs. The production tasks used were two elicited narratives (Task 1 and Task 2) in which the participant s voice was recorded while narrating the actions another person performed in the past as shown on a storyboard For the first task, participants had to use the verbs printed on the page below each drawing, while the second picturestory sequence did not have accompanying verbs. Only the results from the first task are included in the present study. A s with Experiment 1, the number and rate of correct responses were examined by verb type and the nature of errors was discussed My ultimate goal with these t asks was to provide evidence of dissociations in the processing of inflectional morphology based on the degree of regularity of the verb. By looking at participants RTs, accuracy rates and errors when conjugating and/or recognizing complex Spanish regular irregular, and stem change verb forms from the preterite tense, we should find clear dissociations if these verb forms are handled by and stored in different memory systems. If irregular verb forms are stored in the declarative memory system, RTs to high and low frequency irregular verb forms should differ significantly because a memorized item that is more frequently heard and used will be retrieved and accessed faster than an infrequent one. If regular verb forms
101 are handled by the procedural memory sys tem, however, frequency should not play a significant role because the rule of suffix addition should apply at a similar rate for high and low frequency regular verb forms. Regular verb forms do not need to be stored in the declarative memory fully conjuga ted, as is posited to be the case for irregular verbs, although they can be. Vowel stem change verbs are a special kind of verb in Spanish which take the regular Tense/Aspect/Mood (TAM) suffix but change the vowel of the root when in the preterite tense. T his vowel change is predictable and only applies to Spanish verbs from the third conjugation, i.e., verbs which end in ir in their infinitive form. There will be different predictions for this special group of verbs if we think of this vowel change as a n ew rule to be mastered based on its predictability or if we think of it as a new stem to be memorized because it replaces that of the infinitive form. The literature review presented in Chapter 2 showed no study treating this group of verbs separately. Ins tead, they are usually included in the group of irregular verbs. Psycholinguistic Experiment 1 as well as Production Tasks 1 and 2 focused on the production of Spanish preterite forms, while Psycholinguistic Experiment 2 was interested in the speeded recog nition of Spanish verb stems by Englishspeaking participants. In order to uncover the learning mechanisms that L2 learners at varying proficiency levels employ in the production of Spanish preterite forms, t heir cognitive strategies and the types of mista kes they made when verb forms were used incorrectly were looked at (e.g., overregularizations, no vowel stem change, incorrect TAM suffix or hypercorrection, among others). The frequency with which they made such errors was also examined and those rates w ere compared among the proficiency groups to better understand the developmental route for the acquisition of inflectional morphology.
102 Since Spanish has more inflectional endings than English and different ways to refer to an event in the past Spanish L2 learners ability when prod ucing and accessing Spanish preterite forms was examined as the necessary next step in the ongoing debate about the mental representation and produc tion of inflectional morphology Languages with a wider range of morphological tr ansformations force the debate to expand beyond the boundaries to which it has traditionally been limited. For this reason vowel stem change verbs are considered in the present study as a separate group of verbs, a group which is at the heart of the regularity spectrum: these verbs certainly apply a regular rule for TAM but they also undergo an additional internal transformation. Our classification of these verbs as regular or irregular depends, therefore, on whether we believe this stem change is a new rul e to be applied or a new irregular root to be memorized. In addition to the new group of verbs included, another contribution of this study design is that it is crosssectional in nature, providing snapshots of morphology acquisition by different groups of learners at a particular point in time. This methodology aims to provide the same kind of findings which may have been obtained by the moving picture of a longitudinal study involving a series of observations over a longer period of time using the sa me group of participants. The procedure for data collection, the design of all tasks and materials, participants information and recruitment, as well as data analysis are discussed in this chapter. Data coding is explained in the next chapter, with the r esults f or the corresponding tasks. 3.2. Research Questions The specific research questions that guided this study are as follows: (1) What ty pe of dissociations do English/ Spanish bilinguals evidence based on frequency effects, as observed i n the RTs, accuracy, and errors when handling complex verb forms?
103 (2) What type of dissociations do the different groups of late L2 Spanish learners evidence based on frequency effects, as observed i n the RTs, accuracy, and errors when handling complex verb forms? (3) W hat developmental patterns can we find from the different proficiency groups of late L2 learners when handling complex verb forms of various verb types of high and low frequency? (4) What significant differences can we find between the group of advanced late learners and the group of English/Spanish bilinguals (taking age of initial exposure to L2 Spanish a s the main differing variable) in RTs, accuracy, and errors/behaviors when handling complex verb forms? Since, accordin g to the dual mechanism models, NS s generate regular verb forms through rule application, it is hypothesized that there will be no significant difference s in RTs between high and low frequency regular verb forms in the group of bilingual speakers Irregular verb forms by contrast, are sto red in the lexicon in the declarative memory, which is sensitive to input frequency. RTs for high frequency irregular verb forms should be significantly shorter than those for low f requency irregular verb forms. The group of bilinguals in this study is pre dicted to perform just as NSs would because they are expected to depend more on procedural memory for regular verb forms than late learners. For late learners, linguistic forms computed by procedural memory in the L1 are expected to be dependent to a grea ter extent upon declarative memory in L2, which is sensitive to input frequency. RTs for high frequency regular and irregular verb forms should be significantly shorter than those for low frequency verb forms. However, the greater the amount of practice an d
104 experience with the L2 (operationalized by their proficiency level in the L2 ), the smaller the dissociation, as a result of a better learning of grammatical rules in procedural memory. A ccording to the Cognitive Approac h and the DP model, t he initial st ages of SLA are lexical in nature. It is therefore posited that the pressures for syntactization (or the restructuring process by which lexical concepts acquire syntactic category and subcategorization features) and the use of the procedural memory system will be delayed for late L2 learners. This will have direct consequences on learners behavior when handling complex verb forms. Specifically, more proficient adult late L2 learners will evidence fewer strategies of avoidance, lower regularization rates, l ower irregularization rates, as well as fewer error types, a smaller error frequency, and a greater regular/irregular dissociation than those with low er proficiency levels Since English/Spanish bilingual s will make a greater use of the procedural memory from the very beginning, they should behave more like NS s when handling complex verb forms. They will show a greater regular/irregular dissociation, less avoidance, fewer errors, and lower regularization and irregularization rates than the entire group of late L2 learner s. 3.3. Participants 3.3.1. Participant Selection Two key variables for the present study were age of initial exposure to Spanish (early vs. late) and proficiency in Spanish as an L2 (beginning, intermediate, and advanced). All late Spanish L2 learners reported that they started learning Spanish for the first time at age 14 or later. On the other hand, most English/Spanish bilinguals started learning Spanish from birth. The latest a bilingual started learning Spanish for the first time was at age five. Late learners were recruited and assigned to one of three proficiency levels, based on the Spanish course in which they were currently enrolled at the university. Their proficiency was then confirmed with a proficiency exam. The proficiency exam is described further in section 3.4.7. The group of
105 English/ Spanish bilinguals all had an advanced knowledge of Spanish, which allowed for their comparison with the advanced late learners with respect to the role of the age of initial exposure to the L2. By contrast, the different proficiency groups of late learners allowed for an analysis of developmental changes across proficiency levels 3.3.2. Background I nformation Background information for these groups of participants was obtained in a post te s t FL background form (Appendix A ) at the end of the first session of data collection. This short questionnaire is described in more detail in section 184.108.40.206. Additionally, participants age of initial exposure to Spanish and their ethnicity were collected during the first session of data collection. Of the 15 partic ipants in the beginning proficiency group, 13 were females and two were males8 8 This gender distribution represents the typical one in the Spanish classrooms from which these groups of participants were recruited. Their ages ranged from 18 to 23 with a mean age of 19.13 years. The L1 for 14 of these participants was English, and one participant had Haitian Creole as the L1. There were three African Americans in this group. All beginning participants started learning Spanish for the first time at about age 14 or later. There were seven freshmen, four sophom ores, two juniors, an d two seniors in this group None of them had ever lived in a Spanish speaking country and only one reported socializing regularly with friends or family members outside the class in Spanish. When asked about their use of English and Spanish in several contexts, all participants mentioned that at their current home they use only English. All but one participant use only English with family and at work (whenever applicable), with the exceptional student using mostly English. With friends, nine participants r eported using only English and the remaining six reported using mostly English. Finally, 10 participants use mostly English at school and five of them only
106 English. In sum, their use of Spanish outside the class was scant and limited to some friends or the school context, where they use mostly English. All participants were asked to rate their proficiency in Spanish on a variety of skills using a 7point Likert scale on which a 1 represented very poor and a 7 represented hi ghly proficient (see Appen dix A ). Most beginning participants rated their Spanish skills from low to average, as shown by the mean scores from their responses ( arranged from highest to lowest): 4.27 for reading, 3.60 for writing and listening, 3.53 for vocabulary knowledge, 3.40 for grammar knowledge, and 3.07 for speaking. Of the 15 participants in the intermediate group, there were 12 females and three males. The averag e age of the participants in this group was 19.6 years, ranging from 18 to 22. Except for one native speaker of Mandarin Chinese, the L1 for all participants in this group was English. There were four Asians and one Hispanic in this group and their age of initial exposure to Spanish ranged from 14 to 19. There were two freshmen, four sophomores, six juniors, and three seniors. Only two participants had lived in a Spanish speaking country (see Appendix B for additional information about the participants past travel experiences) and four reported that they socialized regularly with friends or family members in Spa nish. When asked about their use of English and Spanish, all participants reported using only English or mostly English with family, with friends, and in their current home; 14 of them said they use only or mostly English at school; and 11 did so at work. Spanish was used most of the time by only one participant at school, and by another participant at work. This group of participants rated most of their skills in Spanish average, with the exception being reading, which obtained a good, as calculated by the mean scores of their responses: 5.00 for reading, 4.87 for listening, 4.47 for writing, 4.40 for grammar knowledge, 4.13 for vocabulary knowle dge, and 3.80 for speaking.
107 In the advanced proficiency group, there were 13 female and two male participants with a mean age of 22.67 years, ranging from 21 to 40. All participants in the advanced group spoke English as their L1 and all of them were university seniors, with the exception of an assistant professor and a graduate student. There was one African Am erican participant, as well as one participant of multicultural ethnicity. Their age of initial exposure to Spanish ranged from 14 to 20. Most of them (11 participants) had lived in a Spanish speaking country (see Appendix B) but only five reported sociali zing regularly with friends or family members in Spanish. English was exclusively or most often used by most participants in all contexts: with family, at home, and at work by all participants, by 14 participants with friends, and by 13 of them at school. Spanish was used most of the time by only two participants at school, and by another participant with friends. The mean scores for the different skills indicated that most of them rated their Spanish skills from average to good: 5.80 for listening, 5.73 for reading, 5.07 for writing, 4.93 for vocabulary knowledge, 4.73 for speaking, and 4.60 for grammar knowledge. In the bilingual group, there were 12 females and three males. Their ages ranged from 18 to 29 wit h a mean age of 20.60 years. About half of them considered Spanish their L1 and the remaining seven participants reported both English and Spanish as their L1s. All participants were of Hispanic origin and most of them learned Span ish from birth or prior to age five There were two freshmen, two so phom ores, four juniors, five seniors, and two graduate students in this group All of them socialize regularly with friends or family members outside of class in Spanish but only nine had ever lived in a Spanish speaking country (Appendix B) Whe n asked ab out their use of English and Spanish in several contexts, most participants mentioned that they use English only or most of the time in four out of the five contexts: 14 of them with friends, 13 at sc hoo l, 13 at work, and 11 at home. Only one use d mostly E nglish with family. Spanish was
108 used most of the time by 10 participants with family, by two participants at home and at school, and by one participant wit h friends. F our participants reporte d using only Spanish with their family and two participants at ho me. In sum, it was in the family context where these bilinguals use d Spanish only or most of the time. Most bilingual participants rated their S panish skills from go od to advanced,: 6.40 for listening, 5.93 for reading and speaking, 5.73 for grammar kn owledge, 5.60 for writing, and 5.53 for vocabulary knowledge. 3.3.3. Recruitment of Participants Most L2 Spanish learners were enrolled in a Spanish course at the University of Florida at the time of data collection. In order to recruit them, different Sp anish courses were contacted, depending on the proficiency level targeted. In some cases, the information about the study was sent to the instructor to read to the class and collect potential participants emails, but for the most part the researcher visit ed the course at the beginning of a class to introduce the topic of the investigation broadly. For the beginning learners SPN2200 sections were visited which is the first semester of second year Spanish, during the first two weeks of the semester Beginning learners were recruited at the start of their second year Spanish to make sure they completed the second semester of first year Spanish where the preterite tense is introduced. For the intermediate group the researcher visited SPN3300 sections a high intermediate Spanish course with emphasis on grammar and composition, required for those learners major ing /minor ing in Spanish For the advanced group fourth year courses were visited. The topic was introduced orally and the class was provided with the r esearchers contact information or asked to write their email on a sheet of paper for those who wanted to get additional information about the study. Participation was completely voluntary, although in some classes the instructor offered an additional rewa rd for their participation, usually an extra point in the final grade.
109 Bilinguals were informed about the study through an advertisement posted in the independent student newspaper The Alligator which is available for free on campus five days a week. So me advertisements were also placed on various campus sites. Once potential bilingual participants sent an email with their willingness to participate in the study, they were contacted back with further details about the study. Some people were disqualified before taking part in the study simply because they did not meet the minimum prerequisites, namely: (a) speaking both English and Spanish, as a native speaker, (b) being 18 years old or older, and (c) having learned Spanish and English from a very early age in the US or another country where Spanish is not the official means of communication. 3.4. Design and Materials Regularity and verb form frequency are the most important linguistic variables in this study because one of the aims of this research is t o find dissociations between different types of verb forms based on their morphological structure. Fol lowing the predictions of the dual mechanism model s we need to find significant differences in the way learners represent and access complex verb forms i n Spanish as an L2. As a consequence, the corpus of verbs used to address the research questions in this study included 20 regular, 20 irregular, and 20 vowel stem change Spanish verbs in the third person singular preterite form, subdivided into 10 high an d 10 low frequency items, for a total of 60. All 60 test items and 60 fillers were used for the two psycholinguistic experiments (Experiments 1 and 2) but only 24 test items were targeted on each of the elicited narratives (Tasks 1 and 2), as is described later. What follows is an explanation of the test items selection as well as information on the different materials used in the study. 3.4.1. Verb Types Spanish has a rich verbal morphology. Aguirre & Dressler (2006) outlined the schema of any Spanish ver b form as follows :
110 [Root + Thematic Vow el] + Suffix1 (TAM) + Suffix2 ( P N ) The first suffix is the tense, aspect, and mood marker. The second suffix is the agreement marker and it refers to person and number (PN). This suffix holds throughout the conjugatio n except for a few exceptional verbs. In the present study, each verb form was analyzed independently rather than regarding all verb forms of a single verb as belonging to either one or the other verb type. In fact, if we look at the internal structure of individual verb forms, a verb can take regular suffixes as well as having irregular forms and even stem change ones. For example, the verb servir (to serve) is regular in servimos (we serve) [serv i] mos but undergoes a vowel stem change in sirvi ( s /he served) [sirv i] Poner (to put) is usually considered an irregular verb in Spanish, but the verb form ponemos (we put) [pone] m os does not exhibit any irregularity. All test forms were in the 3rd person singular of the preterite tense. I nstead of allowing participants to narrate in the past using the person and number of their choice for the test items, their available choices were constrained for several reasons. First, doing so helped in the comparison of participants responses but, mo re importantly, it forced them to use low frequency verb forms, which may have otherwise been completely absent from their production. The behavior they showed when encountering a low frequency test item is of great importance for this investigation becaus e it can inform us about the mental mechanisms at work when coping with different types of verbs in the L2. Additionally, by selecting the 3rd person singular of the preterite the potential role of automaticity for the 1st person singular verb form was avo ided, as being the first item when accessing the rules for the preterite paradigm, which they have been undoubtedly exposed to in their Spanish classes. Finally, this verb form only takes the TAM su ffix because there is no overt P N suffix.
111 T hree broad categories based on regularity were considered to address th e research questions. First, a given verb form was considered regular i f it added the regular TAM and P N suffixes93.4.2. Verb F requencies to the root while keeping unchanged the root of the infinitive form: (1) Regular: pregunt ( s/he asked ) = [ pregunt ] Second, irregular verb forms were those which replaced the whole verb stem of the infinitive, or at least its root. These are usually referred to as suppletive forms or strong preterites because, although some of the ir endings may still be recognized, the root of the infinitive form and the root in the preterite tense are so different from each other that they cannot be derived by general rules at all as Bauer (2003, p. 48) pointed out. (2) Irregular: In tuvo ( s/he had ), the stem [ten e] was replaced by [tuv ] Finally, vowel stem change irregular verb forms were those which suffered a vowel change in the root while adding the regular TAM suffix. As mentioned before this group of verbs is halfway between t he purely irregular and the purely regular verb forms and different predictions can be proposed depending on whether the internal vowel change is thought of as another rule or as an irregular root. (3) Stem change: sinti ( s/he felt ) instead of the pred icted regular form [sent i] The 60 verb forms chosen as test items for this study were selected based on the averaged frequencies from three corpora: two online databases and a printed frequency dictionary. What follows is a brief description of each corpus including their total number of words and representativeness in modern Spanish, followed by the procedure used to collect and average the frequencies for every verb form, as well as some challenges faced during the process. 9 For a detailed list of the regular suffixes see Aguirre & Dressler (2006, p. 92-93)
112 One resource was the online corpus Corpus de Referencia del Espaol Actual ( Reference Corpus of Modern Spanish or CREA), from the Royal Academy of the Spanish Language, which includes an impressive database of more than 150 million words from the 1970s to the current time. It can be accessed for free at the following web address: http://corpus.rae.es/creanet.html The corpus had a total of 150,778,934 words when frequencies for potential test items were collected; 83,939,527 words (55.67%) were from Spain while 66,759,968 words ( 44.28%) were from Latin American countries; 79,558,459 of these words ( 52.67%) came from books, 68,387,717 words ( 45.37%) from printed media, and 2,832,758 words ( 1.88%) from othe r resources including oral data. The second online corpus was Corpus del Espaol ( Spanish Corpus or CDE ) (2002present), created by Mark Davies and funded by the National Endowment for the Humanities. It can be accessed for free at this web address : www.corpusdelespanol.org. Although the complete corpus includes 100 million words from texts from the 1200s to the 1900s, verb form counts for the present study were obtained from the 20 million words portion of the corpus for the late 1900s. This portion of the corpus contained a total of 20,350,000 words in January 2008. Virtually all texts were from 19702000; a quarter of the words came from the academic register, another quarter from newspapers, another quart er from literature, and the final quarter of the words from oral texts. Approximately 43% of the texts came from Spain while the remaining 57% came from Latin American texts. Finally, the third resource consulted was the dictionary Frecuencias del Espaol ( Spanish frequencies ) published by Almela et. al. (2005). This dictionary was based on the linguistic corpus CUMBRE, which contains more than 15,000 texts for a total of 20,662,306 words. Approximately 65% of the texts came from Spain while the other 35% came from Latin
113 American texts. Two thirds of the texts are written (70% of the Spanish texts and 60% of the Latin American texts) and about one third of the texts are oral (30% and 40% of the texts for Spain and Latin America, respectively). The procedur e for obtaining the frequencies for each verb form was the following: Every specific verb form, such as abri (s/he opened), was searched in the three corpora during the month of January 2008. The number of tokens of every verb form in each corpus was recorded. These raw counts were then transformed into frequencies of occurrence per million words, in order to allow comparability among the different corpora, following this formula: N tokens in Corpus X 1, 000,000 Frequency in Corpus X = Total N words in Corpus X For example, this is how the frequency of the verb form abri ( s /he opened) was obtained for the CREA corpus from the raw counts: Corpus and total N words CREA 150,778,934 CDE (1900s) 20,350,000 CUMBRE 20,662,306 Info recorded N cases Freq./m. N cases Freq./m. N cases Freq./m. Abri ( s/he opened ) 7,380 1,177 982 7,380 1,000,000 Frequency on CREA = 150,778,934 Finally, the frequencies from the three corpora were averaged in order to obtain the frequency which, in this study, is used to determine whether a verb form can be considered as either of high or low frequency. Tables 31, 32, and 3 3 display the t oken count s frequencies per million words from the three corpora, and averaged frequencies for the regular, irregular, and vowel stem change verbs.
114 Table 3 1. Token count, frequencies per million words from the three corpora, and averaged frequency for regular test items High frequency regular verb forms CREA CDE CUMBRE Average d Freq Cases Freq. Cases Freq. Cases Freq Empez (s/he started ) 11,129 73.81 2,020 99.26 1,893 91.62 88.23 Pregunt ( s/he asked ) 12,799 84.89 1,578 77.54 1,897 91.81 84.75 Sali ( s/he went out ) 11,651 77.27 1,661 81.62 1,616 78.21 79.03 Llev (s/he carried ) 11,479 76.13 1,872 91.99 1,359 65.77 77.96 Mir (s/he looked ) 6,168 40.91 1,691 83.10 1,336 64.66 62.89 Tom (s/he took) 8,501 56.38 1,477 72.58 962 46.56 58.51 Llam (s/he phoned ) 7,445 49.38 1,119 54.99 1,067 51.64 52.00 Entr (s/he entered ) 6,993 46.38 1,180 57.99 1,037 50.19 51.52 Abri (s/he opened ) 7,380 48.95 1,177 57.84 982 47.53 51.44 Na ci ( s/he was born ) 5,871 38.94 1,125 55.28 1,014 49.07 47.76 Low frequency regular verb forms CREA CDE CUMBRE Averaged Freq. Cases Freq. Cases Freq. Cases Freq. Viaj (s/he travelled ) 1,706 11.31 292 14.35 158 7.65 11.10 Pag (s/he paid ) 1,810 12.00 161 7.91 144 6.97 8.96 Vendi (s/he sold ) 1,294 8.58 210 10.32 154 7.45 8.78 Esper (s/he waited ) 1,114 7.39 222 10.91 161 7.79 8.70 Olvid (s/he forgot ) 1,235 8.19 198 9.73 159 7.70 8.54 Regal ( s/he gave away) 1,281 8.50 183 8.99 150 7.26 8.25 Invent (s/he invented ) 1,021 6.77 185 9.09 143 6.92 7.59 Escap (s/he escaped ) 1,117 7.41 165 8.11 147 7.11 7.54 Bes (s/he kissed ) 1,041 6.90 141 6.93 155 7.50 7.11 Cant (s/he sang ) 729 4.83 128 6.29 146 7.07 6.06
115 Table 3 2. Toke n count, frequencies per million words from the three corpora, and averaged frequency for irregular test items High frequency irregular verb forms CREA CDE CUMBRE Averaged Freq. Cases Freq. Cases Freq. Cases Freq. Dijo (s/he said ) 106,172 704.16 14,308 703.10 13,170 637.39 681.55 Hizo (s/he did ) 49,738 329.87 6,635 326.04 5,696 275.67 310.53 Tuvo (s/he had ) 34,749 230.46 5,122 251.70 3,864 187.01 223.06 Hubo ( there was ) 18,493 122.65 2,949 144.91 2,452 118.67 128.74 Pudo (s/he could ) 20,157 133.69 2,477 121.72 1,993 96.46 117.29 Estuvo (s/he was ) 17,155 113.78 2,576 126.58 2,175 105.26 115.21 Puso (s/he put ) 15,675 103.96 2,254 110.76 2,149 104.01 106.24 Fue (ir) ( s/he went ) N/A N/A 1,872 91.99 N/A N/A 91.99 Vino ( s/he came ) N/ A N/A 1, 229 60.39 N/A N/A 60.39 Produjo (s/he produced ) 9, 432 62.56 1, 292 63.49 943 45.64 57.23 Low frequency irregular verb forms CREA CDE CUMBRE Averaged Freq. Cases Freq. Cases Freq. Cases Freq. Trajo (s/he brought ) 2, 855 18.94 507 24.91 365 17.67 20.51 Introdujo (s/he introduced ) 1, 653 10.96 448 22.01 226 10.94 14.64 Condujo (s/he drove ) 1, 733 11.49 273 13.42 188 9.10 11.34 Redujo (s/he reduced ) 1, 419 9.41 224 11.01 (< 142) (< 6.87) 10.21 Atrajo (s/he attracted ) 594 3.94 129 6.34 ( < 142) (< 6.87) 5.14 Anduvo (s/he walked ) 773 5.13 88 4.32 (< 142) (< 6.87) 4.73 Tradujo (s/he translated ) 614 4.07 101 4.96 (< 142) (< 6.87) 4.52 Extrajo (s/he extracted ) 549 3.64 83 4.08 (< 142) (< 6.87) 3.86 Contrajo (s/he contracted ) 419 2. 78 95 4.67 (< 142) (< 6.87) 3.73 Cupo (s/he fit ) N/A N/A 44 2.16 (< 142) (< 6.87) 2.16
116 Table 3 3. Token count, frequencies per million words from the three corpora, and averaged frequency for vowel stem change test items High frequency stem change for ms CREA CDE CUMBRE Averaged Freq. Cases Freq. Cases Freq. Cases Freq. Convirti (s/he converted) 6,224 41.28 2,136 104.96 782 37.85 61.36 Muri (s/he died) 8,084 53.61 1,404 68.99 1,216 58.85 60.48 Pidi (s/he asked) 10,404 69.00 1,142 56.12 961 46.51 57.21 Sigui (s/he continued) 7,934 52.62 1,228 60.34 914 44.24 52.40 Sinti (s/he felt) 6,222 41.27 1,284 63.10 768 37.17 47.18 Consigui (s/he obtained) 6,003 39.81 1,087 53.42 591 28.60 40.61 Advirti (s/he warned) 4,867 32.28 485 23.83 369 17.86 24.66 Sirvi (s/he served) 3,762 24.95 493 24.23 399 19.31 22.83 Sonri (s/he smiled) 2,531 16.79 379 18.62 390 18.87 18.09 Repiti (s/he repeated) 2,432 16.13 261 12.83 269 13.02 13.99 Low frequency stem change forms CREA CDE CUMBR E Averaged Freq. Cases Freq. Cases Freq. Cases Freq. Se despidi (s/he said bye) 796 5.28 103 5.06 (< 142) (< 6.87) 5.17 Durmi (s/he slept) 630 4.18 108 5.31 (< 142) (< 6.87) 4.75 Se ri (s/he laughed) 566 3.75 109 5.36 (< 142) (< 6.87) 4.56 Hiri (s/he hurt) 406 2.69 32 1.57 (< 142) (< 6.87) 4.26 Minti (s/he lied) 331 2.20 62 3.05 (< 142) (< 6.87) 2.63 Se visti (s/he dressed) 271 1.80 51 2.51 (< 142) (< 6.87) 2.16 Corrigi (s/he corrected) 327 2.17 38 1.87 (< 142) (< 6.87) 2.02 Midi (s/he measured) 197 1.31 35 1.72 (< 142) (< 6.87) 1.52 Se divirti (s/he enjoyed) 66 0.44 9 0.44 (< 142) (< 6.87) 0.44 Virti (s/he poured) 10 0.07 4 0.20 (< 142) (< 6.87) 0.14
117 While gathering the information on the frequencies for the test items, there were some challenges which could only be addressed by the Corpus del Espaol because this corpus is lemmatized and also tagged for part of speech. The lemmatization feature was helpful in distinguishing those cases where a verb form shared it s spelling with another verb form with a different meaning, as in fue (s/he went or s/he was) which can be the preterite form for the verb ser (to be) or for the verb ir (to go). The part of speech feature was useful to distinguish cases where a to ken could either be a verb form or a noun, as in the case of vino (s/he came or wine) and cupo (s/he fit in a place or occupancy). Distinctive token counts taking lemmas or parts of speech into account could not be retrieved from either CREA or CUM BRE. The primary difficulty with CUMBRE was that some verb forms were so infrequent that they were not even part of the 10,000 most frequent words and there was no way to find out their exact number of cases, except for the fact that they would appear less than 142 times. For those cases, the generic frequency (< 6.87), the frequency for the 10,000th most frequent word in this corpus, was indicated on Tables 3 1 to 33, but was not considered towards the averaged frequency for this study. For these verbs, o nly data from CREA a nd CDE were averaged. 3.4.3. Fillers and their F requencies In addition to the 60 test items, the study included the same number of fillers or distractors They belonged to a different grammatical category, to be precise, nouns, and th e inflectional suffixes they took were those for plural formation. Half of the fillers were masculine nouns and the other half were feminine nouns, of high and low frequency. The procedure and the databases used to obtain their frequencies were the same as those used for the 60 stimulus verb forms. The frequencies in the three corpora and the averaged frequency for the high and low frequency fillers for masculine and feminine fillers can be found in Appendix C
118 3.4.4. Audio R ecordings The 60 stimulus verb forms in the preterite tense and the 60 singular noun fillers were digitally recorded by a native speaker of Spanish for use in the design of Experiment 2. T he researcher wr ote a set of instructions for the native speaker to: (1) read loud, clear an d at a normal speed; (2) leave a small pause between words; (3) not to emphasize any syllable of any word in an unnatural way as well as to keep the same intonation throughout the whole recording; and (4) read the whole list of 120 items from start to finish. T he audio recording took place in Spain and the native speaker of Spanish was a young female. She followed the instructions cl osely and the digital recording with the 120 test items was 3:58 minutes long. Ten additional items were recorded on a different au dio file and at a later time for the practice s ession (last column on Table 3 4). These two audio files were edited using the computer program Audacity; each individual word was cut and saved as a different audio file. That way the audio recordings for each single item could be used in the design of Experiment 2. In order to assess the overall quality of the audio recordings, two native speakers of Spanish volunteered to listen to them under experimenta l conditions. They visited the Language and Cognition Lab at the University of Florida. Each native speaker listened to different sets of verb stimuli presented over the headphones. They were instructed to listen to the individual words and write them down on a piece of paper. There was 100% accuracy on thei r responses and they did not provide any additional comments as to the quality of the audio. The audio recordings were deemed appropriate to be included in the design of the second psycholinguistic expe riment, to which we turn now.
119 3.4.5. Design of the P sy cholinguistic E xperiments Before designing the experiments, the test items as well as the fillers were randomized for each experiment separately. The automated formula =rand() was used in Microsoft Excel software and the first randomized order of items w hich fulfilled the following criterion was kept: there could not be more than three items of the same type together, i.e., there could not be four low frequency regular ver bs one after the other. T his pseudo randomization technique using the same list of r andomized items for all participants provided me with better comparability of results. T he psycholinguistic e xperiments were designed using PsyScope X10For Exp eriment 1, participants saw either Spanish verbs in their infinitive form or nouns in the center of a computer screen which remain ed there indefinitely. Words were color coded: verbs were displayed in a blue font and nouns in a black font both in a big font size Participants w e re instructed to provide the 3rd person singular preterite form of verbs or the gender of nouns orally as fast as possible. The list of practice items were included f irst (second column in Table 34), followed by a screen to allow participants to ask questions before starting the experiment, and then the pseudorandomiz ed list of test items and fillers. Instructions were not presented over the computer screen Instead, they were written on a printed laminated page for the participants (Appendix D), which gave the researcher time to set up the experiment while participant s read Dr. Wind Cowles helped me to write the scripts, i.e., o nce she was provided with the randomized list of items for each experiment and the individual audio files, she wrote the scripts of the experiments in PsyScope X based on the needs for the present study. 10 PsyScope X is a program used to design and conduct psychological experiments that runs on Apple Macintosh computers. It was developed at Carnegie Mellon University by Jonathan Cohen, Matthew Flatt, Brian MacWhinney and Jefferson Provost for Mac OS 9 in the 19 90s. PsyScope X is a recent universal version which runs on Intel processors.
120 them. The pseudo randomization of test items and fillers appears in th e protocol checklist (Appendix E). Table 3 4. Practice items for the psycholinguistic experiments Items on instructions for Experiment 1 Practice items for Experiment 1 Items on i nstructions for Experiment 2 Practice items for Experiment 2 Beber (to drink) Comer ( to eat ) Comprar ( to buy ) Hablar ( to speak ) Cocinar ( to cook ) Estudiar ( to study ) Flor (flower) Querer ( to like ) Escuchar ( to listen ) Saber ( to know ) Caer ( to fall ) Ver ( to see ) Impedir ( to impede ) Computadora ( computer ) Elegir ( to choose ) Invertir ( to invest ) Sugerir ( to suggest ) Mesa ( table ) Cuaderno (notebook) Lmpara ( lamp ) Cancin ( song ) Accin ( action ) Car ro ( car ) Zapato ( shoe ) Profesor ( teacher ) Pan ( bread ) Experiment 1 was designed so that t est items would disappear from the computer screen at the precise moment the microphone detect ed any sound. Then, PsyScope X waited for 1,500 millisecond s (ms.) before displaying a green circle at the center of the screen, representing the buttonbox participants had to press to move on to the next item. It wa s a self paced experiment and participants had the control to decide when to move on to the next i tem. This action was assigned to the green button at the center of the buttonbox. PsyScope X was set up to record the RT or elapsed time from the precise moment each item appeared on the computer screen until the beginning of the participant s response, i .e. the software detecting any oral waveform. PsyScope X w as used simultaneously with the recording feature of Audacity to collect not only their RTs but also the ir oral responses for later analyses of the accuracy of the response and the types of errors made As for the noun distractors, they appeared in the singular and participants ha d to say into the microphone their gender, whether by saying masculine or feminine, or by using only the defini te article which would precede t hem i.e., el vs. la.
121 For Experiment 2, participants listened to either a Spanish verb in the preterite form or a Spanish singular noun over the headphones while the computer screen displayed a red cross at the center of the screen to direct participants attention there. Then either the corresponding verb in the infinitive form or the plural noun was displayed on the screen and remained there indefinitely Participants w e re instructed to listen closely to the words and then read the written words and to decide as fast as possible if the two words had the same stem or a different stem. Their responses were identified by pressing one of two buttons at each extreme of the button box: the red button for the different stem option and the purple button for the same stem option. The list of practice items were included first (last column in Table 3 4), followed by a screen to allow participants to ask questions before starting the experiment, and then the pseudorandomized list of test items and fillers. Instructions were also writ ten on a printed laminated page for the partic ipants (Appendix D ). Experiment 2 was designed so that the written words would disappear from the computer screen at the precise moment the participant pressed one of the two buttons mentioned. Then, PsyScop e X waited for 1,500 m s before displaying a text that said Ready? at the center of the screen, representing the moment at which participants decided when to move to the next item by pressing any button from the button box. PsyScope X was set up to record t he RT or elapsed time from the precise moment the written words appeared on the screen until the participants pressed one of the two buttons No other type of data was co llected during this experiment. 3.4.6. Design of the Elicited Narratives In addition to the psycholinguistic experiments just described, the study also incorporated two elicited narratives (Tasks 1 and 2). Both required the use of a visual prompt to elicit participants responses on the different types of Spanish preterite verb forms prese nted in a picture story sequence. Due to the limited space of a page, each picturestory sequence contained
122 24 drawings (about six per row, four rows per page) for a total of 48 picture stimuli, out of the 60 test items. Each sequence elicited an even numb er of verb types, i.e., eight regular, eight irregular and eight vowel stem change verbs, of high and low frequency. For each visual prompt, first the actions were drawn on a white page by hand before scanning them. Then they were edited, colored, and sh aded using the Adobe Photoshop CS computer software. Finally, the 24 drawings were arranged on a page trying to display a logical sequence of events, with the help of some brief text dialogues in Spanish. For the first elicited narrative, participants were told to use the verbs provided on the page (see Appendix F). Therefore, below each drawing participants had the verb for that action in its infinitive form. For the second elicited narrative, participants were free to use the Spanish verb of their choice for the actions depicted (see Appendix G). The hope, however, was that they would use the verbs targeted in the study and depicted in the drawings. For each elicited narrative the participant was asked to select among four choices. The task of the research er during the data collection was to pretend to guess which picture story sequence the participant selected once s/he finished the narrative. A pilot test for the second picturestory sequence the one without the verbs written on the page, was conducted to evaluate the clarity of the drawings and to check if they elicited the intended verb forms. Four college students of Spanish at the intermediate level who were not participants in the study were instructed to write a preterite verb form for each drawin g. At least one student used the verb form being elicited for every drawing except one, which showed a plane with one engine on fire N one of the students used the target verb form tuvo (had) in the expression tuvo un accidente (had an accident). Inste ad, they tried to use an equivalent Spanish verb for the English verb to crash As suggested by these students, when asked about the
123 difficulty of that item, that drawing was changed for a different one showing a woman having an accident with her car. 3.4. 7. Proficiency Exams Besides carefully selecting late learners from specific Spanish courses at the University of Florida, it was necessary to confirm that they had been assigned to the appropriate proficiency level. It is no t surprising to find advanced learners with less advanced knowledge and skills than expected at their proficiency level or beginning learners who could easily be taking an intermediate level course instead. Therefore, a proficiency exam was included in this study to en sure that all par ticipants were assigned to the proper group. After considering the different tests available, a Spanish exam created by native speakers of Spanish in a country where Spanish is the official language was selected T he same exam for all learners was not se lected; instead, an exam which was nuanced to the specific skills and abilities of different groups of participants was preferred A variety of skills through an objective scoring system was also preferred rather than conducting an oral proficiency intervi ew to be rated by others. Finally, the exam had to be easy to conduct in any location and it also had to be easy to grade. The exams used to obtain the different Diploma s de Espaol como Lengua Extranjera (DELE) from the Instituto Cervantes were the ones s elected for this study. These exams are conducted on a regular basis in more than 100 countries to measure the Spanish knowledge of international students and they are accredited by the Ministry of Education, Social Politics, and Sports of Spain. T he DELE exams from the examination session of the month of November 2007 were used which are available online for free. A ll sections from the beginning, intermediate, and advanced exams were used except for the writing and oral tasks. The DELE score used for thi s study was the aggregate from the scores on the grammar vocabulary listening, and reading
124 sections. Every exam was edited in Microsoft Word so that it fit on fewer pages than the originals, but they included all relevant sections, including instructions and pictures. The audio clips were edited to delete long empty spaces in the original waveform. T he exams were conducted following the guidelines from the exam creators very closely, exc e pt for providing long pauses between the different sections. All sec tions are multiple choice activities, which facilitated grading. ( For f urther information about the specific types of activit i es, the format of the readings, or the characteristics of the audio recordings see sample activities in Appendix H or the whole e xam at : http://diplomas.cervantes.es/index.jsp At this web address, there is also access to the complete exams and guidel ines on how to deliver them ) The DELE proficiency exam results corresponded we ll with the proficiency groups of the present investigation, namely: beginn ing, intermediate, and advanced As a consequence, no adjustments in placement were made based on these findings, and no participants were eliminated from the study. 3.4.8. Other M aterials B esides the materials described i n the previous sections, the present study also involved the creation of additional materials, whether to fulfill ethics requirements or to ensure comparability among participants responses during data collection This section briefly describes the following additional materials: the informed consent form, the FL background form, the instructions for the participants, tags for the color coded button box, and the protocol checklist. 220.127.116.11. The informed consent form This document is required before conducting research with human participants at the University of Florida It guarantees participants that their identity will be kept safe, informs them of their rights, and ensures them of their right to withdraw from the study at an y moment. As shown in Appendix I the informed consent form included the following sections: title of the
125 project, purpose of the study, the research methodology, the time required, the anticipated risks, the potential benefits and compensat ion for taking part in the study, a statement on confidentiality of the data collected, contact information for the principal investigator, the supervisor, and the institutional review office, and statements about the rights of the participants. Participan ts received a printed copy of the document during the first session of data collection, before taking part in the study. 18.104.22.168. Foreign language background form The foreign language background form (see Appendix A ) was used at the end of the first data collection session. It was used to collect a wide array of personal information from the participants, such as their age, their self rated proficiency on different skills in the L2, their study abroad experiences (see Appendix B) and the like. It was also used to check if they were familiar with the 60 test items in the study through a translation activity. 3.4.8 .3. Instructions for e xperiments As mentioned in section 3.4.5., the instructions for the psycholinguistic experiments were not included in the design of the experiments in PsyScope X. Instead, they were written in English and printed on a laminated page for the part icipants to read (see Appendix D). These pages gave the researcher time to set up the experiments, save previous data when necessary, check the protocol checklist as well as the volume level and the microphones sensitivity. All participants received the same instructions and they could read them as many times as they wanted, which could have been difficult to implement if delivered tho ugh a series of computer screens. 22.214.171.124. Tags for the button box For Experiment 1, participants decided when to move to the next item by pressing the green button on the button box (see Figure 31).
126 Figure 31. Next tag on top of the green button i n the ioLab Systems button box A laminated tag containing the word next was placed on top of the green button during the data collection. In Experiment 2, participants had to press either the red or the purple button located at each side of the button bo x. The following tags were placed over the corresponding buttons to make very clear the choice they were selecting: 126.96.36.199. Protocol checklist The procedure for the first session of data col lection was pilot tested before officially starting with the data collection. A Spanish student was asked to visit the Language and C ognition L ab and complete all tasks. The procedure outlined in section 3.5. was followed, and some changes were made to the protocol, most notably the inclusion of a protocol checklist, in order to avoid some mistakes in the future. During the pilot test session it became evident that it was very easy to forget any given step because the study involved several tasks and different methods of data collection. For SAME s tem DIFFERENT stem
127 example, I completely forgot to open Audacity and record the participants voice during Experiment 1 because I was paying more attention to setting up the experiment and making sure that the microphone was placed correc tly. At a different moment, her cell phone rang during the data collection. This was considered something that may interrupt the data collection, even if experiments are self paced and participants had the choice of deciding when to move to the following i tem. Therefore, the protocol checklist form was created (see Appendix E ) with several objectives in mind: (1) to prevent forgetting a necessary step in the data collection, (2) to remind participants about turning off their phones and making sure that the microphone is placed correctly over th eir mouths, (3) to have a form o n which to mark items during Experiment 1 that would need a RT adjustment due to potential false starts, hesitations, coughs, or the like, and (4) to have a place to write participants responses during Tasks 1 and 2 to pretend that I was really trying to guess the picture story sequence they selected. In the end, this form facilitated the data collection and there were virtually no events which disrupted the data collection for the 60 pa rt icipants included in the study. 3.5. Data Collection Data were collected on two different occasions for each participant. The first data collection session was con ducted in the Language and Cognition Lab at the University of Florida, while the second session took place in a seminar room on campus. Data collection started on the 29th of May, 2008 and was completed on the 21st of October, 2008. This section describes the procedure followed during each data collection session, outlined in Figure 32. 3.5.1. First Session: Main D ata and B ackground I nformation The first data collection session lasted for about an hour on average and was conducted on an individual basis. My main objective was to collect data on participants behavior in
128 relation to Spanish inf lectional morphology through a series of controlled tasks under laboratory conditions. The researcher was present at all times and the procedure for each data collection session was as follows. Figure 32. Outline of the two data collection s essions Once the participant arrived at the Language and Cognition Lab at the time and date previously arranged by email, s/he signed her/his name on a sign in page. Then participants received two copies of the informed consent form that assured their con fidentiality. They signed one copy for me and received the other for their own files. Most participants had already received a digital copy of this document via an email attachment before deciding whether to participate in the study or not; as a result, mo st of them were familiar with the information contained in this document. Their names were not used in any part of the study; instead, they were assigned a number which was used for all data files and data recordings. These files were PROCEDURE (A) First Session (one on one, about 1 hour total): 1. Sign in page 2. Informed consent form 3. The two Psycholinguistic experiments (one of two orders) Experiment 1 Experiment 2 Oral elicitation task O ral and visual lexical decision task 4. Elicited narrative 1 (drawings with written test items) 5. Elicited narrative 2 (only drawings) 6. Foreign language background form 7. $10 compensation in cash (B) Second Session (mostly in small groups, from less than 1 hour to 2 hours): 1. Proficiency exam 2. $10 compensation in cash
129 always stored on password protected computers. Before signing the informed consent form and starting the tasks, they had the opportunity to ask questions. While the participant was reading the informed consent form, the researcher completed an experiment log sheet for the lab and decided on the order of presentation of the psycholinguistic experiments. The two experiments had the same stimuli and fillers but the role o f potential priming effects was minimized by reversing the order of their pres entation to the participants wit hin the proficiency groups Priming effects occur when there is an increase in retrieval speed and accuracy due to prior exposure. Seven participants per proficiency group completed Experiment 1 before moving on to the second one, and eight participants pe r proficiency group completed these experiments in the reverse order11Prior to data being collected, participants were instructed to turn off their cellular phone s to prevent possible interference with the voice detection system They were informed that the session would consist of two experiments with the computer, the headset with microphones, and the buttonbox, as well as two picturestory descriptions in Spanish. Then the following I n order to provide the greatest consistency among the participants, the protocol checklist was closely followed (Appendix E ) during each individual session of data collection. T he sa me eMac personal computer was used to collect the data from all participants It was located in one of three areas in the Language and Cognition Lab a small sound proof room in which participants were seated at a comfortable distance from the computer scr een. For Experiments 1 and 2, participants had to wear a headset with microphones and use a color coded button box from ioLab Systems (see Figure 33). 11 S ome primin g effects were found in Experiment 1: faster RTs to high frequency irregulars and higher accuracy scores to irregular and stem change verb forms of high and low frequency in the intermediate group as well as faster RTs to regular and irregular verbs of low frequency in the advanced group. However the frequency effect results without those participants in the condition Experiment 2 before Experiment 1 were no different than with the entire sample Therefore, the results of the entire sample are included in the results and discussion chapters
130 introductory text was read to them before making sure that the microphone was at an adequ ate distance from their mouths: Today you will be asked to complete two activities on this computer. But, first of all, relax and sit at a comfortable dist ance from the computer screen. You need to wear a set of headphones with a microphone attached to it. Make sure that it is placed correctly over your head and that the microphone is in front of your mouth. Then the data collection started for either Expe riment 1 or Experiment 2, as predetermined. Figure 33. Language and Cognition Lab at the University of Florida eMac computer ioLab Systems button box headset with microphone sound proof wall
131 188.8.131.52. Experiment 1 Irrespective of the order of presentation of the two psycholinguistic tasks, the procedure for Experiment 1 was as follows. Each participant received a single laminated page (see Appendix D) with written instructions for this task. These were written in English and the main steps were highlighted. Participants had ample time to read the instructions while the experimen t was set up. Participants had time to ask questions before starting the experiment. After making sure that they were wearing the headphones with the microphone placed correctly over their mouths, the Start Recording button was pressed in Audacity and th e program was minimized. Then the button Run was pressed in Psyscope X and the participants code number was typed in the participant number window, as shown in Figure 3 4 Figure 34. Screenshot of PsyScope X initial windows
132 The experiment start ed wi th a set of 10 practice items (second column in Table 3 4). At the end of the set, par ticipants had time to ask questions. When deemed necessary, participants were instructed to speak louder into the microphone for the rest of the experiment. And for nouns only, they were told they could simply say el or la ( the, masculine and feminine, respectively) because it is a shorter word than masculine or feminine, which conveys the same information. Then they started the experiment. For each of the 120 expe riment items, PsyScope X recorded the time from the onset of the written stimulus onscreen to the onset of participants oral response. However, the program took into account any waveform via the voice detection software, whether it was a cough or the part icipants final response. In order to correct these artifacts or inaccurate calculations, for each test item the researcher observed very closely whether or not the written stimulus disappeared from the screen at the precise moment t he participant produced her/his oral response. That information was written on th e protocol checklist (Appendix E ) as either correct ( ) or incorrect ( ). Test items were highlighted on the protocol checklist, while this information was not recorded for the filler items. In the data coding stage of this study, the protocol checklist was used to correct the RTs computed by PsyScope X for the false starts, i.e., the items for which participants provided multiple answers, those cases of verb forms produced with long initial syllable s while buying extra processing time or similar behavior that could bias the results. Once the participant provided an answer for the last experiment item, PsyScope X was closed. Then the voice recording was stopped, the project was saved with the file nam e XXaudiolab and as an .aiff audio file. Finally, Audacity was closed before moving on to the next task.
133 184.108.40.206. Experiment 2 Again, irrespective of the order of presentation of the two psycholinguistic tasks, the procedure for Experiment 2 was as fol lows. Each participant received a singl e laminated page (see Appendix D ) with written instructions for this task. These were written in English and the main steps were highlighted. Participants had ample time to read the instructions while the experiment w as set up. Participants had time to ask questions before starting the experiment. They were also informed that in the instructions there was a mistake and that the tags were placed correctly on the button box (see Figure 35), which was the reverse assignm ent from the instructions. After making sure that they were wearing the headphones, the button Run was pressed in Psyscope X and the participants code number was typed in the participant number window. Figure 35. Same Stem and Different Stem tag s on top of the purple and red buttons, respectively, in the ioLab Systems button box The experiment start ed with a set of 10 practice items after which participants had time to ask questions. They were asked if the volume level was appropriate and adjust ed it for the rest of the experiment when necessary. They were also instructed to inform me about any errors that
134 they were conscious of during th e experiment. Finally, it was recommended that they place both hands on the buttonbox close to each button fo r faster responses. Then they started the experiment. Participants completed this task at their own pace. On the protocol checklist, any conscious error they informed me of during the task was noted. Once they pressed a button for the last experiment item, PsyScope X was closed. 220.127.116.11. Task 1 For each elicited narrative, four laminated color copies of the corresponding picturestory sequence were made. The participants were asked to pick one as if there were different picturestory sequences. This decept ion technique was employed to ensure communicative relevance for the task, i.e., to promote the need to provide a complete description of all the actions depicted in the drawings, as well as to allow comparability among all participants. The following intr oductory text was read to them from the protocol checklist: Please, pick one of these pages but do not show it to me. It will contain a picture story sequence of a character that you need to narrate to me. Please, tell me in Spanish what this person did i n the past. You have to use the verbs provided for each drawing in the preterite tense. My task will be to guess which picture story you have just selected once your narration is complete. You need to provide as much information as possible because all the picture story sequences are very similar and they only differ in a couple of drawings. Do not disclose the name of the character to me, which is at the right bottom of the page, until your narration is complete. Instead, refer to the character as L or EL LA. The participant had time to ask questions while Audacity was being set up. Once it was checked that they were wearing the headphones with the microphone placed correctly over their mouths, the Start Recording button in Audacity was pressed and the program was minimized for the remainder of the data collection, including the second elicited narrative (Task 2). Since the researcher already knew the verbs the participants were required to use, the verb forms the participants produced for each test item were written in the protocol checklist. A
135 table was created with big cells, each containing the Spanish verb partici pants had to use (see Appendix E ). Their responses were written on this chart while the participants were talking to the researcher, as if the researcher were truly trying his best to guess the characters name. Their talking was not interrupted if they skipped any given item but it was interrupted if they started to use a different tense than the preterite for several consecutive items18.104.22.168. Task 2 At the end of the narration the researcher pretended to guess the characters name. The voice recording on Audacity was not stopped. Instead, the researcher moved on to the second elicited narrative without a break. Again, for the second na rrative the participants were asked to pick one of the four picturestory sequences provided. Even though another introductory text was created to read to them, participants were usually told that the instructions remained the same as in the previous task except that they could use the verb of their choice that best described the action depicted in each drawing. This time, the verbs they chose for each drawing were also written down by the researcher Once the participant finished her/his oral narration, th ey were asked if the name of the character was Luisa and after they confirmed that the voice recording was stopped, the project was saved with the file name XXaudiodescr as an .aiff audio file. Then Audacity was closed. Once the second elicited narrati ve was completed, participants were asked to fill out a foreign language background form (Appendix A ). While they were completing this questionnaire, a receipt was written out in the participants name and the $10 bill was prepared. This time was also used to get a copy of the data files just collected (two audio files and two Excel files) and to write a brief entry about the participant in an Excel file at the labs main PC. 12 These corrections were made in English and participants were not asked to go back and redo the previous test items. They simply were asked to go on from that point in the preterite.
136 This file was used to record the participants ID and to report any problem observ ed during the data collection overall. When the student gave me the form, they were handed the cash reward and thanked for participating in my study and completing the first session of data collection successfully. As mentioned in section 3.4.6., the hop e with Task 2 was that participants would use the verbs targeted in the second picturestory sequence. However, that was not always the case, which hindered the comparison of results among participants and groups. Of the 1,909 tokens coded for this task, o nly 1,348 tokens (or 70.61%) were verb forms used for the drawings of the second picturestory sequence; the remaining tokens were new verbs added to make a more descriptive narrative. And from those 1,348 tokens, only 763 tokens (56.60%, or 39.97% of the total) were verb forms with the elicited verbs under investigation. Therefore, less than half of the data coded were the verb forms being elicited. There were many cases in which a given participant provided two or more options for the same test item. For example, participant 5 used se puso enfermo ( s he got sick) and tuvo fiebre (she had temperature) for drawing 17 (see Appendix G) instead of the elicited verb form estuvo enferma (she was sick). Her/his responses are correct bu t they are not the ones I was looking for. Furthermore, claims cannot be made that s/he was trying to avoid the targeted verb form by using her/his choices because her/his responses and the elicited verb forms are all irregular verb forms of high frequency. As a consequence, when a participant failed to use an elicited verb in this task, one cannot claim that the participant did it as a strategy of avoidance for that given test item. It is exactly the nature of this task that made it difficult to compare it to the other three. By itself, Task 2 is a very informative task on verbal morphology production by L2 learners, but it allow ed too much freedom to the participants, which arguably, impeded the comparison of results with the other
137 data collection tasks in this study. Consequentl y, it was decided not to include the results from this task in the rest of the study in order to allow a more consistent comparison of results and triangulation of data from Experiments 1 and 2 as well as Task 1. 3.5. 2. Second Session: Proficiency E xams Th e main goal for the second session was to have an objective score on participants competence in Spanish as an L2, separate from the proficiency group they were assigned to based on the Spanish course they were currently taking at the university. For this session, participants were generally assigned to groups in order to be as expedient as possible. Some students were met individually; the biggest the group surveyed was with eight participants simultaneously. The time participants needed to complete the pr oficiency exam depended on the proficiency exam in question. At the beginning level, participants usually completed the exam in less tha n an hour to an hour and 15 minutes. Intermediate participants completed their exam in 6090 minutes. Participants who t ook the advanced exam needed more than an hour and a half but generally less than two hours. Two participants in the advanced group requested additional time for the completion of the exam. One of them had to return on a later day to complete the readings because she did not have enough time during the two hour window of data collection. The other advanced participant was allowed to take the last pages of her exam home and complete the reading activities without the help of any external resource. She promis ed she did it as requested and provided her last resp onses a couple of days later. For each participant, once s/he arrived at the office at the time and date previously arranged by email, s/he started the corresponding proficiency exam. Her/his ID was wri tten at the top of the first page. Participants usually started with the grammar and vocabulary sections. After the first 15 minutes, and when most participants for that day ha d already arrived, they were asked if they wanted to complete the listening acti vities before continuing with the exam and then the
138 audio files were played using a portable CD player at a comfortable volume. They then completed the rest of the exam. Once each participant finished all the activities, s/he handed me the exam and receive d $10 in compensation. Before leaving the office, they were thanked for participating in the study. 3.6. D ata Analysis S tatistical analys es were carried out with SPSS version 13 .0. for Windows The alpha level for significance in all statistical tests was preset at p < .05, although any statistical result at p < .01 is also reported. The following chapter provides the results of these analyses. First, descriptive statistics and frequencies were used to report the participants production accuracy on the test items for all tasks. These data also helped in evaluating participants errors by providing information about the cognitive strategies that most learners per proficiency group used when coping with the harder test items. Secondly, and only f or the two psycholinguistic experiments a series of independent samples t tests were performed to evaluate the comparisons between verb types of high and low proficiency and to determine the statistical significance of the comparisons. Independent samples t tests wer e also used for all tasks to examine potential differences in the accuracy and mean RT s to test items between the advanced and bilingual groups. Thirdly, one way analysis of variance ( ANOVA ) tests were used to test results for differences among the three proficiency groups of late L2 learners. Additionally, in order to assess which groups obtained significantly different results from which other groups of late L2 learners, the follow up Tukey post hoc test results were checked. Finally, Pearson bivariate correlations were run to test the linear relationships between different variables. These correlations were used to test whether there was a relationship between
139 the frequencies of the test items in the Spanish language (as measured by the three corpora described in section 3.4.2.) and participants accuracy, independently of the items verb type.
140 CHAPTER 4 RESULTS 4.1. Introduction In this chapter the main results from the different tasks are presented including the RT results, the accuracy scores, and the types of errors and/or cogn itive strategies used by Spanish L2 learners. Results are not only descriptive but also include findings from the statistical tests employed At the end of this chapter, section 4.4. briefly discusses whether the results obtained from the different tasks w ere similar to one another and how this triangulation of data can inform the present investigation. The significance of these results and the account of how these data can help in providing answers to the research questions are described in detail in the n ext chapter. 4.2. Psycholinguistic Experiments First, the results of the two psycholinguistic experiments are discussed The order of presentation of Experiments 1 and 2 was counterbalanced across proficiency gro ups, as mentioned in section 3.5 .1. 4.2.1. Results from Experiment 1 There were three main goals of Experiment 1: (a) to compare the elapsed time between the visual stimuli onscreen and the participants oral response for regular, irregular, and vowel stem change verbs, of high and low frequency; (b) to examine participants accuracy in the production of the third person singular preterite tense forms elicited; and (c) to examine participants errors and observed behaviors when dealing with complex verb forms for which they did not know the correc t answer. Each participants data for Experiment 1 were recorded into a single Excel file which included the following information for each token (N = 60): (a) participants code, (b) participants proficiency level, (c) item number, (d) verb type of high or low frequency, (e)
141 participants raw RT, (f) participants adjusted RTs, w hen applicable (see below), (g) participant s final oral response, and (h) participants behavior and/or error type, if applicable. There were a total of 3,600 tokens (60 tokens 60 participants). Even though participants were instructed to produce only their final responses into the microphone, there were cases in which they monitored their responses, paused after lengthening the first syllable of the test item and before provid ing their final response, provided multiple answers, and the like. For that reason, some of the initial RTs had to be adjusted before being entered into the app ropriate data cells. Audacity software was used to measure the additional amount of time between the start of the first oral soundwave produced by the participants and the start of the soundwave corresponding to their final response, as shown in Figure 4 1. Figure 41. Screenshot of Audacity software In this way, a more precise elapsed time between the presentation of the visual stimuli and participants final response is calculated. For example, in Figure 4 1, if A was the beginning of a false start and B the start of the final response, the amount of time measured in milliseconds from A to B was added to the initial RT given by the computer for that test item. Only 3.97% (143/3,600) of the RTs for the total number of forms elicited from all groups of participants had to be adjusted, distributed as follows: in the beginning group 4% (36/900), in t he intermediate
142 group 4.44% (40/900), in the advanced group 3.44% (31/900), and in the bilingual group 4% (36/900). The main verb types adjusted in the beginning and intermediate groups were the irregular and the stem change verbs, whereas most of the adju sted forms in the advanced and the bilingual groups were the irregular verbs. Before running the appropriate statistical tests, the mean RTs and accuracy rates for each participant and for each condition were computed The average RTs in m s for the correc t responses were calculated by verb type and verb type frequency a nd the average accuracy in the oral responses to the different verb types was also calculated, by verb frequencies. 4.2 .1.1. RT results First the statistical results for the RTs for the different verb types across proficiency groups are presented. For the most part, these results did not provide evidence in favor of the expected dissociations and/or frequency effects proposed by dual mechanism models A series of independent samples t tes ts were performed to evaluate the comparisons between verb types of high and low frequency per proficiency group and to determine the statistical significance of the comparisons. These results are displayed in Table 4 1. In the beginning group of late lear ners, the RTs to low frequency items were greater than to the high frequency items of regular and stem change verb forms: 2732 and 2275 m s for high frequency regular verbs and stem change verbs, respectively, and 3340 and 2901 ms. for the corresponding gr oup of low frequency items, in the same order. However, the only significant results observed in the group of beginning learners were found in the group of irregular verbs, although in an unexpected direction. Beginning learners took significantly longer ( t = 2.180, df = 22, p = .040) to produce the correct response to high frequency (M = 3587.59, SD = 1487.74) irregular verbs than to low frequency (M = 2428.24, SD = 913.41) irregular verbs.
143 Table 4 1. Independent samples t test results for the RTs for the correct responses by all groups of learners in Experiment 1: Comparison between test items of high and low frequency. Mean RT (SD) Verb Type Frequency Beginning (N = 15) Intermediate (N = 15) Advanced (N = 15) Bilingual (N = 15) Regular High 2732.21 (1671.25) 1584.46 (459.32) 1420.08 (667.57) 1531.66 (697.33) Low 3340.27 (3357.23) 1688.52 (561.74) 1637.33 (764.85) 1482.13 (615.39) Irregular High 3587.59* (1487.74) 2819.26 (1336.31) 2935.77 (2183.70) 2058.36 (960.09) Low 2428.24* (913.41) 2290.58 (138 8.94) 3364.26 (3215.69) 3033.42 (2326.59) Stem Change High 2275.58 (852.46) 1952.38 (1033.09) 2438.34 (2105.27) 1622.06* (702.90) Low 2901.20 (1395.22) 2675.67 (1361.28) 2200.78 (1494.70) 2251.40* (934.02) *p < .05, two tailed, equal varianc es assumed In the i ntermediate and advanced groups, the differences in the RTs for the high and low frequency tes t items of all verb types failed to reach statistical significance, as shown in Table 4 1. It shows that these groups of learners took about the same amo unt of time to respond to the test items per verb type, irrespective of the elicited test items frequency in Spanish. For the most part, the pattern of results from the intermediate group was similar to those observed in the beginning group. They took lon ger to respond to low frequency than to high frequency regular verbs and stem change verbs but it took them somewhat longer to respond to high frequency than to low frequency irregular verbs, as shown in Table 4 1. The group of advanced learners took longe r to respond to low frequency than to high frequency regular verbs and irregular verbs, but they took slightly longer to respond to high frequency than to low frequency stem change verbs. The group of English/Spanish bilinguals showed a different pattern of results. It took them significantly longer ( t = 2.085, df = 28, p = .046) to respond to low frequency (M = 2251.40, SD = 934.02) than to high frequency (M = 1622.06, SD = 702.90) stem change verbs. Further, it took them longer to respond to high freque ncy than to low frequency regular verbs,
144 while the opposit e was found for irregular verbs; i.e., it took them longer to respond to low frequency than to high frequency irregular verbs, as shown in Table 41. However, neither of these last two results reach ed statistical significance. In sum, RT results from Experiment 1 did not provide strong evidence of dissociations between the different verb types in line with the predictions of the DP model. For the most part, there were not significant results in the R Ts to the test items of high and low frequency on the different verb types in this study. However, beginning learne rs took significantly longer to respond to high frequency than to low frequency irregular verbs, while bilinguals took significantly longer t o respond to low frequency than to high frequency stem change verbs. 1420.08 2732.21 1584.46 0 1000 2000 3000 4000 5000 Beginning Intermediate Advanced Reaction Times (milliseconds) *p < .05, **p < .01, n.s. = non significant F igure 42. One way ANOVA with Tukey post hoc test on the RTs for high frequency regular verbs comparing the beginning, intermediate and advanced late learners for Experiment 1 In order to track RT differences across proficiency levels in the groups of late learners, a series of one way ANOVAs with a Tukey post hoc test were run for the RT results to the diffe rent verb types. Results revealed evidence of dissociations between the RTs to regular verb forms on the one hand (Figure s 42 and 4 3) and the RTs to irregular and stem change verb n.s. **
145 forms on the other hand, since there were significant results for regular verb forms which were absent for irregular and stem change verb forms. Figure 42 shows that the RTs to the high frequency regular verbs d ecreased significantly across proficiency levels (N = 45, F = 6.666, p = .003) Beginning learners took 2732 ms. to respond correctly to high frequency regular verbs while intermediate and advanced learners took a significantly shorter amount of time (1584 ms. and 1420 ms., respectively). Intermediate le arners took significantly less time (MD = 1147.75, SEM = 391.56, p = .015) to respond to high frequency regular verbs than the beginning learners, while the RT differences between the advanced and the beginning learners were even more robust (MD = 1312.13, SEM = 391.56, p = .005) than those between the beginning and the intermediate 1637.33 3340.27 1688.52 -1000 0 1000 2000 3000 4000 5000 Beginning Intermediate Advanced Reaction Times (milliseconds) (*) p < .1 n.s. = non significant Figure 43. One way ANOVA with Tukey post hoc test on the RTs for low frequency regular verbs comparing the beginning, intermediate and advanced late learners for Experiment 1 n.s. ( ) ( )
146 In addition, Figure 43 shows t hat the RTs to the low frequency regular verbs also decreased significantly across proficiency levels (N = 45, F = 3.470, p = .040). However, this time the Tukey post hoc results were not as stro ng as for high frequency verbs. Beginning learners took 3340 ms. to provide the elicited low frequency regular forms while the intermediate (1688 ms.) and advanced (1637 ms.) learners took a signifi cantly shorter amount of time (MD = 1651.75, SEM = 735.50, p = .075 and MD = 1702.94, SEM = 735.50, p = .065, respectively). As mentioned before, none of the patterns of results between the three groups of late learners reached statistical significance for irregular and stem change verb forms. These patterns indicated that the intermediate learners took a shorter amount of time to respond to irregular v erbs and stem change verbs of high and low frequency than the beginning learners, as shown in Table 41. However, the advanced learners took longer than the intermediate group to respond correctly to irregular verbs of high and low frequency and to high fr equency stem change verbs, but they took less time to respond to low frequency stem change verbs thus providing no evidence of a discernable pattern of change among proficiency groups Finally, the RT results between the two groups of advanced speakers we re compared i.e., between the advanced late learners and the English/Spanish bilinguals. The latter usually took less time to respond correctly to any verb type under investigation except for high frequency regular and low frequency stem change ve rb forms The t test results indicated that the RT differences between these two groups of highly proficient learners in L2 Spanish failed to reach statistical significance. In sum, when tracking significant differences between the groups of participants in their RT results to the different verb types, there were only significant differences in the RTs for regular verbs, of high and low frequency, between the beginning groups of learners on the one
147 hand and the intermediate and advanced groups on the other hand. There were no significant differences in RT results for any verb type between the group s of advanced speakers 4.2 .1.2. Accuracy results In this section, the percentage s of correct responses for each verb type of high and low frequency across the proficienc y groups are reported. First, I explore possible frequency effects among the verb types for each group of participants individually and then I examine which groups of participants increased their accuracy on the different verb types in relation to the oth er groups. This analysis allows us to track developmental changes across proficiency levels and to check if there are differences in the production of advanced speakers with different learning experiences. Table 4 2 displays the percentage of correct respo nses for each verb type, of high and low frequency, in the oral production of the diffe rent groups of participants. Table 4 2. Accuracy rates on the different verb types of high and low frequency for Experiment 1 for all groups of participants (3,600 tokens). Percentage (correct responses/test items per group of participants and verb type) Regular Irregular Stem Change High Low High Low High Low Beginning (N = 15) 84.67% (127/150) 83.33% (125/150) 60% (90/150) 22.67% (34/150) 42.67% (64/150) 26% (39/150) Intermediate (N = 15) 98 % (147/150) 91.33% (137/150) 77.33% (116/150) 45.33% (68/150) 57.33% (86/150) 46.67% (70/150) Advanced (N = 15) 98% (147/150) 96.67% (145/150) 86.67% (130/150) 44.67% (67/150) 70.67% (106/150) 56% (84/150) Bilingual (N = 15) 99.33% (149/150) 98% (147/150) 92.67% (139/150) 67.33% (101/150) 85.33% (128/150) 78.67% (118/150) There are three different patterns we can pinpoint from Table 4 2. In the production of late learners, the accuracy rates for regular verbs were generally higher than those for irregular verbs and stem change verbs. Only the advanced learners obtained a similar accuracy rate for high frequency irregular verbs (86.67%) compared to the accuracy results for regular verbs by any proficiency group. Besides this difference in the accuracy scores between regular verbs on
148 the one hand and irregular verbs and stem change verbs on the other hand, late learners always showed higher levels of accuracy on all verb forms, irrespective of their verb type and frequency, if we look at the accuracy rates from one pro ficiency level to the next one. The only exception to this trend was the low frequency irregular verbs, for which the group of advanced learners obtained slightly lower scores than the group of intermediate learners (from 45.33% to 44.67%). However, this difference in accuracy scores was minimal, less than 1%. The last pattern of results we can observe in Table 4 2 is the accuracy scores for any verb type by the English/Spanish bilinguals which were usually higher, i n some cases reaching almost perfect accuracy (99.33% of correct responses for high frequency regular verbs). These speakers produced the elicited verb forms of two thirds of all verb types with more than 85% accuracy, the except ions being low frequency ir regular verbs and stem change verbs (67.33% and 78.67%, respectively). In order to check if there were frequency effects in the three types of verbs, a series of independent samples t tests were run taking frequency as the grouping variable. There were cl ear dissociations for the most part between irregular verbs on the one hand and the other two verb types on the other hand. An unexpected result was the frequency effects for regular verbs in the group of intermedi ate learners only. T he statistical results for each proficiency group are described in the following paragraphs. In the group of beginning learners there were frequency effects in the accuracy rates for irregular verbs only, as shown in Table 4 3. Beginning learners produced irregular verb forms w ith significantly better accuracy ( t = 3.530, df = 28, p = .001) if they were of high frequency (M = 60, SD = 31.40) than if they were of low frequency (M = 22.67, SD = 26.31). Conversely, there were no frequency eff ects in the accuracy scores of regular o r stem change verb forms.
149 In the group of intermediate learners there were frequency effects in the accuracy rates for both regular (t = 3.035, df = 28, p = .005) and irregular ( t = 2.788, df = 28, p = .009) verbs, which is evidence against dissociations of frequency effects based on regularity. Table 4 3 shows that intermediate learners obtained greater accuracy results for high frequency regular verbs (M = 98, SD = 4.14) and high frequency irregular verbs (M = 77.33, SD = 22.19) than for low frequency re gular verbs (M = 91.33, SD = 7.43) and low frequency irregular verbs (M = 45.33, SD = 38.52), respectively. There were no frequency effects in accuracy rates for stem change verbs. Table 4 3. Independent samples t test results for accuracy rates by all groups of learners in Experiment 1: Comparison between test items of high and low frequency. Mean accuracy rate (SD) Verb Type Frequency Beginning (N = 15) Intermediate (N = 15) Advanced (N = 15) Bilingual (N = 15) Regular High 84.67 (25.03) 98.00** (4.14) 97.33 (7.04) 99.33 (2.58) Low 83.33 (28.45) 91.33** (7.43) 96.67 (7.24) 98.00 (4.14) Irregular High 60.00** (31. 40) 77.33** (22.19) 86.67** (13.97) 92.67** (12.23) Low 22.67** (26.31) 45.33** (38.52) 44.67** (37.20) 67.33** (22.51) Stem Change High 42.67 (26.58) 57.33 (26.04) 70.67 (25.20) 85.33 (22.00) Low 26.00 (23.24) 46.67 (27.17) 56.00 (26.94) 78.67 (19.22 ) **p < .01 two tailed, equal varianc es assumed In the group of advanced learners there were frequency effects in the accur acy rates for irregular verbs only (t = 4.094, df = 28, p = .000). This group of participants produced more correct irregular verb forms if they were of high frequency (M = 86.67, SD = 13.97) than if they were of low frequency (M = 44.67, SD = 37.20). Ther e were no frequency effects for regular or stem change verb forms.
150 Again, in the group of bilingual s there were frequency effects in the accuracy scores for irregular (t = 3.830, df = 28, p = .001) verb forms but not for regular or stem change verb forms, as shown in Table 43. This group of participants also obtained significantly higher accuracy results for high frequency irregular verbs (M = 92.67, SD = 12.23) than for low frequency irregular verbs (M = 67.33, SD = 22.51). In order to look for possible developmental c hanges and differences in ultimate attainment based on age of initial exposure to Spanish, several statistical tests were run on these results and we found significant differences in the accuracy rates between the various groups of late learners and also b etween the two groups of advanced speakers First, the significant results found in the accuracy scores across the proficiency groups of late learners are described followed by those between the advanced groups of speakers A series of one way ANOVAs were run with a Tukey post hoc test to compare the accuracy rates on all verb types of high and low frequency between the three groups of late learners. The significant results are displayed in Figures 45 to 48 The first set of significant results was found for stem change verbs, irrespective of their frequencies. There were significant differences between the groups (N = 45, F = 4.370, p = .019) in their accuracy rates for high frequency stem change verbs, and the Tukey post hoc test indicated that the sign ificant differences (MD = 28, SEM = 9.48, p = .014) were found between the beginning (M = 42. 67, SD = 26.58) and the advanced groups (M = 70.67, SD = 25.20), as shown in Figure 44. Howe ver, these results were more robust for the low frequency stem change verbs.
151 70.67 57.33 42.67 0 20 40 60 80 100 120 140 Beginning Intermediate Advanced Accuracy Rates (0-100) *p < .05, n.s. = non significant Figure 44. One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency stem change verbs comparing the beginning, intermediate and advanced learners for Experiment 1 56 46.67 26 0 10 20 30 40 50 60 70 80 90 100 Beginning Intermediate Advanced Accuracy Rates (0-100) **p < .01, n.s. = non significant Figure 45. One way ANOVA with Tukey post hoc test on the accuracy rates for low frequency stem change verbs comparing the beginning, intermediate and advanced learners for Experiment 1 n.s. n.s. ** n.s. n.s.
152 There were also significant differen ces in the accuracy results for low frequency stem change verbs between the groups of late learners (N = 45, F = 5.293, p = .009). The Tukey post hoc test indicated that the differences (MD = 30.00, SEM = 9.44, p = .008) were found between the beginning ( M = 26, SD = 23.24) and the advanced (M = 56, SD = 26.94) proficiency groups only, as shown in Figure 45. There were also significant results between these three groups in the accuracy rates for high frequency regular verbs and for high frequency irregula r verbs Fi rst, for both high frequency regular verbs and high frequency irregular verbs there were significant differences between the groups, as shown by the one way ANOVA results (N = 45, F = 3.949, p = .027 and N = 45, F = 4.924 p = .012, respective ly). 98 98 84.67 0 20 40 60 80 100 120 140 Beginning Intermediate Advanced Accuracy Rates (0-100) (*) p < .1 n.s. = non significant Figure 46. One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency regular verbs comparing the beginning, intermediate and advanced learners for Experiment 1 ( ) ( ) n.s.
153 For high frequency r egular verbs th e Tukey post hoc results, which reached the significance cutoff point, were found between the beginning and the intermediate groups as well as between the beginning and the advanced groups of late learners (MD = 13.33, SEM = 5.48, p = .050 for bot h cases), as shown in Figure 4 6. However, these differences were more robust for high frequency irregular verb forms ( N = 45, F = 4.924, p = .012) and they were found between the beginning and the advanced groups only: M = 60, SD = 31.40 for the b eginning group and M = 86.67, SD = 13.97 for the advanced group (MD = 26.67, SEM = 8.62, p = .010 Tukey post hoc results) as shown in Figure 4 7. 86.67 77.33 60 0 20 40 60 80 100 120 140 Beginning Intermediate Advanced Accuracy Rates (0-100) *p < .05, n.s. = non significant Figure 47. One way ANOVA with Tukey post hoc test on the ac curacy rates for high frequency irregular verbs comparing the beginning, intermediate and advanced learners for Experiment 1 The last set of statistical tests ru n for the accuracy scores in Experiment 1 were a series of independent samples t tests to asse ss if there were significant differences between the advanced groups of speakers Figure 48 shows that the only results that reached statistical significance n.s. n.s.
154 between those groups were found for the low frequency stem change verb forms (N = 30, df = 28, t = 2.653, p = .013). 0 20 40 60 80 100 120 140 High Reg Low Reg High Irreg Low Irreg High SCh Low SCh Accuracy Rates (0-100) Advanced Bilingual *p < .05, (*) p < .1 n.s. = non significant two tailed, equal variances assumed Figure 48. Independent samples t test results for accuracy on the different verb types for Experiment 1: Comparison between the advanced and bilingual groups (N = 30) Additionally, the differences in the accuracy f or the low frequency irregular verb forms between these groups (N = 30, df = 28, t = 2.019, p = .053) approach ed statistical significance. Again, these results point to dissociations between regular verb forms on the one hand and irregular verb forms on the other hand. The bilingual group consistently obtained higher accuracy for low frequency irregular verbs (M = 67.33, SD = 22.51) and low frequency stem change verbs (M = 78.67, SD = 19.22) than the group of advanced learners (M = 44.67, SD = 37.20 for irregular verbs and M = 56, SD = 29.94 for stem change verbs). 4.2.1 .3. Error types and participant s behaviors From the different variables used in the coding stage of this study, the variable with the greatest range of options was the participants behavior and/or error type for the test items. A total of 23 opt ions were classifi ed for the present analysis, in order to capture as much n.s. ( ) n.s. n.s. n.s.
155 information as possible from the participants resp onses. Table 44 presents a brief summary with a specific example to illustrate each behavior/error. This taxonomy was also used in the coding of this type of data for the Production Task considered (section 4.3 .2.). Table 4 4. Summary of behaviors/errors used for the coding of oral data for Experiment 1 and Tasks 1 and 2 # Behavior/error Example Correct form with translation 0 None Cant (s/he sang) Cant ( s/he sang ) 1 Overregularization Reduci Redujo ( s/he reduced ) 2 No vowel change Senti Sint i ( she /he felt ) 3 Nonfinite form/Noun *Para regalar Regal ( in order to give a present ) 4 Hypercorrection Puso > puse Puso ( s/he put ) 5 Buying time Puuuuuudo Pudo ( s/he could ) 6 Different PN suffix Llam Llam ( s/he called ) 7 No respons e Naci ( s/he was born ) 8 Wrong stress Extraj Extrajo ( s/he extracted ) 9 # 2 + Incorrect suffix Vert Virti ( s/he poured ) 10 Different tense Sigo Sigui ( s/he continued ) 11 # 1 + # 5 Veeeeeni Vino ( s/he came ) 12 Monitor Menti > minti Minti ( s/he lied ) 13 Wrong root Esquep Escap ( s/he escaped ) 14 # 12 + # 6 Pongo > puse Puso ( s/he put ) 15 # 6 + # 5 Puuuuude Pudo ( s/he could ) 16 TAM of another class Vini Vino ( s/he came ) 17 Nonexistent/Other verb Se puso Pudo ( s/he was able to ) 18 # 12 + # 1 Andi > and Anduvo ( s/he walked ) 19 # 5 + # 8 Repiiiiito Repiti ( s/he repeated ) 20 # 2 + # 8 Correg o Corrig i ( s/ he corrected ) 21 # 1 + # 8 Reduc o Redujo ( s /he reduced ) 22 # 2 + # 5 Me eeeee di Midi ( s/he measured ) This section provides a descriptive account of the most prevalent cognitive strategies and types of errors that the different groups of participants produced in their oral responses to the test items during Experiment 1. First, the most prevalent behavior and/or error for each verb type and group of participants are described, and then the internal structure of the error and/or the relevance of the observed behavior are explained. Table 4 5 summarizes the main observed behavior/e rror by these groups of parti cipants, arranged by verb type.
156 Table 4 5. Summary of participants main behaviors and errors for all groups of learners (N = 60) and by verb type in Experiment 1. Rate of the main error/behavior (number of errors with the mai n error type/total number of errors per group of participants and verb type) Regular Irregular Stem Change High Low High Low High Low Begin ning (N = 15) Error type 10 10 10 1 2 2 Error rate 34.78% (8/23) 36% (9/25) 23.33% (14/60) 41.38% (48/116) 46. 51% (40/86) 46.85% (52/111) Intermediate (N = 15) Error type 6 13 1 1 2 2 Error rate 66.67% (2/3) 53.85% (7/13) 23.53% (8/34) 43.90% (36/82) 53.13% (34/64) 45% (36/80) Advanc ed (N = 15) Error type Single instances 16 1 1 2 2 Error rate 60% (3/5) 35% (7/20) 68.67% (57/83) 77.27% (34/44) 78.79% (52/66) Biling ual (N = 15) Error type Single instances 16 1 & 10 13 1 2 2 Error rate 66.67% (2/3) 27.27% (3/11) 42.86% (21/49) 90.91% (20/22) 90.63% (29/32) When there is only a single instance for an error type, this error type is not reported as it does not represent a trend in the results. In this section, an asterisk represents an incorrect verb form given the instructions of the experiment (provide the third person singular preterite), regardless of the morphological structure of the verb form. For example, deca (s/he used to say) is considered an incorrect verb form for Experiment 1 although it is a correct verb form for the imperfect tense for the verb decir (to say). For a detailed explanation of the different codes used in the data coding of this experiment, see Appendix J and for a list of the 60 test items and their corresponding elicite d preterite form, see Appendix K ; in the interest of brevity, the following sections only report participant s errors and/or behaviors but the corresponding correct preterite forms are not included. 13 If more than one main behavior on a specific verb type, their rates were the same for all behaviors reported in the corresponding cell, e.g., there were five examples of behavior 10 as well as another five examples of behavior 16 for high frequency irregular verbs by the advanced group of late learners.
157 22.214.171.124.1. Errors associated with regular verb forms Most beginning learners used present forms instead of the elicited preterite forms for regular verbs, irrespect ive of their frequency. The greater the proficiency level, the lower the error rate in the production of regular verb forms and the more difficult it becomes to identify behavioral/error patterns. However, most of the occasional mistakes produced by partic ipants in higher proficiency levels (including the bilingual group ) had to do with the attachment of a wrong TAM suffix or with confusion in the syllable structure of the verb root. In the group of beginning learners, the most prevalent behavior for regul ar verbs was to provide a verb form in another tense (error 10), as shown in Table 4 5. There were eight cases in the group of high frequency regular verbs, half of which were used in the third person singular present tense (* entra s/he enters, llama s /he phones, mira s/he looks, pregunta s/he asks) instead of their corresponding preterite forms There were also three examples of salgo (I go out). The last example was the present subjunctive form pregunte (that s/he asks). In the group of high frequency regular verbs, there were also six examples of nonexistent forms or verb forms that belonged to other verbs, such as: empezco ( s /he started) or deba (from the verb deber ought to) instead of llev (s/he carried). In the group of low frequency regular verbs, beginning learners produced seven forms in the third person singular present tense (* canta s/he sings, escapa s/he escapes, espera s/he waits, inventa s/he invents, olvida s/he forgets, paga s/he pays, regala s/h e gives away) and another two in the present subjunctive tense (* pague that s/he pays, viaje that s/he travels). The other main behaviors were four nonexistent forms and four test items used in nonfinite forms. The nonexistent forms were: basa and beso instead of bes (s/he kissed), and espro and esp rio instead of esper (s/he waited). The nonfinite forms were all used in the infinitive: cantar (to sing),
158 *escapar (to escape), inventar (to invent), and viajar (to travel), in plac e of the preterite verb forms listed in Appendix K The group of intermediate learners showed fewer and qualitatively different errors from the ones used by the group of beginning learners. When they had to produce a preterite for high frequency regular verbs, they erred in only three examples. Out of these errors, two were the use of a wrong PN suffix: llam and tom (I called and I took, respectively). In the group of low frequency regular verbs, out of the 13 errors, there were seven cases of wro ng root, i.e., the participant provided the appropriate TAM suffix but made a mistake in the syllable structure of the verb root. For example: vindi (s/he sold) or esquep ( s /he escaped). Another error with this group of verbs with more than one exa mple was the use of a different tense, with the verb vender (to sell): venda (imperfect) or *vendo (present). Advanced learners were highly accurate in the production of preterite regular forms. They had only three mistakes in the group of high frequen cy regular verbs: (1) the use of a different PN suffix, as in tom (I took), (2) a verb form in the present tense instead of the preterite, as in *salgo (I go out), and (3) the use of a TAM suffix from another verb class: abr (s/he opened). This t hird type of error was the most frequent one with low frequency regular verbs too. Advanced learners made three of these mistakes, out of the five for low frequency regular verbs. There was an example of vend (s/he sold) and two examples of inventi ( s/he invented). Bilingual speakers made virtually no errors with regular verb forms. They had only one example of a wrong TAM suffix in the group of high frequency regular verbs: *miri (s/he looked). Additionally, in the group of low frequency regular verbs there were two examples of
159 that error type, out of the three they made: olvidi (s/he forgot) and *inventi (s/he invented). 126.96.36.199.2. Errors associated with irregular verb forms When participants did not know the correct preterite form for lo w frequency irregular verbs, they usually attached a regular ending to the stem of the infinitive, in the hope that their response would be the verb form being elicited. For irregular verbs of high frequency, the use of verb forms in anothe r tense was also one of the most prevalent error types, besides overregularizations. Specific examples are provided in the following paragraphs. The most prevalent error with high frequency irregular verbs by the beginning learners was to use a verb form in a different tense. Most incorrect forms were verbs used in the present tense and in the first person singular instead of the third person singular. For example, there were three examples of hago (I do), there were two cases of produzco (I produce), two examples of pongo (I put), two digo (I say) forms, and so on. The next error type with a total of 13 examples was to provide a nonexistent form or to use a verb form that belongs to a different verb (error type 17). Some of the nonexistent forms provided we re: vingi ( s/he came ), *pongi (s/he put), etc. Some examples of verb forms taken from verbs other than the test item were hecho (done) for the test item haber (there is /are ) or the preterite form puso (s/he put) for the verb poder (to be ab le to). The last most frequent error type that these learners made with high frequency irregular verbs was overregularization, with a total of 12 examples. However, for the most part this error was limited to the test item producir (to produce), which s even participants overregularized to produci (s/he produced). Beginning learners had a lot of difficulty producing correct irregular preterite forms for low frequency irregular verbs; they were correct in only about one out of five examples, as shown i n Table 4 2. The most frequent error with this type of verbs was overregularization with or without a correct stress assignment to
160 the verb form produced, i.e., error types 1 and 21, for a total of 48 and 24 examples for each of them. Most overregularizati on examples were: seven cases of *introduci (s/he introduced), six cases of atrai ( s/he attracted ) of conduci (s/he drove) and of *contrai ( s/he got ), as well as five cases of cabi (s/he fit) and of reduci (s/he reduced). Most of th e overregularization examples with an additional incorrect stress assignment were the following: four cases of traduco (s/he translated), of cabo (s/he fit), or redcio /* reduc o (s/he reduced), and three cases of ndo ( s/he walked ) and the di fferent versions of incorrect stress assignment for *xtraio / *extr io/* extrao (s/he extracted). Intermediate learners also overregularized most low frequency irregular verbs but they produced most high frequency irregular verbs in incorrect regularized verb forms. In the group of high frequency irregular verbs, overregularization was limited to the verbs producir and venir where there were four cases of produci (s/he produced) as well as three examples of veni (s/he came). The next most frequent error type with high frequency irregular verbs was error 21, or the combination of overregularization and wrong stress assignment, as in haco (s/he did) or veno (s/he came). There were a total of six of these errors. And the last most frequent err or type with high frequency irregular verbs, with a total of six cases as well, was the use of a different PN suffix in the preterite tense. The error was to provide the first person instead of the third person singular preterite form. There were three cas es of estuve (I was), one fui (I went), one dije (I said), and one puse (I put) instead of the corresponding elicited forms. Most intermediate learners overregularized the low frequency irregular verbs when they did not have a trace memory to any irregular verb form with the feature [+ past]. For example, seven participants said *and ( s/he walked ), six participants produced *contrai (s/he got), and there were four cases of conduci (s/he drove) and of *traduci (s/he translated). With this
161 group of verbs, intermediate participants also produced 18 forms in another tense, mostly in the present tense and with the first person singular PN suffix. They produced four atraigo (I attract) and four traigo (I bring) forms as well as two conduzco (I drive) and two *introduzco (I introduce) forms. Finally, intermediate learners also produced 17 examples of overregularization with wrong stress, these being the three cases of *redcio /* reduco (s/he reduced), the most recurrent example of this error type. Advanced learners also regularized irregular verbs of hi gh and low frequency as the most prevalent error for this type of verbs. Interestingly, in the group of high frequency irregular verbs, seven participants regularized the verb prod ucir as in produci (s/he produced). No other high frequency irregular verb was overregularized. The other main error with high frequency irregular verbs was to provide a verb form in another tense. There were a total of six examples: two first person s ingular present forms (* hago I do and pongo I put), two third person singular present forms (* hace s/he does and viene s/he comes), and two singular imperfect forms (* pona s/he used to put and *vena s/he used to come). With the group of low fre quency irregular verbs, the most prevalent error type made by intermediate learners was overregularization, with a rate of 68.67% (57/83) from the total number of errors. All 10 low frequency irregular verbs were overregularized and the advanced learners attached an incorrect regular suffix to the verb stem, which was in agreement with the verb class. Consequently, the advanced learners attached the regular PN marker to the irregular verb andar as in the five examples of and (s/he walked), and t hey attached the regular suffix i to the rest of the low frequency irregular verbs because they belonged to the second and third verb classes, as in the six examples of *extrai (s/he extracted) or the five examples of conduci (s/he drove).
162 The gro up of English/Spanish bilinguals did not have many problems with high frequency irregular verbs. Out of the 11 erroneous forms they produced, they had three cases of verb forms in another tense (* haba there used to be, *tena s /he used to have and pr oduzco I produce) and three cases of overregularized forms (* produci (2) s/he produced and *poni s/he put) In the case of low frequency irregular verbs, about half of the errors (21/49) were examples of regularized forms. Eight out of the 10 test items were regularized in this group of verbs, and the most frequent errors were five examples of atrai (s/he attracted) and four examples of and (s/he walked). The other most common error with low frequency irregular verbs was error 17, with a tot al of nine examples. There were two erroneous forms for each of the following test items: andar to walk, atraer to attract, caber to fit, and extraer to extract. For example, one participant produced the nonexistent form extradujo (s/he extracted ), and another participant used the preterite form hubo (there was) as the response to the test item andar (to walk). 188.8.131.52.3. Errors associated with stem change verb forms A quick look at Table 45 is enough to verify that the recurrent error type by all groups of participants with high and low frequency stem change verbs was to keep the root of the infinitive unchanged. V owel stem change verbs require, besides the attachment of the appropriate TAM suffix an internal vowel change. The lack of this vowel change was made frequently and at very high rates, especially by the bilingual group, e.g., in more than 90% of the error examples. In the group of high frequency stem change verbs, beginning learners forgot to change the vowel of the verb root in most test items: adv e rti (8), *conv e rti (7), sonr e i (7), s e gui (5), and so on. The other main error type in the group of high frequency stem change verbs was number 20, or the combination of no vowel change plus wrong stress assignment. For example, t here were five cases of adverto ( s/he warned ), four cases of sent o (s/he felt), four cases
163 of converto (s /he transformed), etc. The most prevalent error with low frequency stem change verbs by the beginning group was not to change the vowel o f the verb root in the preterite. There were nine cases of m e di, eight cases of *corre gi, six cases of he ri and v e rti five cases of *m e nti and se v e sti four cases of *div e rti and so on. The next most frequent error with low frequency stem cha nge verbs was the combination of no vowel change and wrong stress, with a total of 33 examples such as *se divert o (s/he enjoyed) or *menti (s/he lied). The most prevalent errors the group of intermediate participants made with high frequency stem change verbs were to omit the vowel change and also the combination of no vowel change and incorrect stress assignment. Out of the 34 cases of no vowel change there were six examples of adverti (s/he warned) and of converti (s/he converted), as wel l as four examples of consegui (s/he obtained), of repeti (s/he repeated), of segui (s/he followed), and of servi (s/he served). The main examples with error type 20, i.e., no vowel change and wrong stress, were five cases of adverto (s/h e warned), and four cases each of *converto (s/he converted), of sento (s/he felt), and of servo (s/he served). Similar error types to those for high frequency verb forms were observed in the group of low frequency stem change verbs. There were 36 cases of no vowel change and 30 cases of no vowel change with wrong stress. The main examples of no vowel change were seven corregi (s/he corrected) and verti (s/he poured) forms each six mention (s/he lied), five *se diverti (s/he enjoye d) and four *heri (s/he hurt) forms instead of the elicited preterite forms. The main examples of behavior 20 were four cases of ver t o ( s /he poured), as well as three cases of *hero (s/he hurt), mento (s/he lied), se vesto (s/he dressed up) and *corrego (s/he corrected).
164 Even though the group of advanced learners produced fewer errors when providing stem change preterite forms than the other groups of late learners (44 and 66 errors for high and low frequency items in that order), m ost incorrect examples were of error type 2, or the omission of the corresponding vowel change. Most vowel change omissions in the group of high frequency stem change verb forms were limited to two test items : *adverti (s/he warned) and converti (s/h e converted), with eight and nine examples, respectively. And in the group of low frequency stem change forms there were eight examples of se diverti (s/he enjoyed) and of *medi (s/he measured) each, seven examples of heri (s/he hurt) and six examples of *corregi (s/he corrected) and of verti (s/he poured) each instead of the appropriate preterite forms. When bilinguals made an error with stem change verb forms, of high and low frequency, they usually forgot to make the appropriate vowe l change in the stem of the verb. In the group of high frequency stem change verbs there were five cases of converti s/he converted, four cases of adverti s/he warned, and three cases each of consegui s/he obtained, of segui s/he continued, and of servi s/he served. And in the group of low frequency stem change verbs, the verbs which were produced without the corresponding vowel change by most participants were diverti (6) s/he enjoyed, verti (6) s/he poured, *corregi (4) s/he corrected, medi (4) s/he measured, and menti (4) s/he lied. In sum, if we consider the different error types by most groups of participants for the various verb types of high and low frequency, the following patterns can be identified: (1) regul ar verbs, irrespective of their frequency, were produced in another verb tense by most beginning learners, while the most frequent errors by the rest of the proficiency groups were the use of a different PN suffix or a TAM suffix from other verb classes; ( 2) all groups of
165 participants overregularized most low frequency irregular verbs and for high frequency irregular verbs most participants either overregularized or produced those verbs in another tense, usually the present tense; and (3) all groups of part icipants failed to change the vowel of the verb stem for vowel stem change verbs of both high and low frequency. 4. 2.2. Results from Experiment 2 The re were two primary goals for Experiment 2 : (a) to investigate participants accuracy of stem recognition o f verbs in their infinitive form after the oral presentation of a preterite form with either the same or a different stem; and (b) to compare the elapsed time between the oral stimuli and the button presses for regular, irregular, and vowel stem change ver bs, of high and low frequency across groups. Each participants data for Experiment 2 were recorded in a single Excel file which included the following pieces of information for each token (N = 60): (a) item number, (b) verb type (high or low frequency), ( c) written stimuli which were displayed on the screen, (d) event tag or condition in relation to the expected response, (e) button pressed, and (f) participants RT. The percentage of correct responses to the test items was computed by verb type and verb type frequency. Before running the appropriate statistical tests, the average RT in milliseconds for the correct responses was also calculated, again, by verb type and verb type frequency. 4. 2.2.1. Accuracy results The results for the accuracy rates on the test items are presented before presenting the results for the RTs on the same items. Experiment 2 was purely a recognition task and the participants did not have to provide a written or an oral response. They simply had to press one of two buttons to dec ide as quickly as possible depending on whether the oral stimulus had the same or a different stem from that of the written stimulus which was presented on the computer screen. There were only two options and all groups of learners obtained virtually perfect scores,
166 as shown in Table 4 6. The only exception to this was in the low frequency stem change verbs, among beginning and intermediate learners, where accuracy was slightly lower (88.67% for both groups). Table 4 6. Independent samples t test results fo r accuracy rates (as measured by the button pressed) by all groups of learners in Experiment 2: Comparison between test items of high and low frequency. Mean accuracy rate (SD) Verb Type Frequency Beginning (N = 15) Intermediate (N = 15) Advanced (N = 15) Bilingual (N = 15) Regular High 98 (5.61) 98.67 (3.52) 98.67 (3.52) 98.67 (3.52) Low 99.33 (2.58) 100 (.00) 99.33 (2.58) 100 (.00) Irregular High 94 (10.56) 98.67** (3.52) 97.33 (5.94) 98.67 (3.52) Low 92 (11.46) 92.67** (7.04) 95.33 (5.16) 98 (4.14) Stem Change High 98.67** (3.52) 99.33** (2.58) 99.33* (2.58) 100 (.00) Low 88.67** (11.87) 88.67** (11.87) 96* (5.07) 98.67 (3.52) *p < .05 **p < .01 two tailed, equal varianc es assumed A series of independent samples t tests on the accuracy rates from the different verb types were run, taking frequency as the grouping variable. These tests were run for each proficiency group individually and, as shown in Table 4 6, results indicated that there were significant differences for stem change verb form s in the three groups of late learners and also for irregular verb forms but only for the intermediate group. Beginning learners obtained very high accuracy results for regular and irregular verb forms, with an accuracy rate in the button presses that rang ed from 92% to 99.33%. However, there were frequency effects for stem change verb forms (see Table 4 6). Beginning learners were significantly more accurate ( t = 3.128, df = 28, p = .004) at identifying the verb stems of high frequency stem change verbs (M = 98.67, SD = 3.52) than those of low frequency stem change verbs (M = 88.67, SD = 11.87).
167 In the group of intermediate learners there were frequency effects not only for stem change verb forms ( t = 3.400, df = 28, p = .002) but also for irregular verb f orms ( t = 2.953, df = 28, p = .006) as shown in Table 4 6. Intermediate learners identified the verb stems of irregular verbs and of stem change verbs of high frequency with greater accuracy (M = 98.67, SD = 3.52 and M = 99.33, SD = 2.58, respectively) th an those of test items of the same verb types but of low frequency (M = 92.67, SD = 7.04 and M = 88.67, SD = 11.87, in the same order). These frequency effects were not present in the group of regular verbs. In the group of advanced learners there were frequency effects for stem change verb forms only ( t = 2.269, df = 28, p < .031) They identified the stems of stem change verb forms of high frequency (M = 99.33, SD = 2.58) at a significantly better rate than those of stem change verb forms of low freque ncy (M = 96.00, SD = 5.07), as shown in Table 46. There were no frequency effects for regular or irregular verb forms. Finally, in the group of bilinguals there were no frequency effects in the accuracy rates to any verb type, as shown in Table 4 6. Thes e results may be explained by the high accuracy rates the bilinguals obtained in all verb types, irrespective of their frequency. These accuracy rates ranged from 98% to 100%. W hen looking at participants percentage of correct responses from one proficien cy level to the next one (see Table 4 6), there was a general pattern of greater accuracy for most conditions (high regular, low irregular, high stem change, and low stem change verb forms), as expected. Interestingly, all groups of participants obtained s lightly better results at identifying a low regular verb root than a high regular verb root, while the opposite was true for irregular and vowel stem change verb roots: they obtained slightly better results at identifying a high frequency irregular and ste m change verb form than a low frequency one. However, after
168 running one way ANOVAs and independent samples t tests, no significant differences were found in the accuracy rates on the different conditions, neither between the groups of late learners on the one hand nor between the two groups of advanced speakers on the other hand. This indicates that all groups of learners were highly successful at identifying a verbs stem from its inflected form. There were no improvements across proficiency groups when co mparing the three groups of late learners, which shows that Spanish L2 learners are proficient at identifying a verb root from a given conjugated verb form, even if it is a suppletive verb form, such as the group of irregular verb forms in the present stud y. 4.2 .2.2. RT results The results for the RTs for the different verb types across proficiency groups are now presented. Table 4 7 shows a clear di ssociation between regular verb forms on the one hand, and irregular and stem change verb forms on the other hand. Table 4 7. Independent samples t test results for the RTs for the correct responses by all groups of learners in Experiment 2: Comparison between test items of high and low frequency. Mean RT (SD) Verb Type Frequency Beginning (N = 15) Intermediat e (N = 15) Advanced (N = 15) Bilingual (N = 15) Regular High 921.18 (229.05) 869.32 (251.44) 943.66 (262.33) 830.02 (201.40) Low 918.68 (201.33) 957.73 (289.48) 1003.47 (277.58) 874.27 (202.14) Irregular High 1154.38* (330.73) 978.95 (305.71) 1010.88** (330.74) 945.32 (252.50) Low 1497.75* (496.87) 1136.55 (356.40) 1496.36** (567.61) 1097.23 (381.34) Stem Change High 1043.16* (350.58) 907.13** (252.87) 934.73** (313.17) 876.27 (246.99) Low 1424.68* (431.32) 1207.51** (331.07) 1382.40** (489.57) 110 3.86 (453.46) *p < .05, ** p < .01, two tailed, equal varianc es assumed The beginning learners identified the root of a regular verb form from its conjugated preterite form in less than a second, irrespective of the frequency of the conjugated verb form.
169 There were not significant differences between high (M = 921.18, SD = 229.05) and low frequency regular verb forms (M = 918.68, SD = 201.33). However, the frequency of the test items played a statistically significant role for irregular and stem change ver bs, as predicted by the dual mechanism models. The beginning learners took significantly longer ( t = 2.228, df = 28, p = .034) to identify the roots of low frequency irregular (M = 1497.75, SD = 496.87) verb forms than those of high frequency irregular ve rb forms (M = 1154.38, SD = 330.73). This group of learners also took significantly longer ( t = 2.658, df = 28, p = .013) to identify correctly the roots of low frequency stem change verb forms (M = 1424.68, SD = 431.32) than those of high frequency (M = 1043.16, SD = 350.58). Similar results were found for regular and stem change verb forms by the intermediate group of late learners, but not for the irregular verb forms, as displayed in Table 4 7. Intermediate learners always took longer to respond to l ow frequency test items, irrespective of the verb type. However, these differences were statistically significant ( t = 2.793, df = 28, p = .009) only for stem change verbs: M = 907.13, SD = 252.87 for high frequency items in contrast to the M = 1207.51, S D = 331.07 for low frequency ones. At the intermediate level, no dissociation was found between regular and irregular verb forms, i.e., these participants responded at a similar speed to regular and irregular test items, irrespective of their frequency. Th ese results for the irregular verb forms were not predicted, especially if we consider the results from the group of advanced learners. The group of advanced learners behaved in a similar way to the beginning group. As shown in Table 4 7, there was a clear dissociation between regular verbs on the one hand and irregular and stem change verbs on the other hand. They reacted to regular verb forms at a very similar speed, whether they were of high (M = 943.66, SD = 262.33) or of low frequency (M =
170 1003.47, SD = 277.58). However, the advanced learners reacted significantly faster ( t = 2.862, df = 28, p = .008) for irregular verb forms of high frequency (M = 1010.88, SD = 330.74) than for those of low frequency (M = 1496.36, SD = 567.61), and significantly faste r (t = 2.983, df = 28, p = .006) for stem change verb forms of high frequency (M = 934.73, SD = 313.17) than those of low frequency (M = 1382.40, SD = 489.57) The results for the irregular and stem change forms found for the advanced learners are more ro bust ( p < .01) than those for the beginning learners ( p < .05), even though both were statistically significant. An interesting result i s that the RTs to irregular and stem change verb forms decreased from the beginning to the intermediate group but increa sed again from the intermediate to the advanced group. For regular verb forms, the same pattern was found for high frequency items but for low frequency regular forms participants always took longer from one proficiency level to the next one A series of o ne way ANOVAs were run to compare the RTs to the different verb types by frequency between the three groups of late learners but results indicated that these patterns never reached significance ( p > .05) There were only near significant results (N = 45, F = 2.801, p = .072) for the low frequency irregular verbs. The bilingual group behaved differently compared to the late learners. This group of speakers reacted at a similar speed for all the test items of the different verb types, whether they were of h igh or low frequency. As shown in Table 4 7, the different RTs for the test items per verb type never reached statistical significance, although participants always took longer to react to low frequency items than to high frequency ones.
171 0 200 400 600 800 1000 1200 1400 1600 1800 High Reg Low Reg High Irreg Low Irreg High SCh Low SCh Reaction Times (milliseconds) Advanced Bilingual *p < .05, n.s. = non significant, two tailed, equal variances assumed Figure 49. Independent samples t test results for RTs on the different verb types for Experiment 2: Comparison between the advanced and bilingual groups (N = 30) Finally, the RTs to the differen t verb types of high and low frequency between the advanced late learners a nd the bilingual speakers were compared to check if the differences were statisti cally significant. A series of independent samples t tests were run and results indicated that advanced learners took significantly longer ( t = 2.261, df = 28, p = .032) to react to low frequency irregular verb forms (M = 1496.36, SD = 567.61) than the group of bilingual speakers (M = 1097.23, SD = 381.34) as shown in Figure 4 9. These significant resul ts were not found with any other verb type. In sum, results from Experiment 2 indicated that all groups of participants were very accurate at deciding whether a conjugated verb form presented orally in the L2 had the same or a different stem from an immedi ately preceding infinitive verb form presented on a computer screen. Late learners reacted differently to regular verbs on the one hand and to irregular and stem change verbs on the other hand, this becomes apparent when we consider their RTs for items of high or low frequency. There were no differences for regular items; however, there were n.s. n.s. n.s. n.s. n.s.
172 significant differences in their RTs for high and low frequency irregular and stem change verb forms, except for irregular items in the intermediate group. There were also frequency effects in the accuracy rates on stem change verbs by all groups of late learners as well as on irregular verbs by the intermediate group. In the group of bilinguals, there were no significant differences in RTs for any type of verb forms. Fi nally, bilingual speakers reacted significantly faster than advanced learners to low frequ ency irregular verb forms. 4. 3. R esults from the Production Task In this section, the results for the production task14In this elicited narrative, participants had a picture story sequence of 24 drawings about a male character and they were instructed to narrate in Spanish what that person did in the past (see Appendix F). Each drawing was accompanied by a Spanish verb in its infinitive form and participants were asked to use those verbs in the preterite tense. They were inst r ucted to use all test items and that the researcher would have to guess the characters name at the end of their are described. A main goal of the production ta sk was to obtain participants accuracy levels on the different verb types under investigation by using an oral production task in which verb forms are produced in sentence contexts, in a similar way to habitual Spanish activities used during class time. A nother objective was to observe participants reaction to specific verb forms of high and low frequency, to check if they treated verb forms differently when they did not know the target verb form. Finally, participants errors in specific verb forms for w hich they did not know the correct answer can be v ery informative concerning the mental processes at work when handling complex verb forms in the L2. 14 Results for Production Task 2 are not included in this chapter for the reasons described in section 184.108.40.206.
173 narration. There were an equivalent number of high and low frequency verbs from each of the thre e verb types Again, participants oral data we re collected with Audacity and the recordings were saved as .aiff files. There were a total of 1,440 tokens (24 tokens per participant 60 participants). All tokens were coded on an SPSS data file. The information included for each token was the following : (a) participant code number, (b) participants proficiency level, (c) token number, (d) participant exact oral response, (e) test item, (f) verb type for each test item, (g) accuracy of the response, and (h) participants behavior in relation to the test item. The accuracy of the response variable was coded according to three options: (a) correct, (b) incorrect, or (c) does not apply, used when the participant skipped the test item. The variable with the greatest range of options was the participants be havior in relation to the test item, for which the same codes as in Experiment 1 were us ed (see Table 4 4 and Appendix J for more information). 4.3.1. Accuracy R esults In this section, the accuracy results are reported. Any given verb form was considered correct if it was produced following the grammar rules of Spanish or by using the corresponding memorized item and/or making the appropriate vowel change, even if it was not the first choice of the participant. Table 4 8 displays the accuracy on the differ ent verb types per proficiency group. P articipants obtained greater accuracy results for regular verbs than for the rest of the verb types. In general, they also got better results for irregular verbs than for stem change verbs. For example, the highest accuracy rate on the regular verbs by any group of participants was 95% (for high frequency regular verbs by bilinguals), while the greatest accuracy rate for irregular verbs was 85% (for high frequency irregular verbs by bilinguals) and for stem change verbs it was 83.33% (for high frequency stem change verbs by the advanced group).
174 Table 4 8. Accuracy rates on the different verb types of high and low frequency for the Production Task for all groups of participants (1,440 tokens). Percentage (correct respons es/test items per group of participants and verb type) Regular Irregular Stem Change High Low High Low High Low Late learners Beginning (N = 15) 83.33% (50/60) 68.33% (41/60) 40% (24/60) 25% (15/60) 43.33% (26/60) 30% (18/60) Intermediate (N = 15) 85% (51/60) 78.33% (47/60) 65% (39/60) 50% (30/60) 55% (33/60) 36.67% (22/60) Advanced (N = 15) 90% (54/60) 81.67% (49/60) 71.67% (43/60) 45% (27/60) 83.33% (50/60) 40% (24/60) Bilingual (N = 15) 95% (57/60) 90% (54/60) 85% (51/60) 68 .33% (41/60) 81.67% (49/60) 66.67% (40/60) An interesting result for the present study is the contrast between the regular verbs on the one hand and the irregular and the stem change verbs on the other hand; i.e., participants from all proficiency gro ups consistently obtained the highest accuracy results on the regular verbs. This distinction becomes more evident for the low frequency test items on all verb types, e.g., the beginning group got 68.33% (41/60) of the low frequency regular verb forms corr ect, which, conversely, was in sharp contrast to their accuracy on the low frequency irregular (25%, 15/60) and low frequency stem change verb forms (30%, 18/60). This pattern is also visible for the other groups of participants. When looking at the result s of late L2 learners only, as expected, their accuracy scores on all verb types became progressively higher across the proficiency groups. For example, the beginning learners got 30% of the low frequency stem change verbs correct, while the intermediate l earners obtained an accuracy rate of 36.67% and the advanced learners 40%. The only exception were the low frequency irregular verbs, for which the intermediate group obtained slightly better accuracy results (50% or 30/60 verb forms produced correctly) th an the advanced group (45% or 27/60 verb forms produced correctly).
175 In order to check if there were frequency effects in the accuracy rates to the different verb types in this Production Task, a series of independent samples t tests were run taking frequen cy as the grouping variable. These tests were run for each proficiency group individually. Frequency effects were found for some verb types among the advanced and the bilingual groups only (see Table 4 9). By contrast, there were no frequency effects in th e accuracy rates f o r any verb type in the beginning or intermediate groups, as shown in T able 4 9. Table 4 9. Independent samples t test results for accuracy rates by all groups of learners in the Production Task: Comparison between test items of high and low frequency. Mean accuracy rate (SD) Verb Type Frequency Beginning (N = 15) Intermediate (N = 15) Advanced (N = 15) Bilingual (N = 15) Regular High 83.33 (27.82) 85.00 (22.76) 90.00 (12.68) 95.00 (14.02) Low 68.33 (38.34) 78.33 (35.19) 81.67 (22.09) 90.00 (18.42) Irregular High 40.00 (35.10) 65.00 (22.76) 71.67* (26.50) 85.00(*) (22.76) Low 25.00 (28.35) 50.00 (29.89) 45.00* (31.62) 68.33(*) (22.09) Stem Change High 43.33 (33.36) 55.00 (35.61) 83.33** (24.40) 81.67 (19.97) Low 30.00 (27.06) 36. 67 (28.14) 40.00** (33.81) 66.67 (33.63) *p < .05, **p < .01 two tailed, equal varianc es assumed In the group of advanced learners, there were frequency effects in the accuracy rates of irregular and stem change verb forms (see Table 4 9). Advanced lear ners obtained significantly higher accuracy results ( t = 2.503, df = 28, p = .018) for high frequency irregular (M = 71.67, SD = 26.50) than for low frequency irregular (M = 45, SD = 31.62) verb forms as well as significantly higher accuracy results ( t = 4 .026, df = 28, p = .000) for high frequency stem change (M = 83.33, SD = 24.40) than for low frequency stem change (M = 40, SD = 33.81) verb forms. These frequency effects were not present for the group of regular verbs.
176 In the group of bilingual speakers there were frequency effects which were marginally significant ( t = 2.035, df = 28, p = .051) in the accuracy results of irregular verbs only, as shown in Table 4 9. They produced all high frequency verb forms with greater accuracy than low frequency ver b forms, irrespective of the verb types, but results only near ed significance for irregular verbs. 71.67 65 40 0 20 40 60 80 100 120 140 Beginning Intermediate Advanced Accuracy Rates (0-100) *p < .05, (*) p < .1 n.s. = non significant Fig ure 4 10. One way ANOVA with Tukey post hoc test on the accuracy rates for high frequency irregular verbs comparing the beginning, intermediate and advanced learners for Production Task In order to track developmental changes on the different verb types by the late L2 learners, oneway ANOVAs were run with the post hoc Tukey test. There wer e significant results only for high frequency irregular (N = 45, F = 5.114, p = .010) and high frequency stem change verbs (N = 45, F = 6.398, p = .004), as disp layed in Figures 4 10 and 4 11. No significant differences were found for the regular verb forms, whic h indicates that late learners start with a ( ) n.s.
177 very high accur acy level for those verbs and that this level remains constant across proficiency levels 83.33 55 43.33 0 20 40 60 80 100 120 140 Beginning Intermediate Advanced Accuracy Rates (0-100) *p < .05, **p < .01, n.s. = non significant Figure 411. One way ANOVA with Tukey post hoc tes t on the accuracy rates for high frequency stem change verbs comparing the beginning, intermediate and advanced learners for Production Task These results indicate that across proficiency levels late learners improve more on those items which are sensitiv e to input frequency. High frequency irregular verb forms benefit mostly at the beginning stages of acqu isition (as shown in Figure 4 10), from the beginning to the advanced (MD = 31.667, SEM = 10.440, p = .011) proficiency levels but also from the beginn ing to the intermediate ones (MD = 25.000, SEM = 10.440, p = .054), even though the latter difference only nears significance, as indicated by the Tukey post hoc results mentioned. In contrast, high frequency stem change verb forms improved at later stage s of acqu isition (as shown in Figure 4 11), from the beginning to the advanced (MD = 40.000, SEM = 11.501, p = n.s. **
178 .003) as well as from the intermediate to the advanced (MD = 28.333, SEM = 11.501, p = .046) proficiency levels. I f we compare the results of t he advanced lear ners and those of the bilingual speakers their accuracy rates were very similar for most of the verb types, except for the low frequency irregular and stem change verbs. Statistically speaking, s imilar results were found in terms of a clea r dissociation between regular verb forms on the one hand and the irregular and stem change verb form s on the other hand. I ndependent samples t tests were run and the only significant results foun d are displayed in Figure 4 12. 0 20 40 60 80 100 120 140 High Reg Low Reg High Irreg Low Irreg High SCh Low SCh Accuracy Rates (0-100) Advanced Bilingual *p < .05, n.s. = no n significant, two tailed, equal variances assumed Figure 412. Independent samples t test results for accuracy on the different verb types for Production Task: Comparison between the advanced and bilingual groups (N = 30) N o significant differences were found for regular verbs. There were significant differences between these two groups in their accuracy for irregular ( t = 2.343, df = 28, p = .026) and stem change ( t = 2.166, df = 28, p = .039) verb forms, but this time the differences were found only f or the low frequency forms. Bilingual speakers obtained signific antly better accuracy results for the low frequency irregular (M = 68.33, SD = 22.09) and the low frequency n.s. n.s. n.s. n.s.
179 stem change (M = 66.67, SD = 33.63) verb forms than the advanced learners (M = 45, S D = 31.62 for irregular and M = 40, SD = 33.81 for stem change verb forms). If morphological information is handled and encoded by a single system or pattern associator with no intervening rules, then participants accuracy rates on any verb form should c orrelate with the frequency of the Spanish verb form in question, irrespective of its morphological decomposition and internal structure (section 1.1.1.). These results should be more robust for beginning learners because the initial stages of acquisition are more lexical. In order to test these claims statistically, a Pearson bivariate correlation was run between the frequency of these 24 test items as measured by the averaged frequency of the three corpora, and the participants accuracy on these items. T hese correlations were run taking the accuracy rates from the different proficiency groups separately and irrespective of verb types, i.e., all test items at once without further taxonomies. Results indicated that the frequency of the specific verb forms i n Spanish never correlated with the accuracy levels of any group: beginning group (r = .058, p = .787, n = 24), intermediate group (r = .308, p = .143, n = 24), advanced group (r = .214, p = .315, n = 24), and bilingual group (r = .096, p = .657, n = 24). These results are good evidence against the unitary models of morphological repr esentation. In sum, accuracy results show that all groups of learners obtained high levels of accuracy for regular verb forms and no gains were attained as the proficiency lev el in the target language increased Late learners with greater proficiency levels obtained better results for irregular and stem change verb forms of high frequency while the main production differences between the advanced and the bilingual groups in thi s task was found in the irregular and stem change verb forms of low frequency. No correlations were present between participants accuracy and the overall frequency of the verb f orms under investigation.
180 4.3. 2. Error Types and Participants Behaviors The second half of the data analysis for the Production Task provides a descriptive report on the most prevalent errors that the different groups of participants exhibited in their oral narratives. First, the most prevalent behavior and/or error for each verb type are described and then the internal structure of the error and/or the relevance of the observed behavior are explained. Table 4 10 summarizes the most prevalent behavior/error by these groups of participants, arranged by verb type. In this section, an asterisk represents an incorrect verb form given the instructions of the task (provide the third person singular preterite) instead of given the morphological structure of the verb form. Table 4 10. Summar y of participants main errors for all groups of learners (N = 60) and by verb type in Production Task. Rate of the main error (number of errors with the main error type/total number of errors per group of participants and verb type) Regular Irregular Stem Change High Low High Low High Low Begin ning (N = 15) Error type 3, 6, 10 & 17 3 7 & 10 1 2 2 Error rate 20% (2/10) 26.32% (5/19) 22.22% (8/36) 48.89% (22/45) 41.18% (14/34) 52.38% (22/42) Intermediate (N = 15) Error type 10 13 10 1 2 2 Error rate 44.44% (4/9) 46.15% (6/13) 33.33% (7/21) 60% ( 18/30) 66.67% (18/27) 68.42% (26/38) Advanced (N = 15) Error type 3 3 10 & 16 1 2 2 Error rate 50% (3/6) 54.55% (6/11) 29.41% (5/17) 51.52% (17/33) 80% (8/10) 75% (27/36) Bilingual (N = 15) Error type 3 3 10 1 2 2 Error rate 100% (3/3) 100% (6/ 6) 55.56% (5/9) 36.84% (7/19) 63.64% (7/11) 60% (12/20) 220.127.116.11. Errors associated with regular verb forms It is important to remember that regular verbs were always produced with greater accuracy than irregular verbs and stem change verbs. The lowest accuracy for regular verb forms was observed for low frequency regular verbs by beginning learners, with an accuracy rate of 68.33%. That rate is the same or greater than any accuracy rate for low frequency irregular verbs
181 and stem change verbs by any group of participants, including the bilingual speakers who had 68.33% accuracy for low frequency irregular verbs and 66.67% for low frequency stem change verb forms. As a consequence, there were few examples per observed behavior/error type. Table 4 10 displa ys the main behavior/error for high and low frequency regular verbs by each group of participants. With the exception of the intermediate learners, the main cognitive strategy by the participants was to produce a nonfinite form, mostly the infinitive, afte r another verb in the preterite: pudiera regalar (s/he could give away ) instead of the preterite form Beginning learners showed a variety of errors and behaviors for regular verb forms and there is a lack of consistency in their use. First, the most p revalent errors with high frequency regular verbs are described. For the verb llamar (to call) one participant produced *se llamar (behavior 3), another said llam ( I called ) and another used the present tense llama (s/he calls). Two participants p roduced the nonexistent verb form meri for the verb mirar ( to look ). One participant skipped the verb nacer (to be born), although most of them produced it morphologically correctly, even if they placed the stress on the wrong vowel (behavior 8) as i n *naco (s/he was born), by two participants. With the verb tomar (to take), the observed errors were: tomar (behavior 3), *tomi (error 16), and toma (error 10). Similar results were found in the group of low frequency regular verbs. There were two nonexistent forms for the verb esperar (to wait): espre and *espari The verb inventar (to invent) was produced incorrectly by three participants with a TAM suffix from another verb class, i.e., inventi (s/he invented). And regalar (to give aw ay) was either used as a noun or in a nonfinite form: da un regale or para regalar for example. Finally, the main behaviors with the verb escapar (to escape) were avoidance (behavior 7), a different tense as in se escapa (present), and a TAM suffix from another verb class escapi
182 In the intermediate group there was a greater variety of errors when participants did not provide a correct response. In the group of high frequency regular verbs, llamar (to call) was used in the progressive once llam ando (calling). The verb mirar (to watch) was used once in the present tense mira (s/he watches) and once with an incorrect PN suffix in the preterite, i.e., *mir (I watched). Although nacer (to be born) was used by one student in a passive con struction, the main behavior with this verb was to skip it; two participants did not provide a response for it. And for tomar (to take), two participants used it in the preterite with the incorrect PN suffix: tom ( I took ) In the group of low frequen cy verbs, escapar (to escape) was used by two participants with a wrong root, changing the root vowel from a to e : *es que p ( s /he escaped). A similar error was observed for the verb inventar (to invent), with four participants producing it with the i nitial vowel pronounced as an e not i as in e nvent (s/he invented). Esperar (to wait) was used once in the progressive and once in the imperfect tense as the most prevalent errors. And for regalar (to give away) there were two present tense forms (* regalo and regala: I give away and s/he gives away, respectively), as well as two nonfinite/noun forms (error 3): dio la regala (s/he gave away) and para regalar (in order to give away). In the advanced group there was an increase in be havior 3, or the use of nonfinite verbs for both high and low frequency regular verbs. In the group of high frequency regular verbs, nacer (to be born) was always produced accurately. For the verb llamar (to call) there was an example of the test item used as a noun *hizo una llamada (s/he made a phone call) and one example of wrong root *llo m The verb mirar ( to watch ) was skipped once and also used by another participant in a nonfinite form : quiso mirar ( s/he wanted to watch ). The verb tomar ( to take ) was used once in the plural tomaron ( they took ) and once in a nonfinite form: tuve
183 que tomar ( I had to take ). In the group of low frequency regular verbs, both escap ar ( to escape ) and esperar ( to wait ) were produced in a nonfinite form when used incorrectly as in *ten a que escapar (s/he had to escape), two examples of estaba esperando (s/he was waiting), and tena que esperar (s/he had to wait). The only error with the verb reg alar (to give away) was to provide a wrong root as a result of syllable confusion or metathesis: re lag instead of regal. And for the verb inventar (to invent) there were three errors : a nonfinite form in *decidi de inventar (s/he decided to invent), a wrong root in *e nvent (s/he invented), an d the use of a TAM suffix from another verb class in inventi Bilinguals always produced the regular verbs nacer (to be born), tomar (to take) and escapar (to escape) correctly. The main behavior with the remaining high and low frequency verbs was to use a nonfinite form as in para llamar (in order to call) or estaba esperando (s/he was waiting). 18.104.22.168. Errors associated with irregular verb forms The behaviors and errors observed for the irregular verb forms are different from those for the regular verb forms. Table 4 10 clearly shows that the main error with low frequency irregular verbs was overregularization. This indicates that when L2 learners did not have a memory trace for a given past form related to the infinitive form, they simply a ttach ed a regular ending to the stem of the verb, hoping that the final product would be part of the Spanish language. For example, they said cabi ( it fit ) instead of cupo, or *reduci (s/he reduced) instead of redujo. As for high frequency irregular verbs, the most prevalent behaviors/errors were to avoid providing the corresponding irregular verb form (behavior 7) or to substitute a verb form in the preterite with a different PN suffix (error 16) or in another tense (error 10). Although there was so me room for variability in the errors made by these groups of participants when handling high frequency irregular verb forms, a form in a different tense was the main
184 error and the significance of this error is discussed at the end of this section. First, however, the most prevalent behaviors/errors for the specific verb forms elicited by each proficiency group are described. Beginning learners overregularized the following low frequency irregular verbs: andar (to walk), extraer (to extract), and reduc i r (to reduce). Their incorrect responses show that they have two different rules at work, whether the verb in its infinitive belongs to the first class or the second/third class: they said and (s/he walked); but in the case of extraer and reduc i r th ey said extrai (s/he extracted) and *reduci (s/he reduced) The verb caber (to fit) was the least frequent of all four low frequency irregular verbs and most beginning participants skipped this test item. The main observed behaviors when dealing w ith high frequency irregular verbs were skipping of the test item and providing a form from a different tense. However, beginning learners treated all four high frequency irregular verbs differently. For the most part, the verb decir (to say) was conjuga ted in the present tense, two digo (I say) forms and one *dice (s/he says) form. The verb poder (to be able to) was usually skipped. This test item in particular was difficult to grasp for some participants since it depicted a man on top of a scale with a text bubble which said 83 > 75 and learners should have said something like he was able to lose weight ( pudo bajar de peso). The main error with the verb poner (to put) was to provide a nonexistent verb form or a form that belongs to a differe nt verb, in this case the verb poder (to be able to) because of obvious similarity. The attested forms were: pongi, *pongos(las) and pudo, while the expected preterite verb form was puso. For the last high frequency irregular verb, venir (to come), beginning participants used the correct irregular stem but attached to it the incorrect TAM suffix i : vini ( s /he came) This error is interesting in
185 that it is the opposite case to the observed behavior for most incorrect stem change verbs, where le arners used the correct ending but forgot to change the vowel in the stem. Intermediate learners also overregularized the low frequency irregular verbs, using the two rules described by the beginning group, i.e., and i depending on the verb class: *and (s/he walked) instead of anduvo or reduci (s/he reduced) instead of redujo. For the verb caber (to fit), in half of their errors intermediate participants produced the nonexistent verb form cubo, which is very similar to the expected correct form cupo. As with the beginning learners, the intermediate group employed a variety of behaviors and made a range of errors specific to the high frequency irregular verb forms. Of the two errors with the verb decir (to say), one was a different PN suff ix: dije (I said), and the other was a different tense: dice (s/he says). Most errors with poder (to be able to) were to produce preterite forms of the verb poner (to put): se puso (s/he put) and *puse (I put). This confusion of poder / poner verb forms was not present with the errors for poner (to put), because the main error with poner was to conjugate it with a different PN suffix, such as puse (I put). Two imperfect and one present venir (to come) forms were the most prevalent error type with this verb: *vena (2) and *viene There were also three participants who produced this verb in the preterite with a TAM suffix from another verb class, i.e., the irregular rule was applied but with a wrong ending: vini ( s/he came ). These exam ples are simila r to overregularizations but using the irregular stem [ vin ], which was kept. All advanced learners treated the four low frequency irregular verbs similarly, with the most prevalent error type for each of them being overregularization: and (s/he walked), *cabi/ *cab ( s /he fit), extrai (s/he extracted), and reduci ( s /he reduced). The most prevalent behavior/error for the high frequency irregular verbs was specific to the test item in
186 question. For the verb decir (to say), the main error was to provide the imperfect form deca (s/he used to say). The verb poder (to be able to) was skipped twice and another two advanced participants provided the present tense verb forms puede (s/he can) and puedo (I can) instead of the corresponding preterite verb form. As for the verb poner (to put), there was a range of answers when it was not conjugated correctly, with one example for each: puse (I put) whether as the first answer or the final answer after another failed respons e, pona ( s /he used to put) and *pudo (s/he was able to), examples of behaviors/errors 6, 14, 10, and 17, respectively. Finally, four advanced learners produced the incorrect verb form vini instead of vino ( s /he came). Among the bilinguals, there was a variety of errors for high and low frequency irregular verb forms. However, the most prevalent error for low frequency irregular verbs was overregularization and for high frequency irregular verbs it was to provide a verb form in a different tense. T he verbs andar (to walk) and extraer (to extract) were overregularized: *and and *extrai respectively. For c aber (to fit), there were two cases of nonfinite forms: *pudo caber (s/he could fit) as a way to avoid providing the elicited preterite f orm cupo (s/he fit). The remaining low frequency irregular verb reducir (to reduce) was either overregularized after pr oviding the correct verb form (hypercorrection) as in redujo > reduci (s/he reduced), or it was used in a different tense as in redujiera (wrong form for I would reduce) or reduzco (I reduce). The most prevalent behaviors/errors for the high frequency irregular verbs were the following: the verb poner (to put) was produced correctly in the preterite ( puso) by all participa nts. Two bilingual participants produced the verb form *decidi ( s/he decided ) as the preterite verb form for the verb decir ( to say ) which clearly belongs to a different verb (error 17). One verb form in the imperfect poda (s/he could) and a form from a different verb
187 *subi de peso ( s /he gained weight) were the only two errors with the verb poder (to be able to). Finally, the main error with the verb venir (to come) was to produce it in a different tense, as in vena (s/he used to come) or viniera ( s/he would come). Since the first stages of acquisition of a foreign language are usually considered lexical in nature, according to the predictions of the Cognitive Approach to language learning by Skehan (1998), the frequency of the differ ent verb forms of every verb in this task may be playing a role in the developing system of these groups of learners. Since irregular verb forms are the ones to be stored and handled by the declarative memory system, the verb forms with the greatest freque ncy may be the ones accessed the fastest, irrespective of the tense they belong to. As shown in Table 4 10, most learners who erred in the preterite of high frequency irregular verbs did so by using a verb form from another tense. It is relevant for the pr esent study to check if the verb forms supplied incorrectly were more frequent in the Spanish language than the preterite verb form being elicited, i.e., than the third person preterite form. This will indicate if frequency played an interfering role for h igh frequency irregular verb forms. For each high frequency irregular verb, the most frequent verb forms for that lemma were checked in the Corpus del Espaol This corpus proved to be the most resourceful one of the three used in this study as explained in section 3. 4.2., and it was the only one employed for this report. If we do not take into account plural verb forms (in the picturestory sequence there is only one character and the instructions asked participants to report only what he did), nonfinite verb forms (e.g., infinitive, progressive, past participle), and the remainder of the singular preterite forms (better accounted for by behavior/error 6), these participants used singular verb forms in a tense which is among the most frequent one in Spani sh for that verb in question. For example, there were three examples of the verb form dice (s/he says) instead of the elicited
188 dijo (s/he said), and the raw token count from the CDE showed that dice was the following most frequent singular verb form a fter dijo for that verb in Spanish, followi ng the criteria just described. As shown in Table 4 11, the only exceptions to this pattern were the third person singular conditional tense forms podra and vendra and the third person singular future form *vendr, which were never used in the production of these learners. Table 411. Raw token counts for the most frequent verb forms of the high frequency irregular verbs used in Task 1, ranked according to the data from the Corpus del Espaol15 d ecir ( to say ) (CDE) : Verb forms raw token count and usage by the participants poder ( to be able to ) poner ( to put ) venir ( to come ) form # ? form # ? form # ? form # ? dijo 14308 E puede 18591 puesto 2550 N viene 2757 decir 12599 N poder 8346 N puso 2254 E vino 1229 E dice 7794 pueden 8114 N poner 2136 N venir 1195 N dicho 4159 N poda 5287 pone 1427 venido 1120 N digo 4029 podra 4586 pona 677 vena 1090 deca 3730 puedo 3319 ponen 664 N vienen 1043 N pudo 2477 E puesta 623 N veng a 827 pusieron 597 N venan 467 N ponerse 545 N vinieron 368 N puse 476 N vengo 272 poniendo 472 N vine 267 N puestos 351 N vengan 237 N pongo 343 vendra 175 vendr 166 venimos 154 N viniera 140 E = elicit ed verb form, N = not attested in data and/or not applicable, = used, = not used Arguably, we can conclude that behavior 10, or the use of a different tense, by all groups of participants as the most prevalent beha vior for high frequency irregular verb forms is sensitive to the frequency of the different verb forms used incorrectly. To put it differently, when these groups of L2 learners produced an incorrect verb form from a different tense, they always used the ve rb forms with the highest frequencies in the Spanish language for the test items being elicited, as shown in Table 4 11 with a 15 These token counts were obtained during the month of February 2009.
189 22.214.171.124. Errors associated with stem change verb forms When looking at the results for stem change verb forms only, it becomes evident that the most prevalent error by all groups of learners was to omit the internal vowel change while adding the appropriate regular ending to the verb forms. For example, they produced corregi ( s/he corrected ). The regular TAM suffix i was at tached correctly to the stem of the verb but most participants failed to change the vowel of the root from e to i (all eight stem change verb forms in the picture story sequence required this vowel change). When tracking developmental changes across profic iency levels it also becomes clear that the greater the proficiency level of late learners, the greater the rate of no vowel change with this verb type, irrespective of the frequency of the items. The s e differences between the proficiency groups indicate that even though late L2 learners improve in their accuracy for stem change verbs, the enduring error, when it occurs, is to forget to make the corresponding vowel change. For most bilingual speakers as well, this error is still the main observed behavior for high and low frequency stem change verb forms when produced incorrectly. When looking at the results for the different test items individually, a significant percentage of beginning learners neglected to make the vowel change as the most prevalent erro r for all verb items in the stem change groups except for the verb sentir (to feel). Repeti ( s /he repeated) or *verti (s/he poured) appeared very frequently in their production when they did not produce those verb forms correctly. For sentir (to feel), most beginning learners placed the stress on the high vowel of the final diphthong, in addition to not making the internal vowel change, e.g., sento (s/he felt) Similar results were observed with the intermediate group. The most frequent erro r type for all stem change test items was behavior 2, i.e., not changing the vowel but adding the correct TAM suffix. The only verb that was treated differently was repetir (to repeat) in that half of the
190 errors were behavior 10, or the attachment of a s uffix from a different tense, this time the present tense: repite and repito The main error for the stem change verb forms which were produced incorrectly by the advanced group was again the absence of the vowel change. There was one exception only: th e verb seguir (to follow), which all advanced learners produced correctly. All bilingual speakers always produced the elicited preterite form for the stem change verb conseguir (to get) correctly. Again, the main error for those items produced incorre ctly was the absence of the vowel change. For the verb repetir (to repeat), there were a couple of other observed behaviors besides no vowel change: to use an infinitive form as in volvi a repetirla ( s/he repeated it again ) and to use a different ten se as in que lo repitiera (that s/he repeat it). However, these instances of the verb repetir (to repeat) while coded as incorrect according to the instructions of the task, are correct in Spanish because they were inserted in a more descriptive narrative in the past with a wider variety of past tenses. Bilinguals used a more developed narrative and they did not pay such close attention to the instructions of the task, which limited their description to the preterite tense. In sum, if we consider the m ost prevalent behaviors/errors for the different verb types by all groups of participants, the following patterns were found: (a) all groups forgot to change the vowel of the stem change verbs from e to i in the preterite; (b) they all overregularized the low frequency irregular verbs by applying two different rules: to andar (to walk), and i to caber (to fit), extraer (to extract), and reducir (to reduce); (c) most groups of participants provided verb forms in other tenses for high frequency i rregular verbs; and (d) the use of a noun from the verb stem or of a nonfinite form was usually provided for high and low frequency regular verbs by most groups of participants except for the in termediate group.
191 4.4. Methodological Triangulation of Result s In this last section the main dissociations or frequency effects found in the different tasks between regular verbs on the one hand, and irregular and stem change verbs on the other hand are outlined. T hose developmental changes which were significant across the three proficiency groups of late learners are also mentioned. Finally, the differences which reached statistical significance when comparing the advanced groups of speakers with an early (English/Spanish bilinguals) vs. late (advanced late learn ers) age of initial exposure to Spanish are discussed I n this section those results which were only descriptive are not included i.e., the differences in the error types and observed behaviors when the participants did not provide a correct response. In the design of this study, the frequency of the test items in the Spanish language was a key variable in determining if L2 learners may be representing and/or accessing fully inflected verbal forms via two different mental systems. If so, frequency should play a role for irregular and perhaps, stem change verb forms but it should never play a role for regular verb forms. For the most part these predictions were born out in the RT results for the different ver b types, as shown in Figure 4 13. In Experiment 1, there were frequency effects for irregular verbs by the beginning group but in the reversed direction, i.e. it took them longer to produce correct high frequency than to low f requency irregular verbs (see Table 4 1). There were also frequency effects for stem change verbs by the bilingual group only, this time in the expected direction (see Table 41). And in Experiment 2, frequency effects in the RTs to stem change verb forms played a role for all groups of learners, except the bilinguals (see Table 47). In this experiment, there were also frequency effects in the RTs to irregular verb forms, but only for the beginning and the advanced learners (see Table 47). The frequency effects were more robust ( p < .01) for the intermediate and advanced groups t han for the beginning group ( p < .05), indicated by italics in Figure 4 13.
192 Beginning learners Exp.1 & 2 NO Regular Intermediate learners Exp.1 & 2 NO Verbs Advanced learners Exp.1 & 2 NO Bilingual learners Exp.1 & 2 NO Beginning le arners Exp.1 YES (Table 4 1) Exp.2 YES (Table 4 7) Intermediate learners Exp.1 NO Irregular Exp.2 NO Frequency Verbs Advanced learners Exp.1 NO Effects Exp .2 YES (Table 47) in RTs Bilingual learners Exp.1 NO Exp .2 NO Beginning learners Exp.1 NO Exp. 2 YES (Table 4 7) Stem Intermediate learners Exp.1 NO Change Exp.2 YES (Table 47) Verbs Advanced learners Exp.1 NO Exp.2 YES (Table 47) Bilingual learners Exp.1 YES (Table 4 1) Exp.2 NO = results in the expected direction, = results in the opposite direction Figure 413. Outline of statistically significant frequency effects in RTs for all verb types in Experiments 1 and 2 When looking at possibl e frequency effects in the accuracy rates to the different verb types used in the three tasks, there were interesting results, outlined in Figure 414. All groups of learners obtained significantly better accuracy results for high frequency irregular than for low frequency irregular verb forms in Experiment 1 (Table 43). In this experiment, there were frequency effects for regular verbs in the intermediate group only, which is the only piece of evidence obtained that does not support regular irregular diss ociations (see Table 43). In Experiment 2 the frequency effects were observed for the stem change verb forms among all groups of late learners only, although there were also frequency effects for irregular verbs by the intermediate group (Table 46). And in the Production Task, the only frequency effects found
193 were for irregular verbs by the advanced and bilingual groups as well as for stem change verb forms by the advanced learners only (Table 49). Beginning learners Exp. 12 & Task NO Ex p.1 (Table 43) YES Regular Intermediate learners Exp.2 NO Verbs Task NO Advanced learners Exp.1 2 & Task NO Bilingual learners Exp.1 2 & Task NO Exp.1 (Table 43) YES Beginning learners Exp.2 NO Task NO Ex p.1 (Table 43) YES Intermediate learners Exp.2 (Table 46) YES Irregular Task NO Frequency Verbs Exp.1 (Table 43) YES Effects Advanced learners Exp.2 NO in Accuracy Task (Table 4 9) YES Rates Exp.1 (Table 43) YES Bilingual learners Exp.2 NO Task (Table 4 9) (YES) Exp.1 NO Beginning learners Exp.2 (Table 46) YES Task NO Stem Exp.1 NO Change Intermediate learners Exp.2 (Table 46) YES Verbs Task NO Exp .1 NO Advanced learners Exp.2 (Table 46) YES Task (Table 49) YES Bilingual learners Exp.1 2 & Task NO Figure 414. Outline of statistically significant frequency effects in accuracy rates for all verb types in Experiments 1 and 2 as well as in the Production Task In order to tra ck developmental changes across proficiency levels in this cross sectional study a series of one way ANOVAs were run on the accuracy rate and RT results between the three groups of late learners. There were s ignificant results for Experiment 1 and the Production Task only, as shown in Figure 415. In general, intermediate and advanced learners were faster than the beginning learners at providing a correct preterite form for regular verbs of high and low
194 freque ncy (see Figures 42 and 43). These improvements in RTs were not obtained for irregular or stem change verbs. Regulars High freq. YES B I, B A (Figure 4 2) Low freq. YES (B I), (B A) (Figure 43) RT gains Irregulars High freq. NO i n Exp.1 Low freq. NO Stem High freq. NO change Low freq. NO RT gains Regulars High & low freq. NO in Exp.2 Irregulars High & low freq. NO Stem change High & low freq. NO Regulars High freq. YES (B I), (B A) (Figure 46) Ac curacy Low freq. NO Changes gains in Irregulars High freq. YES B A (Figure 47) Over Exp.1 Low freq. N O Time Stem High freq. YES B A (Figure 44) change Low freq. YES B A (Figure 4 5) Accuracy Regulars High & low freq. NO gains in Irregulars High & low freq. NO Exp.2 Stem change High & low freq. NO Regulars High freq. NO Accuracy Low freq. NO gains in Irregulars High freq. YES (B I) B A (Figure 410) Task Low freq. NO Stem High freq. YES B A I A (Figure 411) change Low freq. NO B = beginning, I = intermediate, A = advanced Figure 415. Outline of statistically significant developmental changes in RTs and accuracy scores between the three groups of late learners in Experiments 1 and 2 as well as in the Production Task When tracking developmental changes in the accuracy of their responses to the test items, in both Experiment 1 and the Production Task the advanced learners obtained significantly better results than beginning lea rners when producing correct high frequency irregular forms (see Figures 48 and 411 for Experiment 1 and Production Task, respectively) and correct high frequency stem change forms (see Figures 45 and 412, in the same order ). In addition, in the Produc tion Task
195 advanced learners obtained significantly better accuracy results than the intermediate learners for high frequency stem change verbs (see Figure 411). The advanced group also obtained significantly better accuracy results than the beginning grou p in Experiment 1 for low frequency stem change verbs (see Figure 45) and for high frequency regular verbs (see Figure 46). Finally, the intermediate learners also obtained better accuracy results than the beginning group in Experiment 1 for high frequency regular verbs (see Figure 46). The last set of significant results was found between the advanced grou p of late learners and the English/Spanish bilinguals (see Figure 416). Regulars High freq. Exp.1 2 NO Low freq. Exp.1 2 NO RT Irregulars High freq. Exp.1 2 NO Differences Low freq. Exp.1 NO Exp.2 YES (Figure 49) Stem High freq. Exp.1 2 NO change Low freq. Exp.1 2 NO Age of Regulars High freq. Exp.1 2 & Task NO Initial Low freq. Exp.1 2 & Tas k NO Exposure High freq. Exp.1 2 & Task NO Irregulars Exp.1 (YES) (Figure 48) Accuracy Low freq. Exp.2 NO Differences Task YES (Figure 412) High freq. Exp.1 2 & Task NO Stem Exp.1 YES (Figure 48) change Low fr eq. Exp.2 NO Task YES (Figure 412) Figure 416. Outline of statistically significant differences in RTs and accuracy scores between the group of advanced late learners and the bilingual group in Experiments 1 and 2 as well as in the Production Task There was also consistency across the different tasks used in this study because in Experiment 1 as well as in the Production Task, bilingual speakers obtained significantly better results in the accuracy to low frequency irregular verb forms and in t he accuracy to low frequency stem change verb forms (see Figures 49 and 413, respectively). In this case, it did not matter if these
196 verb forms were elicited in isolation (Experiment 1) or if they were elicited in sentence contexts (Task). The s e results also provide a new set of evidence of dissociations between regular verbs and the other verb types in this study because the statistical tests performed never showed significantly better results for bilinguals than for the advanced learners in their accuracy to regular verbs In Experiment 2, the only significant result was that advanced learners took significantly longer to react to low frequency irregular verb forms than the group of bilingual speakers (Figure 49). In sum, statistical results from the di fferent tasks in this study provided evidence of dissociations in the frequency effects to the different verb types. However, some of these results were in the reversed direction and/or were not observed consistently in all groups of learners. The main imp rovements between the three groups of late learners were faster RTs for regular verbs when they had to produce them in isolation and significant improvements in accuracy rates to high frequency irregular and stem change verb forms, whether in isolation or in sentence contexts. There were also significantly better accuracy results for high frequency regular verbs and low frequency irregular verbs when participants had to provide preterite forms in isolation only. The bilingual speakers had a better knowledge and responded faster than advanced late learners to low frequency irregular verb forms. They also obtained better accuracy results when asked to produce low frequency stem change verb forms.
197 CHAPTER 5 DISCUSSION In this chapter, the results summarized in Chapter 4 are evaluated and discussed based on the research questions and the main goals of my study. The discussion is organized in three mai n sections. The first section (5 .1.) answers the first and second research questions, which examined the regular irregular15.1. Regular Irregular Dissociations and Frequency Effects dissociations that may be found in the accuracy scores, RTs, and/or error types for complex verb forms, due to their regularity and the frequency of their occurrence in the target language. Special attention is paid to the presence and signifi cance of frequency effects in the results reported in the previou s chapter. The second section (5 .2.) addresses the main goal of the study : It provides answers to the third research question by describing the main developmental patterns in the acquisition of inflection morphology by late L2 learners. There is an emphasis on the mental representation and processing among the various groups of L2 learners included in the study and an attempt is made to explain how the L2 proficiency variable can account for d ifferences in the accuracy scores, RTs, and/or error types in the verb types included in my i nvestigation. Finally section 5 .3. describes the main differences between the advanced group of late L2 learners and the bilingual group in how they handle comple x verb forms. This section addresses the fourth research question and helps in understanding what the main effects of greater amount of practice in the L2 and the age of initial exposure to the L2 are once a high proficiency level has been reached in the t arget language The DP model (Ullman, 2001b) proposes that late L2 learners will depend to a greater extent upon declarative memory (the memory system which handles arbitrary associations) in the 1 When there is a mention in th is chapter to regular -irreg ular dissociations, the latter term includes the stem change verbs since stem change verbs are usually considered a subgroup of irregular verbs, based on the vowel change that is absent in the regular paradigm.
198 L2, as a function of age of initial exposure to the target language. But practice with the L2, besides an earlier initial age of exposure, should affect both grammatical proficiency and the degree of dependence on procedural memory for grammatical computations (e.g., the sequencing of morphemes for regular v erbs). As mentioned in section 1.2.5 ., the later the initial age of exposure to the L2 and the less the practice with it, the weaker the dissociations should be (perhaps even to the point of absence) between the declarative and pro cedural memory systems because the procedural memory system is not used for regular verb forms yet Forms which are normally compositionally computed by the procedural memory system in the L1 may simply be stored as memor ized forms in the L2 lexicon, i.e., regular verb forms may be stored in the lexicon and accessed from the declarative memory system rather than being computed by the mental grammar. Some rules may be consciously/explicitly learned (especially in a formal i nstructional context, as was the case for the participants in the current study) and applied in the declarative memory, although they need not be. These learned explicit rules may be used by L2 learners to construct linguistic forms. Based on the se theore tical clai ms, the DP model suggests that in SLA both regular verbs and irregular verbs may be learned in and computed over associative memory, at least in the initial stages of acquisition, and both may show associative memory effects, e.g., frequency effe cts. Consequently, L2 speakers should show less separability between regular and irregular verb forms than L1 speakers since only for L1 processing do regular and irregular verb forms distinctively rely on either grammar or lexicon, respectively. The resul ts of the present study are evaluated in light of the se predictions in the following sections. The first res earch question looked at the types of regular irregular dissociations and frequency effects among the English/Spanish bilinguals for accuracy scores RTs, and erro r types
199 for complex verb forms. T he second research question investigated the same dissociations and frequency effects but among late L2 learners of varying proficiency levels. 5.1.1. Frequency Effects in the Accuracy Scores On the one hand if the DP model is accurate, there should be frequency effects by all the groups of participants especially in the accuracy scores for irregular verbs, as they are supposed to be computed in the declarative memory. For the bilinguals, there should only be frequency effects for irregular verbs, whereas for the late L2 learners, there may be frequency effects for all verb forms. On the other hand, the higher the proficiency of the late L2 learners, the more probable it should be to find regular irregular d issociations in the frequency effects for the accuracy scores, if they are present at all The group of bilinguals produced high frequency irregular verb forms with significantly greater accuracy than those of low fre quency for Experiment 1 an d for the Pro duction Task. These frequency effects were not observed for the regular or the stem change verbs. These results support the claims of the DP model in that only irregular verbs showed associative memory effects namely frequency effects. They also indicate that the accuracy scores for stem change verbs did not differ based on the frequency of the test it ems. This finding is in sharp contrast with the frequency effects found in the RTs to the stem change verbs in Experiment 1, which are discussed in section 5 .1.2. When regular verb forms are produced correctly by the bilingual participants it is because they are computing them online using the procedural memory. The mental operation of sequencing should apply to all regular verbs irrespective of their freque ncy and, not surprisingly, no frequency eff ects were found for these verbs. An account of the absence of frequency effects for the stem change verbs is described in section 5 .1.2
200 Finally, in Experiment 2 there were no frequency effe cts among the group of bilinguals f o r any v erb type This can be explained by the ceiling effect that the task design provoked, i. e., most participants achieved 100% accuracy for all verb types due to the small number of options (only two) and task diffic ulty (not a production task). This is discussed in greater detail in section 5.2 .1. In sum, for this group of participants frequency effects were only observed in the accuracy scores to irregular verbs, as predicted by the DP model, because those forms are the ones retrieved fr om the declarative memory and are the only ones show ing associative memory effects. Turning to the various groups of late L2 learners in Experiment 1, all groups of late L2 learners showed frequency effects in the accuracy scores to irregular verbs. Thes e participants produced with greater accuracy the irregular verbs they encountered more often in the L2 input, so that the arbitrary mental associations between the verb roots and the irregular preterite forms were strengthened and the high frequency irreg ular forms were retrieved with greater accuracy. Additionally, intermediate learners produced regular verb forms with greater accuracy if they were of high frequency than those of low frequency. The DP model suggested that for late L2 learners there should be less separability between regular irregular verbs because both could be initially computed in the declarative memory system and both should show frequency effects. The finding that only the intermediate group showed such frequency effects for the regul ar verbs may be explained in part b y the task demands. L2 learners were asked to produce preterite forms in an oral elicitation task where verb forms we re elicited individually and without a sentence context. The beginning learners were those who have been introduced to the preterite rules most recently in the classroom for the first time and they may have been the ones using those rules explicitly from the declarative memory. That is why all regular verbs of high and low frequency
201 were produced with about the same accuracy scores (84.67% for high and 83.33% for low regular verbs). But the learners in the intermediate group were taking grammar and composition courses in which the introduction of the preterite rules was not part of the curriculum and this te nse was not going to be explicitly tested in upcoming exams. The production by this group of learners showed that frequency played a role for regular verbs because they may have been retrieving individual verbs from the de clarative memory ( before a restruc turing process for L2 learners with higher proficiency). The use of the declarative memory leads the L2 learners to greater productivity, which most intermediate Spanish courses emhasize: greater production in the L2 ( For a discussion of the shift or res tructuring from the declarative to the procedural memory systems for the regular verbs in the groups of late L2 learners, see section 5 .2. where the main developmental patterns in the acquisition of verbal morphology by late L2 learners are discussed. ) Th e accuracy scores by the advanced learners, similar to those by the bilinguals described before, indicate that regular verb production is very accurate for both high and low frequency items (97.33% and 96.67% respectively ), and this may be the result of t he default regular rules being computed online in the procedural memory system, because they have already been internalized and are now being used implicitly. In Experiment 2, all groups of late L2 learners were more accurate at identifying stem change ro ots of high frequency verbs than thos e of low frequency When these participants did not identify the vowel change in those verbs, they may have thought that it was due to the task design. They had to decide whether the nonfinite and the preterite forms be longed to the same verb or not, and that led them to press the different stem button for low frequency stem change verbs incorrectly if they interpreted the change in the vowel of those pair s of words as a trick from the task. The lowest accuracy scores were for the low frequency stem change verbs,
202 indicating that late L2 learners were exposed to the vowel changes of those verbs less often than to those of high frequency stem change verbs, hindering their recognition by these learners. As for the irregula r verbs, frequency effects were only found among the group of intermediate learners, the only group which also showed frequency effects for regular verb forms in Experiment 1. Both results may point to the interpretation that this group of learners is retr ieving and accessing most verb forms from the declarative memory, since memorization helps them to produce more forms in the L2 with accuracy, and, as a consequence, there were more frequency effects found in the intermediate group than in the other groups of late L2 learners. N one of the groups displayed frequency effects for regular verb forms, probably because of the ease of the task and because the root and the preterite forms for regular verbs are more similar in their morphological structure. Finally in the Production Task, there were frequency effects only among the group of advanced learners for both irregular and stem change verbs. Late L2 learners paid less attention to the verb forms when they had to insert them in a sentence context as eviden ced by the fact that the accuracy scores in this task were usually lower than in Experiment 1, the other production task. The fact that only the group of advanced learners showed frequency effects for the irregular and stem change verbs points towards grea ter regular irregular dissociations as the proficiency of the late L2 learners increased, due to the use of two different mental systems in the computation of verb forms at this proficiency level, similar to the native speakers of Spanish and to the bilingual group (which also showed regular irregular dissociations in the frequency ef fects ). 5.1.2. Frequency Effects in the RTs When verb forms are stored in the declarative memory, frequency effects should be observed based on the frequency of the test item, because the retrieval of arbitrary root preterite
203 form associations of low frequency items should take longer than those of items encountered more often. Irregular, and maybe stem change verb forms too, may show these associative memory effects. The time needed to attach an inflectional suffix to a regular root should not differ based on the frequency of the test item because the concatenation of pieces of language can be performed quickly via online mechanisms Therefore, frequency effects are not expect ed in the RTs for regular verbs. The frequency effects for all groups of participants in the two psycholinguistic experiments are discussed in the next paragraphs first for the bilingual speakers and then for the late L2 learners. In the recognition task (Experiment 2), the bilingual participants did not show any frequency effect for any verb type. This finding contrasts with the frequency effects observed in all groups of late L2 learners, which are discussed below. The best explanation for this lies in t he role of automaticity (DeKeyser, 1997) I n this experiment participants were asked to press one of two buttons as quickly as possible once they had decided whether or not each pair of oral and visual stimuli belonged to the same verb. Since the bilingua l participants had been exposed to L2 Spanish from a very early age, they seemed to be cap able of recognizing L2 words at a very fast speed, which may explain why no frequency effects were observed for the irregular and/or the stem change verbs. Since the bilingual participants have somehow automatized the processing of L2 word recognition, their RTs to any verb type did no t vary much. In fact, this group of participants showed the fastest RTs to any verb type when comparing the data from all the groups in the present study. In the production task (Experiment 1), the bilingual participants showed frequency effects in the RTs to the stem change verbs only. They were faster at mak ing a correct decision for those stem change verbs which were of high frequency in Spanish than for those which were of low
204 frequency. Coupled with the absence of frequency effects in the accuracy scores to stem change verbs in the same experiment, the following explanation is suggested: The bilingual participants, as well as the rest of the groups in the present study, produced the low frequency stem change verb forms with the lowest accuracy scores for the most part. This indicates that they had problems when retrieving them from memory, based on the vowel change in the root, as indicated by the most prevalent error type for these verb forms (see section 5.1.3.) It seems that the bilinguals knew that these verbs need a vowel change in the preterite and they accessed that information from the declarative memory, where different verb s tems are stored for every stem change verb. For example, the verb root of dormir to sleep has three allomorphs based on the tense and subject in which it appear s : dorm duerm durm Bilingual participants must retrieve the correct stem from the declar ative memory and, as a consequence, they need more time to retrieve those for low frequency items. However, once they accessed the correct root, the computation of the correct PN suffix was fast and accurate, as indicated by the absence of frequency effect s in the accuracy scores As for the absence of frequency effects for the irregular verbs, the bilingual participants may have internalized all the irregular verbs used in the study and when they knew the correct response for them, they retrieved them fro m the declarative memory quickly. This may be the opposite case to the stem change verb forms in that the difficulty for the bilingual participants was not to retrieve the irregular verbs with ease but to retrieve them with accuracy as indicated by the st atistically significant accuracy differences for the irregular items In fact, they had the lowest accuracy scores for low frequency irregular verbs in Experiment 1 (67.33%) while in the Production Task their accuracy scores for high frequency stem change verbs were v ery similar (68.33%) to those for low frequency stem change verbs (66.67%).
205 The strongest evidence in support of the DP model was found for late L2 learners in the second experiment, i.e., when participants were asked to identify complex verb f orms in Spanish. All groups of late L2 learners showed frequency effects for irregular as well as for stem chang e verb forms in this experiment except for the irregular verbs by the intermediate learners These results indicate that when late L2 learners l istened to a Spanish preterite form and then were presented with an infinitive root of the same or a different verb stem, they searched the mental lexicon in their developing grammars to look for the possible arbitrary connection between those two words be fore making a decision. When the irregular and stem change preterite forms were of low frequency, it took them longer to search for that possible connection in their mental lexicon than if they were of high frequency. These frequency effects were not found for the regular verbs and the explanation can be found not only in the predictions of the DP model for the processing of complex verb forms in an L2 but also in the design o f this experiment The frequency effects for stem change verbs in this experiment may indicate that late L2 learners perceived the vowel change in the pair of words from the stimuli and quickly searched their mental lexicon before making a decision in order to make sure that the vowel in the preterite form was actually a possibility in the target language. To put it differently, these learners stopped for a longer moment when faced with the low frequency stem change test items because they noticed a disagreement between the infinitive and the preterite vowels and they quickly checked in their declarative memory that both verb stems belonged to the same verb, following the instructions of the experiment. The results from the first experiment, on the other hand do not support the predictions of the DP model. In this experiment there wer e no frequency effects for any verb type in the groups of late L2 learners The only exception was the RTs to irregular verbs by the beginning learners.
206 However, these effects were in the oppos ite direction of that predicted; i.e., the beginning learners t ook significantly longer to produce irregular verbs of high frequency than those of low frequency. These results cannot be accounted for by singlemechanism models (e.g., Bybee, 1995; Rumelhart & McClelland, 1986) either because frequency never played a r ole for most of the test items irrespective of the verb type. However, t he frequency effects in the RTs to some verb types were evidenced in the results from the second experiment for the irregular and the stem change verbs for most groups of participants and as mentioned before. 5.1.3. Regular Irregular Dissociations in the Error Types The last type of data coded from the participants responses was the errors they made for the verb types being elicited. The pattern of results observed in the error type s i n Experiment 1 and the P roduction T ask also help in providing a different way to capture the participants mental representations for the production of complex verb forms. I rrespective of the participants proficiency level or even the task, the same m ain error types were made for the low frequency irregular verb forms as well as for the groups of stem change verb forms. E rror type 1, or the attachment of a regular suffix to a verb which required an irregular root was the most repeated error type for l ow frequency irregular verbs. Error type 2 or the lack of the corresponding vowel change was made by all groups of participants for stem change verb forms of either hi gh or low frequency in Spanish. The number of errors for regular verb forms and even f or high frequency irregular verb forms was very low (less than 10 errors for the same error type) in all groups of participants. Th e se latter results are only informative rather than displaying trends in the production of L2 learners. Due to the similarit ies in the errors produced by the groups of participants in the present study, the error types an d error rates by all groups of participants are assessed at the same time in this section.
207 Even if participants obtained high accuracy scores in Experiment 1 and the Production Task for Spanish morphology in general, they always obtained lower accuracy scores for the low frequency irregular and the low frequency stem change verbs. It is posited here that t hese verb types are being retrieved from the declarative memory (the whole preterite form for irregular verbs vs. the irregular verb root for stem change verbs) and their frequency in the target language affected how well the participants retrieved them. On the other hand, regular verb forms can be computed onl ine and no such frequency effects were found for them in general, as discussed in the previous sections. The most prevalent error type for stem change verbs of high and low frequency was to omit the corresponding vowel change, in both tasks. These verb fo rms were produced as if they were considered regular verb forms because the root was kept unchanged while the corresponding regular su ffix, necessary for stem change verbs, was added. Among the group of lat e L2 learners there was a constant increase in the error rate for the absence of the vowel change as the proficiency level increased This is an indication that the greater the proficiency of the learner in Spanish as an L2, the more evident it becomes that the lack of vowel stem changes are persistent er rors for stem change verbs in the production of late L2 learners. Under singlemechanism models, this finding may be a sign that a single associative memory mechanism can account for this error type. However, upon closer examination the lack of the necess ary vowel change may not provide evidence in favor of such associative memory mechanisms. The reasoning is as follows. In the present study there are two regular rules considered, namely: Rule 1: Add to ar verbs when [3rd singular prete rite] must be e xpressed overtly Rule 2: Add i to er / ir verbs when [3rd singular prete rite] must be expressed overtly
208 Rule 1 appears very frequently in the input because 88% of all Spanish verbs follow this rule, according to Aguirre & Dressler (2006). Consequently if the participants were not sure how to construct the preterite form for the stem change verbs, they should have erred by attaching rule 1 to the unmodified verb root by analogy. How ever, the regular suffix attached to these stem change verbs was not or rule 1, i.e., the ending from the most frequent verbs in Spanish, but i or rule 2, following the conditions of this second mental rule, even though it ha d been encountered less frequently in the L2 input. Under dual mechanism models, including the DP model, this error type can be accounted for as the incorrect retrieval of the corresponding irregular verb root (which is kept intact instead because it could not be retrieved from the declarative memory or because it has not been noticed yet) and the a ttachment of the default regular rule in accordance to the verb class. This selection of the appropriate regular rule in accordance to the verb class could not be observed in previous work using languages with poorer inflectional systems such as English and, as just discussed, this error type is bett er explained under the DP model: mental rule attached in accordance to verb class and retrieval of an irregular root from the declarative memory system. T he other most prevalent error type in the group of high frequency stem change verbs was the combination of no vowel change plus wrong stress assignment, e.g., adverto s/he warned. Verb forms in the imperfect tense (the other simple past tense in Spanish) do not require changing the vowel of the root. The misplacement of stress in stem change preterite forms may be due in part to confusion with the imperfect tense forms, which require the adding of the suffix a in the third person singular for er / ir verbs, as in peda s/he used to order. If that is the case for these errors, then two options may explain this error type: (a) participants believed the preterite is similar to the imperfect tense but with a different vowel in the TAM
209 suffix, i.e., the o (wrong stress in the preterite marker ) suffix was bel ieved to be similar to the a (correct imperfect tense suffix) one, or (b) participants did not know how to conjugate these verbs in the preterite and they tried to produce the imperfect tense instead because they knew that, even though they would be wrong, at least the imperfect has a past reference. The TAM morpheme they attached to the root was actually a blend between the semantically incorrect imperfect tense suffix and the preterite tense suffix being elicited. The most prevalent error type for low f requency irregular verbs in both tasks was overregularization, i.e., the attachment of a regular suffix to a form which is expected to take an irregular root. Especially in Experiment 1, there was usually a constant increase in the error rate of overregula rizations as the proficiency level of the groups of participants increased again indicating that this error type lingered in the production of L2 learners. If frequency alone is the only predictor of the observed behaviors in L2 learners, how can single m echanism models account for this error? First of all, these form s are not grammatical in Spanish and the late L2 learners had not been exposed to these verb forms in the input Second, the fact that this error rate is the most prevalent behavior in all gro ups of participants when they did not produce low frequency irregular verbs correctly means that these are not sporadic errors in the production of some L2 learners but strong and consistent patterns of errors by most participants ; e.g., the incorrect form *produc i s/he produced appeared consistently in the producti on of all groups of participants in the present study. Finally, if the frequency of the suffix attached incorrectly at t he end of the verb is what caused this error, then why is it that in mo st of the cases the regular rule attached (1 or 2) agreed with the verb class ( ar vs. er / ir verbs)? The DP model offers a better explanation for this error type in low frequency irregular verbs than the single mechanism models These verbs are computed from the declarative
210 memory system, which is sensitive to input frequency. The tasks asked the participants to produce a preterite form and when the elicited verb form was of low frequency (and sometimes even if they were of high frequency) the participant s occasionally failed to retrieve the corresponding irregular form from the declarative memory Due to the pressures of the tasks to provide the preterit e form as quickly as possible without skipping items the participants computed the irregular verb via the procedural memory system, attaching the default regular rule ( 1 or 2 ) based on the verb class of each test item. That is why even the beginning learners incorrectly attached the suffix i (instead of the most frequent one in the Spanish input) to er / ir verbs such as caber to fit, traducir to translate, extraer to extract, or conducir to drive. However, when dealing with the ar verb andar to walk, they produced *and s/he walked. In both cases the L2 participants closely followed the conditions of the (over)regular ization rules 1 and 2 stated before. Additional evidence in favor of this mental process can be found in error 13 or the use of a wrong root. In those cases the participants chose an incorrect root but attached the appropri ate suffix, again, based on the verb class e.g., some intermediate participants produced: *vind i s/he sold (rule 2 correct but with a wrong root), *esquep s/he escaped (rule 1 correct but with a wrong root). Even if it may be argued that those exa mples could be the result of syllable confusion, why was that confusion present in the production of some L2 learners when in Experiment 1 they had the nonfinite verbs displayed onscreen for as much time as it took them to produce their oral response? The regular rule attached to the verb root of the irregular verbs is better accounted for by dual mechanism models and in languages with a rich morphological system such as Spanish. This is one of the advances made by considering the stem change verb forms as a separate group of verbs fro m the irregular ones It helped to understand the attachment of the default regular rule in the L2 learners IL as a
211 compensatory strategy for low frequency irregular items when the declarative memory failed to retrieve the eli cited form. O verregularization s were also found in the responses to high frequency irregular verbs but only in Experiment 1, and with the exception of the beginning learners. The beginning learners produced most of the incorrect high frequency irregular ve rbs in a different tense instead The same occurred in their production of regular verbs as well This provides additional evidence that at this proficiency level the procedural memory is not at work because the regular rules have not been extracted and a re not being used implicitly yet. Instead, associative memory mechanisms led these learners to the retrieval of the mo st frequent verb forms they had encountered up to that moment to compensate for the lack of the preterite forms in the declarative memory system. In the case of the Production Task, error type 10 or the use of a diff erent tense, mainly the present indicative was found in the production o f all groups of participants In this case this error type can be attributed to the task design as it h as been found in previous studies that when L2 learners are asked to recollect what other people have done in the past with the help of some type of visual stimuli, and especially for those L2 learners at high proficiency levels, they tend to use the prese nt tense. For example, the native and the near native participants in Salaberrys (1999) study on Spanish past tense verbal morphology narrated the events from the clips of a silent movie using the present tense instead of the past tense. In the present st udy, and as mentioned in section 4 3.2.2., when these groups of participant s produced an incorrect verb form from a different tense, they always used the verb forms with the highest frequencies in the Spanish language for the te st items being elicited whi ch indicates that for irregular verb forms the frequency of the items conditioned not only the production of L2 learners but also their most prevalent error types.
212 Finally, the number of cases per most prevalent error type for the regular verbs was limite d and the results can only be considered informative in nature. There were interesting results though, especially for the main claims of the Cognitive Approach to language learni ng (Skehan, 1998). T he use of verb forms in another tense usually appeared as the most prevalent error for these verbs in the beginning and the intermediate groups of late L2 l earners These results can be explained by the higher frequencies of those forms retrieved from the declarative memory system, based on the claim that the ini tial stages of SLA are usually lexical, according to the Cognitive Approach. In the Production Task, the use of a nonfinite form or even a noun appeared often in the production of most groups of participants as the most prevalent error type. These cases usually indicated that the elicited verb form was used in the infinitive inside a fixed expression or a chunk of language with a previous auxiliary verb in the preterite tense, e.g., para + infinitive in order to ( used to link two sentences into a singl e one ) dio un regalo s/he gave a present and hizo una llamada s/he made a phone call ( the use of a noun to skip the computation of a complex verb form), ten a que + infinitive s/he had to estaba + present participle s/he was and s o on Thes e chunks of language are readily available in the declarative memory system of the L2 participants and they could retrieve them very quickly to aid their narration of the events depicted in the picture story sequence. The Cognitive Approach suggests that m ost language production relies on them and a s mentioned in the previous chapter, these examples were considered errors based on t he instructions of the tasks even though it would be more accurate to refer to them as an avoidance strategy. It may be even possible that the participants did not realize that they did not use the verb provided in the elicited tense In any case, these examples support the notion from the Cognitive Approach that most language
213 production relies on chunks of language readily avail able in the declarative memory that can be retrieved and computed online quickly to increase fluency during language production 5.2. Developmental Patterns in the Acquisition of Verbal Morphology This section addresses the third research question of the present study : What are the developmental patterns among the groups of late L2 learners when handling complex verb forms in L2 Spanish? The main significant results in the accuracy scor es (section 5.2.1.), the RTs (section 5 .2.2.) and the error types ( 5.2. 3.) are addressed separately and evaluated to the extent that they can provide cumulative evidence either in favor of the dual me chanism models in general and t he DP model in particular or in favor of the single mechanism models 5.2.1. Improvements in the Accuracy Scores In successful SLA the expected developmental pattern is to start with an incomplete representation of the L2 showing less than perfect accuracy scores followed by subsequent stages of increased accuracy rates as a result of higher profici ency levels. This developmental route may not necessarily be present in all L2 learners and/or for all grammatical features in the same way. In fact, there are usual ly instances of backsliding which, interestingly, may indicate an increase in the complexit y of the learners developing grammar or IL The learning context for the participants in my study may have condition ed the developmental patterns found in the accuracy scores for specific verb forms as they have been exposed explicitly to some Spanish gra mmar rules However, the expected pattern is that high frequency items will be acquired before the low frequency ones because they have also been encountered more frequently in the L2 class room Additionally, verb forms handled by the declarative memory sy stem should be acquired first because, as mentioned by the DP model, the learning speed of this memory system is fast leading to increased/enhanced productivity. In the case of L2 learners, the DP model suggests that all verb forms are initially handled b y the declarative memory system and, as a
214 consequence, all verb types may show the same pattern of acquisition in relation to the accuracy scores. The results from Experiment 2 are evaluated before those from the other two production tasks (Experiment 1 an d the Production Task) analyzed in my study. In Experiment 2 two factors favored a ceiling effect in the average amount of correct responses per group of participants: only two options were available and the task itself was not difficult since participant s did not have to produce complex verb forms. Most of the scores from all groups of participants were at or near the maximum possible for the experiment (100% accuracy) and the utility of the data collection measure used was therefore compromised because o f the lack of variability. As predicted due to the ceiling effect, the oneway ANOVA results indicated that there were no significant differences among these groups. In Experiment 1 and the Production Task, the main statistically significant differences w ere found in the accuracy scores on the high frequency verb forms, mostly between the beginning and the advanced groups and mostly for the irreg ular and stem change verb forms However it is important to remember that the accuracy scor es from Experiment 1 and the Production Task show a clear regular irregular dissociation in that all groups of participants obtained high accuracy scores for the regular verb forms compared to those for the irregular and stem chan ge verb forms Late L2 learners always obtaine d better accuracy scores for the regular verbs in Experiment 1 than in the Production Task. On the other hand, the accuracy scores for the irregular and stem change verb forms were lower than for the regular verbs, and sometimes with very low accuracy, suc h as only producing a quarter of them accurately. If the initial stages of SLA are usually lexical (Housen, 2002) and if learners initially handle all verb forms via the declarative memory system which is sensitive to factors such as input frequency (Ullm an, 2001b), then why is it that regular verbs are produced with better accuracy scores than the other
215 verb types, especially when they are supposed to be handled with a memory system di stinct fr o m the one that native speakers of the language employ (the pr ocedural memory system)? There are several factors which can explain the results for regular ver bs. Forty five participants were learners of Spanish in a US college and their exposure to Spanish was mostly in the classroom. The textbooks always present the paradigms of regular verbs before covering the irregularities and learners tend to memorize these paradigms for the exams (it is not that rare to find them on the back of the exam of some students). The L2 learners can explicitly access the regular rules in their declarative memory system to produce a given verb form, whereas the retrieval of some irregular verb forms from the declarative memory system can easily fail due to their limited number of exposure s Even if the regular rules are not yet used impl icitly at the early stages of SLA, the L2 learners can produce new verb forms by analogy from other verbs. And since the group of regular verbs is larger than the different groups of irregular verbs the L2 learners are more successful at retrieving an app ropriate verb ending by analogy than whole word irregular verbs Additionally, the fact that regular verb forms do not have changes in the verb root also favors a high accuracy score; i.e., if L2 learners keep intact the internal segment structure of the v erb while attaching a regular suffix to a verb root that they are not completely sure about how to conjugate in the preterite, they will always be correct for the group of regular verbs, but not for the groups of irregular verbs The microclass of ir stem change verbs in Spanish with the vowel alternation pattern e/i (e.g., pedir pidi to ask for, elegir eligi to choose) contains more than 50 verbs, according to Aguirre & Dressler (2006), while 88% of all Spanish verbs belong to the default micro class of ar regular verbs (e.g., cantar cant to sing, amar am to love).
216 This observation leads to two implications for the findings in the present study. First, even though some high frequency irregular verbs appear more often in the target la nguage than most regular verbs, the frequency with which L2 learners are exposed to regular ver b forms is higher, a nd 80% of the regular verb forms used in my study were ar verbs (see Appendix K ). The second implication is that if L2 learners handle all v erb forms initially via the declarative memory which can lead to some degree of productivity by analogy, regular verb forms are also favored in comparison to the irregular ones because the latter contain just some members from which to extract a verb patt ern and they also have many different microclasses, whereas the former can only belong to three default macroclasses ( ar er ir ) with many members per macroclass. Finally, Experiment 1 and the Production Task asked the participants to provide a respons e for each stimulus. If L2 learners were not sure what the expected response was, the task still asked them to produce something, and so if their declarative memory failed to retriev e a preterite form, they could resort to an explicit default rule and attach a regular suffix to the verb root. This is how they could compensate for the lack of knowledge in the declarative memory and this can also explain why their accuracy scores for regular verbs were the highest, because L2 learners can produce correct regu lar verb forms even when they are not sure w hat the elicited verb form i s. The advanced learners obtained significantly higher accuracy scores for high frequency irregular and stem change verb forms in both production tasks than the group of beginning lear n ers The Tukey post hoc results also indicated that the accuracy differences for high frequency stem change verbs in the Production Task between the intermediate and the advanced groups reached statistical significance. These results show that those verb forms which are handled by the declarative memory system are improved in the production of late L2 learners
217 When the test items were elicited in a sentence context (the Production Task), there were clear regular irregular diss ociations in the improvement s in the accuracy scores as indicated by the absence of accuracy changes for the regular verbs But when the test items were elicited in isolation (Experiment 1), there were significant improvements in the accuracy scores to high frequency regular verbs a s well, even though the Tukey post hoc test did not reach the significance level between the beginning and the intermediate or the advanced groups. The results for the accuracy improvements among the late L2 learners just mentioned lend partial support fo r the DP model. This model predicts that L2 frequency plays a role for those verb forms handled by the declarative memory system and the accuracy scores from these participants showed that high frequency irregular items (including the stem change verbs) ar e acquired earlier than the low frequency ones because they have been encountered more frequently and they are sensitive to input frequency. Additionally, if the task asked the learners to concentrate on the verb forms and the verb forms only (Experiment 1 ), not in producing full sentences in order to describ e a picture story in which the test items are somehow less attended to ; results showed that low frequency stem change verb forms also benefited as well as high frequency regular verbs These results may be explained by the demands of Experiment 1. If advanced learners pay close attention to the base form of low frequency stem change verb forms, they will be able to retrieve the vowel stem change pattern with greater accuracy than if they are asked to pro duce those verbs in sentence contexts. This can also indicate that beginning L2 learners have not yet perceived the vowel change in some preterite for ms and may still think that those verbs are regular because they attach the corresponding regular suffix. Only the third person subjects require a vowel change in the preterite tense, and if L2 learners do not pay close attention to those forms in the input, they may think the verb is fully regular in the preterite. In
218 the case of high frequency regular verb f orms in Experiment 1, on the other hand, intermediate and advanced learners reached almost perfect accuracy scores for those when the task was limited to the verb forms, so that the application of the default regular rule was facilitated via analogy of oth er regular verbs encountered frequently. Even though the DP model implied that L2 learners are initially handling all verb forms via the declarative memory system (whether for a whole word verb form such as cant s/he sang or via the explicit kn owledge o f a regular rule like add ), the resul ts in th e Production Task indicated that the frequency of the Spanish verb forms used in the present study, irrespective of whether it was high or low never correlated with the accuracy scores of any group. This i s a good piece of evidence against singlemechanism models and in favor of dual mechanism models because it indicates that the accuracy of recognition or production of the verb forms was not only the result of the frequency with which L2 le arners may have encountered those forms in the input. To p ut it differently, something else was driving the acquisition of inflectional morphology besides the frequency of whole words store d in the declarative memory. I propose that that something else is the application of mental rules for verb roots, whether explicitly via analogy from the extensive regular verb forms in the input, or procedurally via online mechanisms for more advanced L2 learners and the native speakers of the language. The regular irregular dissociat ions in the accuracy scores in the Production Task, the higher accuracy scores for regular verb forms only in the first stages of acquisition, and the frequency effects that were generally attested in the accuracy improvements to irreg ular and stem change verb forms are all better accounted for by the dual me chanism models and the DP model than by single mechanism models of SLA. However, there are data which may be even more informative concerning the mental mechanisms late L2 learners employ when handling complex
219 verb forms: the time they need to retrieve complex verb forms based on their frequency and/or their regularity. These results are discussed in the next section. 5.2.2. Improvements in the RTs A significant decrease in the RTs to a specific grou p of verb forms from one group of late L2 learners to the next one, e.g., from the beginning to the intermediate group, can be deemed as an index for automatization. In his study of automatization in the acquisition of L2 morphosyntax rules, DeKeyser (1997 ) used Andersons (1983) Adaptive Control of Thought theory when defining automatization : () knowledge typically starts out as explicit (declarative) information, knowledge that, which is turned into specialized procedural rules, and knowledge how, fo r very specific behaviors through analogy with a series of examples and with the help of very general behavioral rules. It is then fine tuned over time as a function of cost effectiveness (probability of being correct and cost in terms of mental resources) The result of this last process is a gradual dropoff in reaction time and error rate (DeKeyser, 1997, p. 196197). Additionally, DeKeyser (1997) stated that the amount of improvement decreases as a function of the amount of practice. In E xperiment 1, R Ts were recorded from the visual presentation of an infinitive verb and the participants oral production of it s preterite form. In E xperiment 2, RTs measured the elapsed time it took the participants to decide whether a preterite verb form presented orall y and a written infinitive verb presented onscreen immediately afterward belonged to the same stem/verb or not. When the RTs for all verb forms decrease significantly from one proficiency group of late L2 learners to the next one, it can be hypothesized th at all verb forms have been automatized over time as a result of the amount of practice in the L2, irrespective of the regularity of the verb forms. But if the statistical analyses reveal that the RTs for regular verb forms are the only ones to decrease si gnificantly, it can be taken as strong support for the DP model (Ullman, 2001b) because this model posits that the main developmental pattern in the acquisition of complex verb
220 forms should be a shift from reliance on the declarative memory system to the p rocedural memory system for those verb forms which are handled by the procedural memory system by native speakers of the target language. A drop off in the RTs for regular verb forms only can be the result of function of cost effectiveness in the use of me ntal resources. In other words, what was initially memorized as a whole word (e.g., cant s he sang) is now handled by the mental rule add to ar verbs when the grammatical feature [3rd sing. Preterite] must be expressed overtly. The greater the amo unt of exposure or input for this rule, the lower the RT and the greater the accuracy for getting regular verb forms right when little or no attention is being paid to them The decision making task for Experiment 2 only required participants to press one of two buttons and the RTs from the groups of late L2 learners to press the button did not differ significantly The task was easy to perform and all groups of L2 learners were able to make a decision quickly, irrespective of their proficiency level, probably because of the design of the task since, for most of the cases, when the two words belonged to the same verb, these forms started with the same syllables (e.g., compr comprar s/he bought to buy) T his may therefore have contributed to quick R Ts even for the beginning participants. Similarly, when the two words belonged to different verbs, they had different initial syllables even if the intonation pattern was the same (i.e., both verb forms had the same number of syllables and the main accent occurred in the same position) and this may have also contributed to quick RTs by all groups of participants. The only statistically significant differences in the RTs by the three groups of late L2 learners were observed in Exper imen t 1 More importantly the dropoff in the RTs was found only for the regular verbs, regardless of their frequency in the target language. A quick look at
221 the averaged RTs for high and low frequency regular verbs by the intermediate and advanced groups of learners in compariso n to those by the beginning group shows that the former took about half t he time (measured in m s ) to produce those verb forms For example, while the beginning learners took 2732.21 ms. on average to produce the correct preterite forms for the high freque ncy regular verbs, the advanced group of learners only took 1420.08 ms. The significant decrease in these RTs can be the result of the shift f rom reliance on the declarative memory system to the procedural one in the mental representation of regular verb forms by late L2 learners. This shift would be taking place to maximize the effectiveness of the L2 learners mental representations of inflectional morphology because they would no longer be storing each regular verb form as a whole word; instead, they wo uld be computing and retrieving regular verbs via the online attachment of an inflectional suffix to an appropriate verb root. As mentioned by the DP model the processing of sequences is faster than the retrieval of arbitrary associations although the le arning speed for the processing of sequences is gradual and incremental rather than fast (hence the intermediate and especially the advanced groups of L2 learners showed significantly decreased R T scores) It is important to remember that if the significan t decrease in the RTs to the regular verb forms simply implied a process of automaticity of those verb forms as a result of the intermediate and the advanced groups of L2 learners having received more input and having had more opportunities to use them, th en there would be no reason not to find similar results for the other verb types. To p ut it differently, if the results simply indicate d a faster access of verb forms based on the frequency with which they ha d been encountered previously, then irregular an d stem change verb forms should have been accessed significantly faster by the intermediate and advanced groups too. These results were not found, lending support for the developmental patterns posited by the dual mechanism models, in
222 this case the shift o f reliance only for the regular verbs from the declarative to the procedural memory system as posited by the DP model (Ullman, 2001b). The developmental shift in the acquisition of the English irregular past forms ( went > *goed > went ) is considered a classic example of the developmental transition from exemplar based representations to more rulebased representations (McLaughlin, 1990, p. 118). The mental mechanisms to modify internalized cognitive representations are usually referred to as restructuring (McLaughlin, 1990). When this concept was applied to SLA, Lightbown (1985) suggested that the L2 learners own system of form function relationships or IL was characterized by backsliding and loss of forms that seemingly were mastered: [restructuring] occu rs because language is a complex hierarchical system whose components interact in nonlinear ways. Seen in these terms, an increase in error rate in one area may reflect an increase in complexity or accuracy in another, followed by overgeneralization of a newly acquired structure, or simply by a sort of overload of complexity which forces a restructuring, or at least a simplification in another part of the system (p. 177). Another way of understanding the results presented in Chapter 4 is to look for evid ence of restructuring or a transitional shift between two stages in the development of form function mappings, following E llis (1985) study (see Figure 51) 1. Internalization of new linguistic forms >>> assimilation + non systematic variation 2.a. Prog ressive organization of form function relationships >>> RESTRUCTURING 2.b. Elimination of redundant forms >>> economy principle + systematic variation Figure 51. Interlanguage processes during SLA [Adapted from: Ellis, R. (1985). Sources of variability in interlanguage. Applied Linguistics, 6 (Page 129).] The first stage is an assimilation stage in which L2 learners are open to new forms and develop hypotheses that may or may not correspond to the target language form function relationships ; e.g., a begi nning learner may incorporate the form puso s/he put as the preterite form for the verb poder to be able to. The L2 learners form function system has not yet integrated the new forms and, as a consequence, different forms may coexist for a given funct ion and hence the
223 number of errors is high due to the nature of the non systematic variation of L2 verb forms. In the second stage, the L2 learner accommodates her/his whole L2 system by restructuring the existing form function relationships in order to gi ve the new forms their own meanings to perform. Sometimes this restructuring stage may trigger the elimination of redundant forms, usually referred to as the economy principle, because those verb forms cannot be integrated into the new system by ensuring t hat they contribute to distinguish ing new meanings, e.g., an intermediate L2 learner may decide that deci s/he said does not have any distinguishable meaning and since it cannot be incorporated to any form function relationship it is eliminated from th e current IL, increasing her/his performance in the L2 since now the verb form dijo s/he said is the only form stored in the declarative memory system when the L2 learner needs to satisfy the grammar feature [+ preterite] for the verb form decir to say. This new systematic variation is what ultimately can l ead L2 learners towards a high proficiency level in the target language. L2 learners restructure their knowledge about the L2 until they sort out all the form function relationships. However, not all learners reach the last stage, which probably leads to fossilization of the current state of their IL system. The current results point towards some restructuring in the acquisition of verbal morphology by late L2 learners. As mentioned earlier, this process of restructuring occurs for regular verb forms only, which were initially assimilated as whole words and later on retrieved online via the attachment of a regular mental rule. If, as Lightbown (1985) suggested, an increase in error rate in one area may reflect an increase in complexity or accuracy in another (p. 177), then this new dual mechanism mental representation for complex verb forms (declarative memory for irregular verbs and procedural memory for regular verbs ) via the restructuring of the reg ular verb forms should trigger an increase in error rate for irregular verb
224 forms. Interestingly, that increase was present in the two production tasks even if it did not reach significance. For low frequency irregular verb forms the advanced learners ob tained lower accuracy scores than the intermediate learners in Experiment 1 ( M = 44.67, SD = 37.20 for the former, M = 45.33, SD = 38.52 for the latter ) as well as in the Production Task ( M = 45, SD = 31.62, and M = 50, SD = 29.89, in the same order ) Aga in, although the differences found in the accuracy scores were small, this result stands out in sharp contrast to the accuracy differences for the same verb type between the beginning (M = 22.67, SD = 26.31 in Experiment 1 and M = 25, SD = 28.35 in the Pro duction Task) and the intermediate learners since in those cases the accuracy scores doubled from one proficiency level to the next. Based on the averaged accuracy scores from these groups of part icipants there are reasons to believe that if true developm ental changes could be tracked for the same groups of participants over a long period of time, a backsliding in the accuracy scores for irregular verbs would accompany the abovementioned restructuring process for regular verb forms in a more clear way. Th e main issue with these results is that they can only be regarded as apparent, since the participants in each proficiency group were different, i.e., my study did not have a group of L2 learners from which data were collected at regular intervals for a lon g period of time. Interestingly, the developmental patterns found in favor of the DP model can be considered as an indication of what future L2 studies may expect from a tr ue longitudinal investigation. 5.2.3. Changes in the Rates and Types of Errors The results from the error types and error rates are less informative about the developmental patterns by L2 learners in the acquisition of verbal morphology because there were not many changes and those found in the data were not very strong. This last sectio n addressing the developmental patterns in the acquisition of verbal morphology evaluates the data on the participants main behaviors and error types in Experi ment 1 and the Production Task
225 In Experiment 1, all groups of late L2 learners made overregularizations for low frequency irregular verbs and did not change the vowel in the verb root for stem change verbs as the most prevalent error types for those test items. The only change in the error types among the proficiency groups for the remaining verb t ypes (high frequency irregular verbs and regular verbs) was that the beginning group produced those forms in a different tense, reflecting a selection of a verb form found more often in the L2 input than the elicited preterite form and since the processing of inflectional morphology by beginning learners is still restricted to the declarative memory system. The remaining results are not very informative due to the limited number of errors produced for high frequency irregular and regular verb forms by late L2 learners of intermediate or advanced proficiency levels. Perhaps worth mentioning is the fact that only the intermediate group made error 13, or the selection of a wrong root, for most regular verbs of low frequency and in both tasks. Even though this e rror could be simply interpreted as syllable confusion or isolated cases in the production of some L2 learners, it could also be indicative of the wrong selection of a verb root to which the corresponding regular suffix (rule 1 o r rule 2, mentioned in sect ion 5 .1.3.) was attached correctly, indicative of a step forward towards the use of both the declarative and the procedural memory systems, i.e., wrong selection of a verb root from the long term memory and online attachment of the regular TAM suffix based on the verb class. Similar results to those in Experiment 1 were found in the Production Task for stem change verbs (no vowel change) and low frequency irregular verbs (overregularization), which indicates that these error types linger in the production of late L2 learners and even in the production of English/Spanish bilinguals Finally, in this Production Task, most groups of participants used the regular verbs in nonfinite forms or even as nouns, which was p reviously
226 explained in section 5 .1.3. under s ome c laims of the Cognitive Approach: These examples may reflect the use of memorized chunks of language readily available in the declarative memory system which were retrieved to insert the elicited verb forms with fluency and in easier to compute forms ( i.e., unchanged as in the case of infinitives or as nouns in available fixed expressions such as: hice una llamada I made a phone call). 5.3. The Role of Age of Initial Exposure and Amount of Practice in the L2 The greater the amount of practice in the L 2, the greater the regular irregular dissociations in terms of accuracy scores, RTs and frequency effects as claimed by t he DP model (Ullma n, 2001b) for highly proficient L2 learners or the final stages of SLA In section 1.2.5 it was mentioned that, a ccording to this model, practice with the L2 in addition to age of exposure should affect both grammatical proficiency and the degree of dependence on procedural memory for grammatical computations. The later the initial age of exposure to the L2 and the l ess practice with it, the weaker the dissociations should be between the declarative/procedural memory systems for the computation of verb forms. That is the interest of the fourth research question, the differences between the two groups of advanced speak ers (late L2 learners vs. English/Spanish bilinguals ). The present study included a group of English/Spanish bilinguals who performed the same tasks as the groups of late L2 learners. The main differences between the two groups of advanced speakers are t heir initial age of exposure to L2 Spanish (around puberty for late learners vs. at age 5 or earlier for the bilinguals ), as well as their amount of exposure to the L2 (limited to the Spanish classroom for most late learners vs. frequently at home/ with fri ends, etc. for the bilinguals). If the DP model is correct in the claims related to the differences based on the initial age of exposure and the amount of practice, then there should be differences in the grammatical proficiency between the two groups of advanced speakers but there should not be
227 in relation to the dependence on procedural memory for grammatical computatio ns if, as suggested in section 5 .2.2., the advanced late L2 learners have already shifted reliance to the procedural memory for the compu tation of the regular verb forms. This latter claim is discussed in section 5 .3.2. The significant differences between these two gro ups and the differences in error types and error rates for specific verb forms are evaluated to the extent that they provide evidence in support of the single mechanism or the dual mechanism models. 5.3.1. Differences in the Accuracy Scores First, no significant results were found between the groups of advanced speakers in Experiment 2 due to the ceili ng effect described in section 5 .2.1. Second, in Experiment 1 bilinguals obtained statistically significant higher accuracy results for low frequency stem change verbs than the advanced learners and results almost reached significance in the same direction for low frequency irre gul ar verbs And third, in the Production Task the bilinguals obtained significantly higher accuracy scores than the advanced group for both low frequency irregular and low frequency stem chang e verbs The only verb forms in which the bilingual group show ed an advantage in comparison to the advanced group were those claimed to be handled by the declarative memory system, according to the DP model, provided they were of low frequency in the target language. The amount of exposure to Spanish by the bilingual group was greater than the advanced learners based on their age of initial exposure to Spanish. They had been exposed to Spanish for a greater number of years and more often and, as a result, they were more successful at retrieving low frequency forms fr om the declarative memory because they had undoubtedly received more input for them. Even though the averaged scores in the proficiency exam for the bilinguals (M = 31, SD = 4.58) and the advanced groups (M = 27.73, SD = 6.27) did not differ significantly ( t = 1.632, df = 28, p = .114) the bilinguals showed a better grammatical proficiency for irregular
228 and stem change verbs. Those are the verb forms handled by the declarative memory and those are the verb forms in which the bilinguals showed an advantage over the advanced L2 learners because irregular and stem change verbs are the verb types which benefit the most from input frequency. 5.3.2. Differences in the RTs It was in Experiment 1 where the intermediate but especially the advanced groups of late L 2 learners showed a significant decrease in the RTs for produc ing regular verb forms in comparison to the RTs by the beginning group. These results were explained as a possible change in the mental processing by late L2 learners from the declarative to the procedural memory systems, especially when these differences were not present for the other verb types and when the accuracy scores for low frequency irregular verbs decreased from the intermediate to the advanced proficiency groups. D ecreases in the accu racy scores are better understood in this type of data as being triggered by a restructuring of the learners IL. Since the group of bilinguals started their acquisition of Spanish at a very early age, their processing of the L2 should pattern more with th at of native speakers of the target language ; i.e., they should show greater dissociations between the declarative and procedural memory systems for the computation of L2 verb forms. T he independent samples t test results for the RTs on all the conditions show ed that the advanced L2 learners and the bilingual speakers did not differ in the ir RTs to any verb type. These results provide additional evidence that the advanced learners are processing the regular verb forms procedurally, as their RTs are similar to th ose of the bilingual group a nd are faster among both groups of participants than those for the irregular and stem change verb forms. Faster RTs for the regular verb forms are expected because the computation of those forms is fast and online rather th an being retrieved from the declarative memory system.
229 In Experiment 2 ther e were significant differences between the advanced learners and the bilingual gro ups for low frequency irregular verb forms. The se verb forms are retrieved from the declarative me mory system and since they a re of low frequency in the target language, late L2 learners did not have as much exposure to them as the bilinguals and, as a consequence, had to search longer in their mental lexicon for the appropriate verb form before making a decision. If the restructuring for the processing of regular verb forms is complete by this proficiency level ( rather than at the intermediate proficiency level ) as suggested in the present study ba sed on the observed patterns of results, then when the L2 learners listened to the low frequency irregular preterite forms in Experiment 2 they tried to make a decision based on whether the two forms would really fit under verb root regular form vs. verb root irregular form because both mental systems (declarative and procedural) are now active and work cooperatively in the computation of complex verb forms. To put it differently, they may be processing the low frequency irregular verbs via both the declarative and the procedural memory systems as poss ible whole word associations or rule like products That processing is different for the advanced learners since they already know when most verb root and irregular forms belong to the same verb but they are not sure about low frequency irregular items sin ce they had been exposed to those verbs less often and they evaluated both possibilities (rule like products vs. whole word associations) implicitly before making their final decision. This explains the significant differences in the RTs by the bilingual s peakers who, thanks to their greater amount of practice in Spanish, were capable of identifying the verb root irregular verb form associations of low frequency items quickly through the declarative memory alone 5.3.3. Differences in Error Types and/or E rror Rates The differences in the error types and/or error rates between the advanced learners and the bilinguals are evaluated qualitatively and quantitatively in the following paragraphs.
230 For stem change verb forms, regardless of their frequency in Spani sh, the main error type in both groups was not to change the vowel in the preterite forms elicited. The bilingual participants had fewer raw numbers of errors than the advanced L2 learners but the error rate of this error type in Experiment 1 was hi gher in the group of bilinguals than in the group of advanced learners. If the verb root was not identified as an irregular one by these groups of participants, they may have simply attached the default regular suffix for stem change verb forms. The total number of errors for stem change verb forms was the greatest for all the verb types and a possible explanation, besides the lower accuracy scores for stem change verbs in general, is that the vowel change in the root of those verbs only occurs for third person su bjects rather than in the whole preterite tense paradigm Additionally those verb forms are not encountered very often in Spanish, and frequency may have hindered L2 learners noticing of the stem change patterns, which also contributes to the most prevalent error type just mentioned. Both groups of participants made overregularizations as the most prevalent error type for low frequency irregular verb forms in both tasks. This error type shows that when L2 learners are not able to retrieve the appropriate irregular preterite form, they can attach to the verb root a regular ending by default in order to complete the task, although the accuracy scores do not reach high levels. Again, the bilingual group had fewer errors than the group of advanced learners. Th ese errors fully support the predictions of the DP model in that when an irregular verb form is not retrieved from the declarative memory system, the procedural memory system can compensate for that lack of knowledge by resorting to a default rule. The def ault rule attached incorrectly to these irregular verbs usually agreed with the verb conjugational class ; e.g., was attached to ar verbs, whereas i was attached to er / ir verbs. If frequency were the only factor responsible for these overregularizat ions, as single mechanism models suggest, then only
231 the most frequent regular rule should have been used by these participants since it appears in 88% of all the Spanish verbs Overregularizations are also produced in Experiment 1 by the advanced and the b ilingual speakers for high frequency irregular verbs. Additionally, the bilingual group produced some of these verbs in a different tense, although the number of occurrences was small: three cases. Furthermore, in the Production Task, both groups of partic ipants chose a form from a different tense instead of the elicited preterite form. The advanced learners also attached an incorrect TAM suffix to some of the correct irregular roots. These errors suggest that when these groups of participants narrate in th e L2, the frequency of specific verb forms for a given irregular verb may affect the selection of an incorrect verb form. Those verb forms chosen incorrectly were usually in the present tense, a tense which is introduced earlier and practiced more frequent ly in the classroom setting. Additionally, previous research has shown that even when participants are asked to produce a narrative in the past based on a visual prompt, advanced L2 learners and native speakers tend to use the present tense instead of the expected preterite tense, as mentioned in section 5 .1.3. In the case of regular verbs, both groups of participants showed the same pattern of results. In Experiment 1, the number of errors for regular verb forms was very low. The most prevalent error type for the low frequency items was the use of a TAM suffix from another verb class, e.g., attaching the suffix i to a verb from the first conjugation instead of to a verb from the second or third conjugations, as in invent i instead of invent s/he inve nted. However, in the Production Task, the most prevalent error type for regular verbs by both groups was the use of nonfinite forms, usually after a previous verb in the preterite tense. This error type can be explained by the task design in that partici pants were asked to narrate what a character did in the past based on a picturestory sequence. Sometimes the participants used a chunk of information
232 stored in their declarative memory and used it online in the ongoing narrative, in addition to avoiding t he computation of some verbs. For example, most of the advanced and bilingual participants used the sentence starts: tuvo que s/he had to decidi s/he decided to, estaba s/he was, followed by the regular verb in the infinitive form for the fir st two or followed by the present participle for the third one. These results can also be accounted for under the Cognitive Approach and the role of unanalyzed chunks of language o n which ongoi ng communication usually relies, as mentioned earlier in secti on 5.1.3. In sum, results indicated that the bilingual participants produced fewer errors than the advanced L2 learners although, for the most part, when comparing their most prevalent error types for each verb type (regular, irregular and stem change) th e groups did not differ qualitatively, which may indicate that the use of the same mental systems in the processing of complex verb forms lead s to the same error types even though the amount of practice helps in making fewer total errors. 5.4. Outline of Main Findings In the present study, the first and second research questions asked about possible frequency effects in the accuracy scores, RTs, and/or error types for complex verb forms in Spanish by different groups of participants Results indicated that there were regular irregular dissociations in the frequency effects in accuracy scores as well as RTs supporting the main predictions of the DP model. Specifically, irregular and stem change verb forms of high frequency were accessed and/or retrieved wi th faster RTs and greater accuracy than irregular and stem change verb forms of low frequency in L2 Spanish. These results were not observed in all tasks or for all groups of participants, but frequency effects were generally absent in the results of the r egular verbs. Taken together, frequency effects were indicative of associative memory effects for those items being handled by the declarative memory system, whereas the absence of
233 such frequency effects, especially in the RTs, supported the view that regu lar verb forms were computed online in the procedural memory system. The most prevalent error types also supported this dual mechanism processing, especially the overregularizations of low frequency irregular verb forms : The incorrect TAM suffixes attached to the verb root of irregular verb forms agreed with the conjugational class of the verb itself, rather than simply reflecting the attachment of the TAM suffix of the most frequent verbs in the target language ( ar regular verbs). The third research ques tion investigated the main developmental patterns in the acquisition of Spanish verbal morphology by late L2 learners. There were accuracy improvements for most high frequency items, especially those of irregular and stem change verbs, suggesting that the initial stages of SLA are usually lexical. However, the most interesting finding was the improvements in the RTs for regular verb forms only, pointing to a restructuring processing in the mental mechanisms used to retrieve them, from the use of explicit ru les memorized in the declarative memory system to the implicit use of mental rules that the L2 learners us ed quickly online. This shift from reliance on the declarative to the procedural memory systems for regular verb forms was claimed by the DP model and evidenced for the first time in the results of the present study. The fourth research question investigated the differences in L2 achievement for complex verb forms between two groups of advanced speakers, namely adult late L2 learners and English/Spanis h bilinguals Since both groups were using the two memory systems implicated in the processing of regular irregular verb forms, the only differences were observed in those items for which the greater amount of exposure by the bilinguals could have affected their retrieval:
234 low frequency irregular and low frequency stem change verb forms. The bilinguals had greater memory traces for these items than the group of advanced L2 learners. Finally, I have argue d that most of the data analyzed here can be better accounted for by the DP model than by the single mechanism models. Most research results stemmed from the predictions posited by the DP model, the dual mechanism models in broader terms, and even by the Cognitive Approach to language learning.
235 CHAPTER 6 CONCLUDING REMARKS T he present study examined the developmental changes in the acquisition of complex verbal morphology in L2 Spanish and the mental operations L2 participants at varying proficiency levels employed when storing retrieving and producing these verb s in the target language Data on the accuracy of and RTs for complex verb forms of different regularity types were collected through a production task as well as through a psycholinguistic experiment using recognition, in addition to an oral eli cited narrative. The results from this cross sectional study showed that, for the most part, all groups of participants displayed frequency effects (i.e., faster RTs and better accuracy scores for high frequency items) for those verb forms claimed by the D P model to be handled by the declarative memory system. No such frequency effects were found in the computation of regular verb forms, as measured by the RTs. Additionally, the data from the groups of late L2 learners indicated that the RTs for the regular verb forms only were significantly faster in the groups of advanced and even intermediate learners. These results support the claim that in the acquisition of an L2, late learners shift reliance from the declarative memory system to the procedural system for regular verb forms which, for late L2 learners at higher proficiency levels, were processed very quickly, irrespective of their frequency in the target language, a sign that these verb forms were computed online and via the use of implicit mental rules In previous studies, the use of mental rules for regular verbs as well as arbitrary associations in the computation of irregular verbs and stem change verb roots, i.e., dual mechanism processing, shows that L2 learners increased the complexity of the men tal representations of this grammar feature as their proficiency level increased No such improvements in the RTs for the retrieval of irregular and stem change verbs were observed among the gr oups of participants here even though the accuracy for the hig h
236 frequency items was greater for L2 learners at higher proficiency levels. The lack of those RT differences among the groups of late L2 learners suggests that the RT improvements for regular verbs were not the result of automatization in the processing of verb forms in general. Overregularizations in the production of low and even high frequency verbs as well as the absence of vowel changes for stem change verbs were the main error types found in the data of all groups of participants. These error types c oupled with the main regular irregular dissociations described in the previous chapter support the main claims of the DP model for SLA. Frequency and associative memory effects were found in the data, as predicted by singlemechanism models, but not to th e extent of fully accounting for all the acquisitional patterns observed in the data. Instead, dual mechanism models can better explain most of the results presented in the previous chapters. Results support the notion that the regular rules are not simply a descriptor for linguists but also a reality in the mental representations of complex, but structured, verb forms. In the following section, the main limitations of my study are des cribed, followed by suggestions for future research studies in the acquis ition of verbal morphology by L2 learners. The chapter ends with so me pedagogical implications and sugges ted activities to be used in Spanish classroom s of varying levels 6.1. Limitations and Future Directions The conclusions of this research investigatio n, while highly promising should nonetheless be interpreted cautiously as, with any research study, there are a number of limitations that should be addressed in future work One limitation was that the study was crosssectional and all developmental patt erns described in the results chapter can only be considered apparent. The participants in the different proficiency groups were independent samples and data from the L2 learners were not collected
237 more than once. A replication of the present study, but us ing one group longitudinally, would provide a better picture of the developmental changes in the learners mental representations of complex verb forms. R unning a longitudinal study like this would take several years of data collection however, which is b eyond the scope of a dissertation. Second, the design of the second psycholinguistic experiment favored the observed ceiling effect in the accuracy scores from all the groups of participants. A wider range of options than the two buttons same stem and d ifferent stem should have been considered in the design of the task. For example, pictures c ould have been provided with five drawings similar to those of the picturestor y sequences used in other tasks, and participants c ould have select ed the one that c orresponded to th e preterite form they had just heard. Or they may have been asked to produce orally the infinitive of the verb as quickly as possible after hearing the preterite form. Otherwise, this type of data collection measure should be used in simil ar studies only to collect RTs for which the measure provided very informative results, and not accuracy scores Third, the two picture story sequences had similar objectives and the design of the second (drawings only ) made it difficult to compare with t he rest of the tasks because the L2 participants could use any verb form they wanted. In that second production task, the absence of low frequency items and the recurrent use o f high frequency items w e re not very informative with respect to the mental mech anisms of L2 learners when handling complex verb forms. That is why although data were collected via this measure and coded accordingly, the information from this task was not included in the analysis In the future, researchers may consider providing the verbs for these types of tasks using picturestory sequences. Finally, the proficiency exams the participants completed were very difficult as well as long, especially for the advanced groups of learners (late L2 learners and the bilinguals). Even
238 though it was considered that the tests were us eful and informative in assess ing some of the language skills from the participants, most participants felt unhappy about the exam and informed the researcher that the exam was the hardest they had ever taken. Futur e studies might consider using other standardized proficiency exams. However, it should be noted that participants were not discouraged to the extent of dropping out of the study and no data were collected after the proficiency exams. These exams helped to truly distinguish between the proficiency groups and ma d e sure that the participant s in each proficiency level had comparable L2 skills. Even with these limitations, there are many reasons to have confidence in the validity of this study. The materials we re carefully designed from the selection of the test items based on multiple frequency dictionaries to the audio recordings by a native female speaker from Spain. T he se materials and the whole protocol were pilot tested and additional changes were made, s uch as the protocol checklist, to prevent data loss or artifacts during the data collection. In the psycholinguistic experiments, the participants decided when to move to the next test item and the RTs were corrected during the coding process as needed. T h e statistical methods used were selected in order to capture as much information as possible from the students responses. Finally, the results from the tasks were combined because each task was measuring different variables. This triangulation of results proved to be an e ffective way to provide more robust results than if a single production or recognition task was used. This study provided evidence in favor of the shift of reliance from the declarative memory system to the procedural memory system for reg ular verb forms for late L2 learners, as suggested by the DP model (Ullman, 2001b). T hese results in the averaged RTs for and accuracy scores of the test items were observed in the data from different groups of participants at varying
239 proficiency levels. T here is a need for longitudinal s tudies in this type of research, which is the only way to faithfully track changes in the mental processing of complex verb forms by developing L2 learners. Additionally, t here is a need to include some tasks in which littl e to no conscious attention (on the part of the participants) is paid to the verb forms based on the results from Experiment 2 In doing so, the RTs in the processing of regular verb forms only should not be affected, especially for L2 learners at the hig her proficiency levels. Future studies may consider including a task in which participants are presented drawings on the screen with the verb elicited but with a wider range of subjects (i.e., different PN suffixes elicited) to test the role of automaticit y in relation to verb types and their frequencies in the target language. The use of Spanish proved to be a successful choice in the study of complex verb forms as it has a wider range of verb types than English, as well as a wider variety of regular rules The overapplication of regular rules based on the verb class helped in providing stronger support for the dual mechanism models, as discussed in the previous chapter. The use of languages other than English may continue to provide better insights about t he processing of inflectional morphology. 6.2 Pedagogical Implications and Suggested Classroom Activities The main goal of this study was to account for the developmental changes in the acquisition of complex verbal morphology in Spanish as well as to in vestigate the mental mechanisms by L2 learners when processing inflectional morphology. This is a study about the developing grammars of L2 learners and the implications of these developments in the ongoing theoretical debate about the processing and menta l representation of language in broad terms. However, the findings of the study also have pedagogical implications because the participants were primarily classroom learners whose knowledge of Spanish came from formal instruction. W hat follows is an outlin e of the main implications with sample activities that can be
240 incorporated in the L2 classroom to aid learners in the acquisition of inflectional morphology, based on the way they handle it and the recurr ing errors in their production. Irregular verb form s as well as the vowel changes in the stem change verbs were posited as being handled by the declarative memory and, as a consequence, there were frequency effects in the accuracy and the RTs Retrieval of low frequency items was challenging for all groups of L2 learners and the lack of vowel change for stem change verb forms was a recurring error, even for highly proficient learners. The former was the result of limited exposure to those forms, which hindered the arbitrary connections between the verb root and the corresponding irregular form/irregular verb stem ; the latter was t he result of lack of noticing of the vowel change for some subjects in the preterite paradigm of stem change verbs. Since irregularities are handled in the declarative memory system L2 learners may benefit from activities that strengthen the verb root preterite form connections explicitly. Since the vowel change has to be perceived first before it can be stored in the declarative memory system, inductive tasks are proposed to help L 2 learners discover the grammar rules by themselves through guided activities. As for regular verbs, consciousness raising tasks as well as speeded activities may aid L2 learners to detect the regular patterns and apply them more quickly so that the regular rules can be computed procedurally instead of explicitly. The following exercises are proposed keeping in mind the discussion points just mentioned. 6.2.1. Concentration Game L2 learners need to memorize the irregular verb forms so that in their produc tion they do not fail to retrieve them. The associative connections between the base form (infinitive) and the irregular forms (in this case, the preterite tense) can be strengthened with the use of a concentration game, as shown in Figure 61. The teacher d ivide s the class in small groups of about 4 or 5 students. Each group will need a set of about 30 40 cards. On half of the cards, the
241 teacher write s the infinitive form of the verb. On the other half of the cards, the teacher write s the irregular preteri te forms of those verbs. After shuffling the cards they are placed in a grid fac ing down Students will take turns turning over two cards at a time. If the two cards match (i.e., if they belong to the same verb, as shown in Figure 6 1), the student keeps t he pair of cards; otherwise, they are replaced face down in their original positions. Figure 61. Concentration game for irregular verb forms The winner of the game is the student with the most cards at the end of the game. This concentrati on game can be played from time to time as new irregular verbs, especially those of lower frequency in Spanish, are introduced in the course. A more difficult version of the game for students at the intermediate/advanced proficiency levels can be to includ e baseirregular form connections with different subjects, i.e., from the whole preterite paradigm. That way, one can ask the students to state the subject of the base form irregular preterite matching cards before they can keep the pair of cards. 6.2.2. R eading Passages with Input Enhancement The regular rules as well as the vowel stem changes in the Spanish input can be somehow enhanced (Sharwood Smith, 1993) to make the verb endings and the vowel changes more salient. vino venir
242 In 2008, Lee & Huang conducted a me ta analysis in this research domain with a systematic synthesis of 16 studies and they found that learners exposed to enhanced texts outperformed those who read unenhanced texts. However, the authors indicated that the observed effect was smallsized ( d = 0.22) and that the comparisons were provided with another experimental groups (input flood) rather than with true control groups (Lee & Huang, 2008, p. 323). One can f ind a real text in Spanish, or create a short reading passage, with most of the verbs bei ng regular and stem change and enhance them by boldfacing, italicizing or underlining their endings and vowel changes, as shown in Figure 62. Figure 62. Reading passage with input enhanced The teacher can ask the students comprehension questions onc e they have read the passage and then the teacher can draw their attention to the particular basepreterite changes by using guided yes/no questions such as: En qu terminan los verbos 1 y 4? Whats the ending for verbs 1 and 4?, Cu l es su infinitivo ? Whats their infinitive form?, Qu podemos concluir sobre las formas de pretrito de los verbos 1 y 4? What can we deduce about the preterite forms of verbs 1 and 4? Juan Pablo (1) nac i el 15 de mayo de 1979. (2) Durmi casi todo el da despus de abrir sus ojitos por prime ra vez. (3 ) Sinti hambre y (4) bebi leche materna cada 3 horas. Su mam lo (5) bes mucho y su hermano mayor le (6) regal un peluche. Juan Pablo (1) was born on the 15th of May, 1979. He (2) slept most of the day after opening his eyes for the first ti me. He (3) was hungry and (4) drank milk every 3 hours. His mom (5) kissed him a lot and his older brother (6) gave him a cuddly toy as a gift.
243 6.2.3. Multiple Choice Activity with Stem Change Preterite Forms Another activity t hat can help L2 students to pay attention to the vowel changes for some subjects of stem change verbs in the preterite is a multiplechoice exercise. Teachers can create a reading passage with a variety of stem verbs in the preterite and make sure it inclu des low frequency stem change preterite forms as well as fully regular verb forms from the preterite paradigm, e.g., ment I lied (regular verb form) as well as minti s/he lied (stem change verb form) preterite forms, as shown in the following example: La semana pasada sal con mi mejor amiga al cine. Ella estuvo muy feliz todo el da y (1) _______________ de cada chiste o historia que le cont. Fuimos al cine pero como la pelcula fue tan aburrida (2) _______________. Para cenar fuimos al nuevo restaurante mexicano del centro comercial Yo (3) _______________ una quesadilla de pollo y ella (4) _______________ unos tamales. Pero el camarero (5) _______________ la orden de otra mesa. Last week I went out with my best friend to the movies. She was very happy all day long and (1) __________ every joke or story I told her. We went o the movies but since the movie was so boring we (2) __________. For dinner we went to the new Mexican restaurant in the mall. I (3) __________ a chicken quesadilla and she (4) __________ tamales. But the waiter (5) __________ the order from another table. (1) a. se ri b. se rei c. me re d. se re (2) a. nos durmimos b. se duerme c. dormirnos d. nos dormimos (3) a. pido b. pid c. ped d. pidi (4) a. pedi b. pide c. pidi d. ped (5) a. nos servi b. nos sirve c. nos servimos d. nos sirvi The teacher w rite s the regular and stem change choices for each verb in addition to the incorrect answers so that L2 students are indirectly reminded that those verbs can h ave a vowel change in their root. At the end of the activity, one can ask the students to report how many verbs from the correct answers changed the vowel of their roots in the preterite forms. The teacher can h elp them induce that the vowel change only oc curs for some subjects and the teacher can ask the students to provide some examples from the text.
244 6.2.4. Verb Tree Poster for Patterns of Changes Students can practice all types of morphological changes in different groups of Spanish verbs, whether thes e patterns are considered regular (for most Spanish verbs of the same class) or irregular (for a limited number of verbs), using a verb tree poster as shown in Figure 6 3. Figure 63. Verb tree poster and suggested flashcards dormir durmi contraer contrajo atraer atrajo conducir condujo introducir introdujo caber cupo haber hubo poder pudo seguir sigui sentir sinti servir sirvi mentir minti comprar compr cantar cant llamar llam abrir abri vivir vivi pedir pid i pagar pag morir muri traer trajo reducir redujo poner puso salir sali nacer naci vender vendi comer comi beber bebi subir subi
245 Teachers can m ake a poster with a tree with many branches Each tree branch must have an infinitive preterite example for the verb conjugation pattern that the branch represents. Each student is given a flashcard that must be placed into one of the remaining spots in the tree branches, based on the morphological changes in the preterite form. A sample verb tree and complete set of suggested flashcards are included in Figure 63. A follow up activity can be to ask them write a story in pairs or small groups in Spanish using eight preterite forms, each from a different tree branch, and then have them read it to the class or act it out in front of the class. 6.2.5. Quiz Show Game Regular verb forms can be practiced in speeded tasks with the use of the Powerpoint software program to force the students to retrieve or simply recognize those forms as quickly as possible. Teachers can d esign a Powerpoint presentation of about 30 40 slides with some drawings for regular verbs. Teachers d ivide the class in to two big groups and have one member from each come to the front of the class to represent her/his group. For half of the slides, one can provide a drawing and four written options with only one correct choice. For the other half of the slides, one can simply provide a drawing and a personal subject pronoun, as shown in Figure 64. 1. pagu 2. paga 3. pag 4. pagar [ ellos ] Figure 64. Samples of Quiz Show Game slides
246 For the former, the student who chose the correct option the fastest wins a point. For the latter, the student who provides the correct preterite form the fastest using the appropriate verb suggested by the drawing wins a point. Once a student provides an answer, s/he will not be able to provide additional ones and the turn is h anded to the other group. If both groups get the answer wrong, the correct answer is provided by the teacher and the game continues, with no points being deducted. The group with the greatest number of points at the end of the quiz wi ns. 6.2.6. Correct the Errors The final suggested activity can be used for advanced L2 learners to offer them more input in low frequency items as well as to direct their attention to frequent errors they make when producing some preterite forms. Teachers can f ind a reading pa ssage such as a piece of news which contains a lot of low frequency irregular and stem change verbs. One can c hange some of the verbs to erroneous verb forms, mainly regularizing irregular verbs and keeping unchanged the vowel in the root of stem change ve rb forms. The instructions can tell the students that they have been hired as proofreaders for a newspaper in Spanish. Their task is to r ead the passage, find the typos and spot errors and correct them. Follow up discussion questions can be included so th at students practice the verb forms just read in the passage. Based on the results from this study, beginning learners seem to use explicit grammar rules to compute regular verbs whereas irregular verb forms are stored in and retrieved from long term memor y. Activities that aid in the extraction of the regular patterns of regular verbs (e.g., r eading passages with i nput e nhancement ) or activities that promote the arbitrary connections between roots and preterite forms of irregular verbs ( e.g., concentration games ) are best suited for their needs. Intermediate learners use the declarative memory system to a great extent because it helps them in the retrieval of greater numbers of forms. If they are able to identify patterns in the language (whether in the reg ular, irregular, or vowel stem change verbs),
247 that may force them to restructure their IL system ( e.g ., m ultiple choice activit ies and verb tree posters ). Additionally, speeded tasks in the computation of regular verbs can help them retrieve those forms wi th greater automaticity ( e.g., quiz show games ). Advanced and bilingual learners may not benefit from additional practice of regular rules. Their main problems stem from low frequency irregular and stem change items. They need more input with them as well as explicit awareness of the main error types for those forms to avoid fossilization of erroneous patterns in their developing grammars (e.g., activities to correct errors ). The previous activities are not meant to change the route of acquisition of Spanis h verbal morphology by L2 learners. Instead, they are meant to address some of the most challenging areas in which even highly proficient learners fail to achieve nativelike competence as well as to provide precise aid in the mental mechanisms L2 learners use in the processing of complex verb forms, based on their profi ciency level and the main findings of this study. 6.3. Final Remarks The present study has contributed to our understanding of the mental representation of verbal morphology in the acquisit ion of Spanish as an L2. The late L2 learners started their route of acquisition memorizing verb forms and explicit rules taught in the classroom, but the advanced learners were able to extract patterns from the input, deduce the rules for the regular verbs, and compute those forms online by concatenating the correct TAM suffix to the verb root. In doing so, the declarative memory system of the se advanced L2 learners was not overload ed with hundreds of redundant regular verb forms, even though that was obse rved in the beginning group. This restructuring of verb forms indicate s that L2 learners have achieved a more complex representational system capable of computing complex verb forms declaratively or procedurally. This is the first time this shift in relian ce from the declarative memory system to the procedural memory system for regular verb forms has been tested and evidenced in the production of L2
248 learners. This is an important contribution to the debate on the processing of verbal morphology because it evidences that not all verb types are processed equally by L2 learners at varying proficiency levels.
249 APPENDIX A FOREIGN L ANGUAGE B ACKGROUND F ORM Gender : Male Female Age: Your current rank at UF is (circle one if app licable ): Freshman Sophomore Junior Senior Graduate When writing, I am Right handed Left handed Do you need some visual aids when reading from a computer screen or a printed material (circle one)? No Yes, I need contact lenses Yes, I need glasses Do you have any known learning disabilities (circle all that apply)? None Dyslexia Dysgraphia Auditory and/or processing disorders ADHD (attention deficit/hyperactivity disorder) Other. Please, specify __________________________________________ What is your native(s) language(s)? _________________________________________________ Have you ever studied Spanish in a formal instruction setting before entering the University of Florida ? Yes No If yes: Whe re? ____________________________________________ For how many years? _________________________________ Do you usually socialize with anyone outside the class (friends or family members) who only speaks Spanish with you? Yes No If yes: How often? _________________________________________ Where do they live? __________________________________ Have you lived in a Spanish speaking country? Yes No If yes: Where? (country/s) ___________________________________ For how many mon ths/years ? ___________________________
250 After this class, do you have plans to continue studying Spanish (if applicable ) ? Yes No How would you rate your Spanish proficiency in the following skills (circle one option for each skill according to the scale): very highly poor poor low average good advanc. proficient Spanish listening 1 2 3 4 5 6 7 Spanish reading 1 2 3 4 5 6 7 Spanish speaking 1 2 3 4 5 6 7 Spanish writing 1 2 3 4 5 6 7 Spanish vocabulary 1 2 3 4 5 6 7 Spanish grammar 1 2 3 4 5 6 7 Thinking only about English and Spanish, what language(s) do you use the most in the following contexts (circle one for each context)? At current home: only Engl. mostly Engl. mostly Span. only Span. At school: only Engl. mostly Engl. mostly Span. only Span. At work (if applicable): only Engl. mostly Engl. mostly Span. only Span. With friends: only Engl. mostly Engl. mostly Span. only Span. With family: only Engl. mostly Engl. mo stly Span. only Span. Please, translate the following list of Spanish verbs into English. If you do NOT know the meaning of a verb but you have an intuition or a vague idea of its possible meaning, write that down too. If you do not have any idea, leave it blank. For example: Probar > a movement of some sort______ Abrir > ___________________________________________________ Advertir > ___________________________________________________ Andar > ___________________________________________________ Atraer > ___________________________________________________ Besar > ___________________________________________________ Caber > ___________________________________________________ Cantar > ___________________________________________________ Conducir > ___________________________________________________ Conseguir > ___________________________________________________ Contraer > ___________________________________________________ Convertir > ___________________________________________________ Corregir > ___________________________________________________ Decir > ___________________________________________________ Despedirse > ___________________________________________________ Divertirse > ___________________________________________________ Dormir > ___________________________________________________
251 Empezar > ___________________________________________________ Entrar > ___________________________________________________ Escapar > ___________________________________________________ Esperar > ___________________________________________________ Estar > ___________________________________________________ Extraer > ___________________________________________________ Haber > ___________________________________________________ Hacer > ___________________________________________________ Herir > ___________________________________________________ Introducir > ___________________________________________________ Inventar > ___________________________________________________ Ir > ___________________________________________________ Llamar > ___________________________________________________ Llevar > ___________________________________________________ Medir > ___________________________________________________ Mentir > ___________________________________________________ Mirar > ___________________________________________________ Morir > ___________________________________________________ Nacer > ___________________________________________________ Olvidar > ___________________________________________________ Pagar > ___________________________________________________ Pedir > ___________________________________________________ Poder > ___________________________________________________ Poner > ___________________________________________________ Preguntar > ___________________________________________________ Producir > ___________________________________________________ Reducir > ___________________________________________________ Regalar > ___________________________________________________ Rerse > ___________________________________________________ Repetir > ___________________________________________________ Salir > ___________________________________________________ Seguir > ___________________________________________________ Sentir > ___________________________________________________ Servir > ___________________________________________________ Sonrer > ___________________________________________________ Tener > ___________________________________________________ Tomar > ___________________________________________________ Traducir > ___________________________________________________ Traer > ___________________________________________________ Vender > ___________________________________________________ Venir > ___________________________________________________ Vestirse > ___________________________________________________ Verter > ___________________________________________________ Viajar > ___________________________________________________
252 APPENDIX B PARTICIPANTS PAST T RAVEL EXPERIEN CES Summary of participants destinations in the Spanishspeaking country and duration of the stay Group and N Participant ID Destination Duration in months Average d travel experiences for the group Beginning (0/15) NA NA NA 0 months Intermediate (2/15) Participant 18 Spain 1.0 1.25 months Participant 30 Spain 1.5 Advanced (11/15) Participant 3 Peru 5.0 3.0 months Participant 7 Honduras 2.0 Participant 8 Mexico 1.5 El Salvador 0.5 Participant 37 Spain 4.0 Participant 39 Spain 5.0 Par ticipant 40 Mexico 1.5 Participant 41 Spain 1.5 Participant 42 Spain 5.0 Participant 43 Spain 4.0 Participant 44 Spain 1.5 Participant 46 Spain 1.5
253 APPENDIX C FREQUENCIES FOR THE FILLER ITEMS Token count, frequencies per million words f rom the three corpora, and averaged frequency for high frequency masculine fillers CREA CDE CUMBRE Averaged Freq Cases Freq. Cases Freq. Cases Freq. Mundo ( world ) 94,895 629.37 12,409 609.78 1,374 66.50 435.22 Momento ( moment ) 87,145 577.97 10,486 515.28 1,536 74.34 389.20 Hombre (man ) 79,463 527.02 9,958 489.34 1, 802 87.21 367.86 Lugar ( place ) 79,008 524.00 9,885 485.75 971 46.99 352.25 Estado ( state ) 47,009 311.77 11,957 587.57 1,570 75.98 325.11 Fin ( end ) 60,671 402.38 7,443 365.75 879 42.54 270.22 Grupo ( group ) 57,289 379.95 6,396 314.30 1,007 48.74 247.66 Padre ( father ) 42,155 279.58 6,536 321.18 1,059 51.25 217.34 Problema ( problem ) 43,790 290.43 5,739 282.01 1,273 61.61 211.35 Seor (man ) 44,293 293.76 5,244 257.69 1,338 64.76 205.40 Pueblo ( village ) 39,898 264.61 4,990 245.21 879 42.54 184.12 Hijo ( son ) 29,550 195.98 4,420 217.20 1,186 57.40 156.86 Proyecto ( project ) 39,353 261.00 3,217 158.08 884 42.78 153.95 Programa ( program ) 31,450 208.58 3,250 159.71 973 47.09 138.46 Mes (month ) 27,743 184.00 2,867 140.88 992 48.01 124.30
254 Token count, frequencies per million words from the three corpora, and averaged frequency for high frequency feminine fillers CREA CDE CUMBRE Averaged Freq Cases Fre q. Cases Freq. Cases Freq. Vida (life ) 124,349 824.71 14,957 734.99 1, 516 73.37 544.36 Forma ( shape ) 98,006 650.00 13,561 666.39 1,097 53.09 456.49 Casa ( house ) 80,327 532.75 12,433 610.96 1, 373 66.45 403.39 Poltica ( politics ) 73,650 488.46 7,964 391.35 792 38.33 306.05 Gente ( people ) 52,769 349.98 8,896 437.15 1,041 50.38 279.17 Manera (manner ) 61,045 404.86 7,474 367.27 926 44.82 272.32 Mujer ( woman ) 60,221 399.40 6,945 341.28 1, 253 60.64 267.11 Noche ( night ) 56,184 372.62 7,152 351.45 1,101 53.29 259.12 Verdad ( truth ) 46,560 308.80 7,323 359.85 839 40.61 236.42 Mano ( hand ) 46,813 310.47 5,550 272.73 1, 310 63.40 215.53 Situacin ( situation ) 55,307 366.81 4,788 235.28 912 44.14 215.41 Hora ( hour ) 40,878 271.11 4,410 216.71 1, 175 56.87 181.56 Persona ( person ) 34,611 229.55 4,745 233.17 1, 230 59.53 174.08 Relacin ( relation ) 44,921 297.93 3,534 124.52 873 42.25 154.90 Economa ( economy ) 19,599 129.99 3,118 153.22 782 37.85 107.02
255 Token count, frequenc ies per million words from the three corpora, and averaged frequency for low frequency masculine fillers CREA CDE CUMBRE Average d Freq. Cases Freq. Cases Freq. Cases Freq. Dictador ( dictator ) 2,172 14.41 274 13.46 (< 142) (< 6.87) 13.94 Fotgrafo ( photographer ) 2,221 14.73 235 11.55 (< 142) (< 6.87) 13.14 Prrafo ( paragraph ) 3,075 20.39 116 5.70 (< 142) (< 6.87) 13.05 Postre ( dessert ) 2,583 17.13 150 7.37 (< 142) (< 6.87) 12.25 Perfume (pe rfum e) 1,655 10.98 259 12.73 (< 142) (< 6.87) 11.86 Terremoto ( earthquake ) 1,679 11.14 174 8.55 (< 142) (< 6.87) 9.85 Lpiz ( pencil ) 1,222 8.10 218 10.71 (< 142) (< 6.87) 9.41 Armario ( closet ) 1,563 10.37 112 5.50 (< 142) (< 6.87) 7.94 Cheque ( check ) 1,195 7.93 118 5.80 (< 142) (< 6.87) 6.87 Hebreo ( Hebrew ) 529 3.51 181 8.89 (< 142) (< 6.87) 6.20 Diamante ( diamond ) 484 3.21 142 6.98 (< 142) (< 6.87) 5.10 Insulto ( insult ) 755 5.01 83 4.08 (< 142) (< 6.87) 4.55 Notario ( notary ) 877 5.82 66 3.24 (< 142) (< 6.87) 4.53 Coloquio ( colloq uium ) 698 4.63 61 3.00 (< 142) (< 6.87) 3.82 Veterinario ( veterinarian ) 446 2.96 35 1.72 (< 142) (< 6.87) 2.34
256 Token count, frequencies per million words from the three corpora, and averaged frequency for low frequency feminine fillers CR EA CDE CUMBRE Averaged Freq Cases Freq. Cases Freq. Cases Freq. Tolerancia ( tolerance ) 2,448 16.24 279 13.71 (< 142) (< 6.87) 14.98 Nacionalidad ( nationality ) 2,677 17.75 191 9.39 (< 142) (< 6.87) 13.57 Discriminacin ( discrimination ) 2,372 15.73 185 9.09 (< 142) (< 6.87) 12.41 Biografa ( biography ) 2,125 14.09 211 10.37 (< 142) (< 6.87) 12.23 Vitamina ( vitamin ) 2,487 16.49 129 6.34 (< 142) (< 6.87) 11.42 Caridad ( charity ) 1,513 10.03 252 12.38 (< 142) (< 6.87) 11.21 Medianoche ( midn ight ) 1,728 11.46 208 10.22 (< 142) (< 6.87) 10.84 Calificacin ( grade ) 1,750 11.61 197 9.68 (< 142) (< 6.87) 10.65 Cueva ( cave ) 1,367 9.07 241 11.84 (< 142) (< 6.87) 10.46 Chaqueta ( jacket ) 1,918 12.72 164 8.06 (< 142) (< 6.87) 10.39 Motivacin (motivation ) 1,648 10.93 157 7.71 (< 142) (< 6.87) 9.32 Blusa ( blouse ) 981 6.51 140 6.88 (< 142) (< 6.87) 6.70 Partcula ( particle ) 558 3.70 150 7.37 (< 142) (< 6.87) 5.54 Pastilla ( pill ) 459 3.04 81 3.98 (< 142) (< 6.87) 3.51 Motocicleta ( mo torcycle ) 377 2.50 61 3.00 (< 142) (< 6.87) 2.75
257 APPENDIX D INSTRUCTIONS FOR EXP ERIMENTS 1 AND 2 Instructions for Experiment 1 The screen will display a small green circle ( ). Press the green button from the color coded button box to start the activity. The computer will display either a verb or a noun in Spanish at the center of the screen. Words are color coded. VERBS will be displayed in a blue font and in their infinitive form. NOUNS will be displayed in a black font and in their singular fo rm. There are a total of 120 words. First, read the verb or the noun silently Words wil l remain onscreen indefinitely. Second, if the word is a VERB you need to say into the microphone its third person singular PRETERITE as quickly as possible For exampl e, if the verb BEBER is displayed on the screen, you need to say BEBI into the microphone as quickly as possible. Please, say it clear and loud If the word is a NOUN you need to say into the microphone its GENDER as quickly as possible Only two options are possible: masculine or feminine. For example, if the Spanish noun FLOR is displayed on the screen, you need to say FEMENINE into the microphone as quickly as possible. Again, please say it clear and loud Words will disappear from the screen at the mom ent you start providing an answer into the microphone. Please, do not say anything into the microphone but your final response that is, the one you want to provide. It is important that you do not cough, sigh, mumble, murmur or alike as this will automati cally trigger the disappearance of the word from the screen before you had a cha nce to think about your answer. If you realize you made a mistake right after you provided an answer, you can say a new answer into the microphone before moving on to the next word But if you realize about your mistake later during the experiment, do not worry about it. Better concentrate on the upcoming words Finally, after 1,5 seconds from the moment you provided your answer for each word into the microphone, the small gree n circle ( ) will automatically appear at the center of the screen. When you are ready for the next word, press the green button again If you need a small pause or break during the experiment, you can stop at any of the screens with a small green circle before moving on to the next word. If you have any questions, please ask the experimenter right now Before starting the activity, lets make sure you understand the procedure by working with 10 practice words So get ready and start these 10 practice w ords!
258 In structions for Experiment 2 The computer screen will display the question: Ready? Press any button from the color coded button box to start the activity. Right after pressing a button, the computer will display a small red cross ( + ) at the center of the screen while you listen to a word in Spanish via the headphones. It can be either a verb in the preterite or a singular noun. You will listen to the word only once. There are a total of 120 words. Right after the word is pronounced in full, a wr itten word will be displayed on the computer screen. It can be either a verb in the infinitive or a plural noun in Spanish. You have to decide if the written word shares the same stem as the word you just listened to or if it has a different stem The writ ten word will remain onscreen indefinitely until you make a decision by either pressing the SAME STEM red button ( ) or the DIFFERENT STEM purple button ( ) fr om the color coded button box. For example, if you heard the word CUADERNO and the screen displayed COMPUTADORAS, you need to press the DIFFERENT STEM purple button However, if you heard CUADERNO and then CUADERNOS was displayed on the screen, the words shared the same stem even though the written version was in its plural form and, consequent ly, you need to press the SAME STEM red button Similarly, if you heard COMPR and the screen displayed ESCUCH, press the DIFFERENT STEM purple button But if you heard COMPR and then COMPRAR was displayed on the screen, the words shared the same stem ev en though the written version was in its infinitive form so you need to press the SAME STEM red button. Please, press any of the two buttons as quickly as possible If you realize you made a mistake right after you pressed a button, inform about it to the experimenter immediately But if you realize about your mistake later during the experiment, do not worry about it. Better concentrate on the upcoming words Right after pressing one of the two buttons, the screen will display the Ready? question again. When you are ready for the next word, press any button f rom the color coded button box. If you need a small pause or break during the experiment, you can stop at any of the Ready? screens before moving on to the next word. If you have any questions, please ask the experimenter right now Before starting the activity, lets make sure you understand the procedure by working with 10 practice words So get ready and start these 10 practice words
259 APPENDIX E PROTOCOL C HECKLIST Participant ID: _________________ Date: __________ Group: __________ Before the participant arrives Check that the headphones work, the volume level and microphones sensitivity With the participant Make the participant complete the sign in page. Make the participant to read and sign the consent form Write his/her ID code on top of the consent form page. Give him/her a copy. Complete the experiment log sheet Instruct the participant to turn off the cell phone during the data collection. Turn OFF the network on the c omputer. Decide the order of the 2 psycholinguistic tasks: __________ Place the tags on the appropriate buttons. Read this introductory text : Today you will be asked to complete two activities on this computer. But, first of all, relax and sit at a comfortable distance from the computer screen. You need to wear a set of headphones with a microphone attached to it. Make sure that it is placed correctly over your head and that the microphone is in front of your mouth For TASK 1 : Read next page. For TASK 2 : Have the participant read the instructions. (Change the tags, if applicable) Open Psyscope dragging the Main Script file. Run the experiment by typing the participants code number: ______ Additional comments here For task 1: Have the participant read the instructions. (Change the tags, if applicable: NEXT tag) Start Audacity with file > new > (start recording). Open Psyscope dragging the Main Experiment Script file. Run the experiment by typing the participants code number: ______
260 Mesa Additional Comments here Caer Impedir Profesor Carro Invertir Cocinar Cancin Querer Comer Prrafo Hijo Extraer Calificacin Economa Tomar Pedir Hebreo Traer Vitamina Blusa Regalar Veterinario Postre In sulto Poltica Situacin Seguir Proyecto Escapar Traducir Poder Relacin Manera Inventar Llevar Caber Corregir Noche Tener Problema Hacer Poner Salir Hombre Venir Introducir Hora Pagar Morir Andar Decir Producir Conducir Esperar Vida Perfume Mundo Gente Mano Dictador Sonrer Chaqueta Advertir Motocicleta Besar Terremoto Notario Mirar Estado Empezar Mes Despedirse Diamante Preguntar Abrir Mujer Tolerancia Momento Pastilla Divertirse Dormi r Discriminacin Haber Persona Cantar Estar Conseguir Forma Verter Caridad Biografa Entrar Seor Ir Cueva Padre Verdad Coloquio Vender Fin Viajar Armario Reducir Lpiz Contraer Rerse Sentir Fotgrafo Pueblo Llamar He rir Programa Servir Lugar Cheque Partcula Repetir Mentir Vestirse Olvidar Medir Atraer Motivacin Medianoche Nacionalidad Casa Nacer Convertir grupo Stop the recording and then file > save project as > IDaudiolab.aup
261 Make t he student pick one picture story among the ones for the man Read the instructions for Picture description activity 1 : Please, pick one of these pages but do not show it to me. It will contain a picturestory sequence of a character that you need to narrate to me. Please, tell me in Spanish what this person did in the past. You have to use the verbs provided for each drawing in the preterite tense. My task will be to guess which picture story you have just selected once your narration is complete. You n eed to provide as much information as possible because all the picture story sequences are very similar and they only differ in a couple of drawings. Do not disclose the name of the character to me, which is at the right bottom of the page, until your narr ation is complete. Instead, refer to the character as L or ELLA. Again, use the verbs provided in the preterite form and provide as much information as possible. There is no time limit to complete this task. Start Audacity with file > new > (start re cording). Write the responses in the following chart during the data collection: Nacer Llamar Esperar Venir Regalar Poner Verter Tomar Mirar Medir Corregir Mentir Escapar Caber Poder Decir Repetir Conseguir Inventar Reducir Andar Sentir Extr aer Seguir Pretend you guessed the characters name: Ral Make the student pick one picture story among the ones for the woman Read the instructions for Picture description activity 2 : As before, pick one of these pages but do not show it to m e. Again, tell me in Spanish what this person did last month. This time there will not be written verbs on the page but you still have to narrate all the events by using the preterite tense. My task will be to guess which picture story you have just select ed once your narration is complete. Do not disclose the name of the character to me, which is at the right bottom of the page, until your narration is complete. Instead, refer to the character as L or ELLA. Again, use verbs in the preterite form and provi de as much information as possible. There is no time limit to complete this task. Write the responses in the following chart during the data collection:
262 Dormir Vestirse Entrar Hacer Salir Abrir Conducir Introducir Pedir Traer Servir Pagar Cantar Viajar Traducir Ir Estar Sonrer Rer Preguntar Besar Despedirse Tener Morir Pretend you guessed the characters name: Luisa Read the instructions for the questions: In the previous tasks, you had to use some verbs in the preterite. I would like you to tell me what is it you did inside your mind right before you produced different verbs in Spanish in their preterite forms. You can answer to me in English now. 1. When you had a verb like CANTAR or BEBER and you wanted to provide a preterite form what is it you did mentally before producing the verb forms CANT or BEBI? 2. And when you had a verb like PEDIR or DORMIR and you wanted to provide a preterite form, what is it you did mentally before producing the verb forms PIDI or DURMI? 3. Finally, whe n you had a verb like HACER or IR and you wanted to provide a preterite form, what is it you did mentally before producing the verb forms HIZO or FUE? Stop the recording and then file > save project as > IDaudiodescr.aup Make the participant fill ou t the Foreign Language Background form Write a check and make the participant sign it. Give him/her the money Before leaving the lab Fill out the labs excel document Desktop > Keys > participantkeys.xls. When complete, eject it with the butto n on the left. Copy all 4 data items to the flash drive Turn ON the network. Make sure all PCs are in sleep mode and that I have completed the sign in page, the experiment log sheet and the labs excel document
263 APPENDIX F P ICTURE S TORY S EQUENCE 1 nacer llamar esperar venir regalar poner verter tomar mirar medir corregir mentir escapar caber poder decir repetir conseguir inventar reducir andar sentir extraer segu ir RAL Tengo 25 aos. Tengo 25 aos. 83 > 75 No tengo celular
264 APPENDIX G P ICTURE S TORY S EQUENCE 2 Adis! Una sopa, por favor Cuntos aos tienes? LUISA
265 APPENDIX H SAMPLE ACTIVITIES FR OM THE DELE PROFICIE NCY EXAMS NIVEL INICIAL Interpretacin de Textos Orales. Parte 1. A continuacin escuchar 10 dilogos br eves entre dos personas. La persona que responde lo hace de tres formas distintas pero solamente una es adecuada. Oir cada dilogo dos veces. Despus de la segunda audicin, marque la opcin correcta en la hoja. Dilogo nmero 1: HOMBRE: MUJER: a) El mes que viene voy. b) Gire la segunda a la derecha. c) Pues no lo encuentro. Interpretacin de Textos Orales. Parte 2. A continuacin escuchar 7 dilogos muy breves. Se le har una pregunta acerca de cada uno de ellos. Escoja una de las tres respuestas que se le proponen. Oir cada dilogo dos veces. Despus de la segunda audicin, marque la opcin correcta en la hoja. Texto 1: 11. Qu objeto lleva esta pareja a la playa? Interpretacin de Textos Orales. Parte 3. A continuacin ust ed va a or una noticia radiofnica. La oir dos veces. Despus de la segunda audicin, marque la opcin correcta en la hoja. 18. Segn esta noticia, estos consejos del Ministerio de Sanidad son exclusivos para el verano. a) Verdadero. b) Falso. Inter pretacin de Textos Orales. Parte 4. A continuacin usted va a or una conversacin entre dos personas. La oir dos veces. Despus de la segunda audicin, marque la opcin correcta en la hoja. 21. Segn la grabacin, el hombre a) desea baarse con los d elfines. b) dispone de una semana de vacaciones. c) quiere realizar un viaje cultural y de ocio.
266 Conciencia Comunicativa. Parte 1. En qu situacin dira usted las siguientes expresiones? Marque la opcin correcta en la hoja. 1. Que cumplas muchos ms Usted __________ a un amigo. a) saluda. b) llama. c) felicita. Conciencia Comunicativa. Parte 2. A la izquierda tiene usted 10 frases. En cada frase hay en letra negrita una o dos palabras que no son adecuadas. Sustityala por alguna de las palabras de la lista que aparecen en el cuadro de la derecha. Puede utilizar esta hoja como borrador. Marque la opcin correcta en las casillas con nmeros al final de la hoja. 6. Todas las noches son conciertos en las calles de la ciudad. 7. Ustedes decs que este libro es muy interesante. 8.  a) muy b) hay c)  6. 7. 8. 9. 10. 11. 12. 13. 14. 15. Conciencia Comunicativa. Parte 3. Rellene los huecos del texto siguiente con una de las tres opciones que se le proponen. Marque la opcin correcta en la hoja. UNA CONVERSACIN TELEFNICA HOMBRE: Hola, est Ido ia ? MUJER: S, soy _____16_____. Eres Mikel, no? HOMBRE: S, no te haba _____17_____. MUJER: Es que estoy un poco resfriada. Qu queras?  Opciones: 16. a) mi b) me c) yo 17. a) reconocido b) reconociendo c) reconocer 18.  Interpretacin de Textos Escritos. Parte 1. A continuacin encontrar usted un texto y tres preguntas sobre l. Marque la opcin correcta en la hoja. EL CEBICHE Segn todos los indicios, fue en Per donde se empez a consumir el pescado y el marisco en crudo. Al principio se maceraba con una especie de naranjillas, pero con la llegada de los espaoles y de los portugueses se introdujeron los limones y el perejil, que enriquecieron esta
267 preparacin y dieron lugar al ceb iche, que, en la actualidad y gracias en gran medida al xito que tuvo entre los marineros ingleses, se consume en todo el mundo.  (282 words) (Adaptado de I Congreso de Cocina Iberoamericana. Espaa) Preguntas: 1. Segn el texto, el cebiche es un plat o de pescado que a) tiene su origen en Per. b) inventaron los espaoles. c) perfeccionaron los ingleses. Interpretacin de Textos Escritos. Parte 2. A continuacin le presentamos una serie de textos breves. Conteste a las preguntas que se le hacen. M arque la opcin correcta en la hoja. Texto A: 4. Segn el texto, puede ir a clase de guitarra cualquier da de la semana. a) Verdadero. b) Falso. Interpretacin de Textos Escritos. Parte 3. A continuacin encontrar usted un texto y diez preguntas s obre l. Marque la opcin correcta en la hoja. (text selection)
268 Preguntas: 11. Hay una tienda de msica que vende instrumentos a) de percusin. b) para especialistas. c) del siglo XX. NIVEL INTERMEDIO Comprensin auditiva. Usted va a or cuatro textos. Oir cada uno de ellos dos veces. Al final de la segunda audicin, dispondr de tiempo para contestar a las preguntas que se le formulan. Hay dos modalidades de pregunta: Primer tipo: a) Verdadero. b) Falso. Segundo tipo: Seleccin de una resp uesta entre tres opciones: a) b) c) Marque la opcin correcta en la hoja. Texto 1: EXPOSICIN DE CMICS A continuacin escuchar una noticia sobre una exposicin de comics. (Adaptado de Radio Espectador Uruguay) Preguntas: 1. Segn la audicin los dibujos de Santiago Gonzlez se vieron en un programa de televisin. a) Verdadero. b) Falso. Texto 2: EL AJEDREZ EN LA ESCUELA A continuacin una noticia sobre el ajedrez en la escuela. (Adaptado de Radio Rebelde Cuba) Preguntas: 4. Segn la aud icin, la prctica del ajedrez en las escuelas cubanas a) es una novedad en este curso escolar. b) se imparte en todos los niveles educativos. c) es una asignatura escolar.
269 Gramtica y Vocabulario. Parte 1. Complete el siguiente texto eligiendo para cada uno de los huecos una de las tres opciones que se le ofrecen. Puede utilizar esta pgina como borrador, si lo estima conveniente. Marque la opcin elegida en la siguiente hoja. GUAS DE VIAJES Pocas personas se aventuran a visitar por primera vez u n lugar sin ir acompaadas de una buena gua de viajes. _____1_____ con el pasaporte, la moneda extranjera o el billete de avin, este producto _____2_____ en parte fundamental del equipaje y ha pasado de ser un artculo minoritario a _____3_____ las posic iones ms altas en las listas de los libros ms _____4_____.  (Adaptado de Consumer Espaa) Opciones: 1. a) Adems b) Al lado c) Junto 2. a) se ha hecho b) se ha vuelto c) se ha convertido 3. a) estar b) ocupar c) ser 4. a) vendidos b) ventas c) venden 5.  Gramtica y Vocabulario. Parte 2. En cada una de las frases siguientes se ha marcado con letra negrita un fragmento. Elija, de entre las tres opciones de respuesta, aqulla que tenga un significado equivalente al del fragment o marcado. Por ejemplo: No he hablado todava con Javier porque el telfono est comunicando a) est estropeado b) no da seal c) est ocupado La respuesta correcta es la C Marque la opcin correcta en la hoja. 21. Qu te parece Dolores? Es muy graciosa a) amable b) agradecida c) divertida Gramtica y Vocabulario. Parte 3. Complete las frases siguientes con el trmino adecuado de los dos o cuatro que se ofrecen. Marque la opcin correcta en la hoja. 31. Dnde __________ la fiesta? En casa de Patric ia a las once. a) es b) est
270 43. Qu piensas __________ Yolanda? Que es encantadora. a) en b) a c) de d) con Comprensin de lectura. A continuacin encontrar usted cuatro textos y una serie de preguntas relativas a cada uno de ellos. Hay dos modalidades de pregunta: Pr imer tipo: a) Verdadero. b) Falso. Segundo tipo: Seleccin de una respuesta entre tres opciones: a) b) c) Marque la opcin correcta en la hoja. TEXTO 1: PROGRAMA APRENDICES El Programa Aprendices, administrado por el Servicio de Capacitacin de Chile, es un pro yecto que tiene el objetivo de fomentar y apoyar la contratacin de jvenes para que aprendan un oficio. En concreto, ofrece una bonificacin mensual por cada aprendiz contratado durante los primeros doce meses de vigencia del contrato. Este beneficio tien e como finalidad financiar los costos de formacin de los aprendices en la empresa.  (372 words) (Adaptado de El Mercurio.com Chile) Preguntas: 1. Segn el texto, el Programa Aprendices ofrece a los jvenes empleados una bonificacin mensual. a) Ver dadero. b) Falso. TEXTO 2: FOTOGRAFAS DE ESCRITORES COLOMBIANOS Ayer fue inaugurada en Madrid la exposicin fotogrfica Cuarenta escritores colombianos. En esta exposicin, el fotgrafo colombiano Jorge Mario Mnera recoge los rostros de los escritores del momento, ejemplos de la diversidad y el colorido cultural de Colombia. El artista reconoce que lleva toda la vida preparndose para fotografiar a escritores como Hctor Abad, Laura Restrepo, Antonio Caballero, scar Collazos o Fernando Vallejo, entre otros, puesto que conoce sus obras desde nio, hecho que le ha permitido acercarse a ellos con conocimiento y con el deseo de plasmar la esencia de cada uno en una imagen en blanco y negro.  (391 words) (Adaptado de El Tiempo Colombia)
271 Preguntas: 4. Segn el texto, los escritores retratados por Mnera a) son contemporneos. b) son famosos. c) son conocidos suyos. NIVEL SUPERIOR Comprensin Auditiva. Usted va a or cuatro textos. Oir cada uno de ellos dos veces. Al final de la segunda audicin de cada uno de los textos, dispondr de tiempo para contestar a las preguntas que se le formulan. Hay dos modalidades de pregunta: Primer tipo: a) Verdadero. b) Falso. Segundo tipo: Seleccin de una respuesta entre tres opciones: a) b) c) Marque la opcin correc ta en la hoja. Texto 1: HISTORIA DEL ELEVADOR A continuacin escuchar una noticia sobre los elevadores. (Adaptado de Radio Frmula. Venezuela) Preguntas: 1. Segn la grabacin, el seor Graves Otis tuvo un problema con los cables de su elevador en su pr esentacin pblica. a) Verdadero. b) Falso. Texto 3: TURISMO EN BUENOS AIRES A continuacin escuchar una entrevista a un secretario de Turismo. (Adaptado de Radio Continental Argentina) Preguntas: 9. Segn la grabacin, la provincia de Buenos Aires a) ha tenido desde siempre varios destinos tursticos. b) prcticamente solo recibe turismo nacional. c) cuenta con atractivos diversificados.
272 Gramtica y Vocabulario. Parte 1. Complete el siguiente texto eligiendo para cada uno de los huecos una de las tres opciones que se le ofrecen. Puede utilizar esta pgina como borrador, si lo estima conveniente. Marque la opcin elegida en la siguiente hoja. DEMASIADOS ENVOLTORIOS En los ltimos diez aos he percibido un notorio crecimiento de la produccin y el consumo de envases para alimentos y bebidas. No dudo que el unicel est viviendo su poca dorada. Todava _____1_____ unos aos un atole lo servan en un jarro y el tamal lo daban en una hoja de papel reciclado (el jarro se lavaba y el papel era consum ido presto _____2_____ la tierra). Hoy da prcticamente cualquier bebida la sirven en unicel; y no solo eso, _____3_____ que lo adornan una tapa y un popote agregados, de los que bien se puede prescindir, puesto que si el vaso es nuevo y _____4_____ ha to cado su borde y se va a desechar, seguramente, qu caso _____5_____ usar un popote? Ms basura.  (Adaptado de www.floresdenieve.cepe.unam.mx Mxico) Opciones: 1. a) hay b) hacen c) hace 2. a) para b) por c) sobre 3. a) pero b) sino c) pues 4. a) nadie b) alguien c) ningn 5. a) tuviera b) tenga c) tiene 6.  Gramtica y Vocabulario. Parte 2. En cada una de las frases siguientes se ha marcado con letra negrita un fragmento. E lija, de entre las tres opciones de respuesta, aqulla que tenga un significado equivalente al del fragmento marcado. Por ejemplo: Marta no ha venido: se encuentra un poco indispuesta pero se le pasar. a) est ocupada b) est algo enferma c) se ha enfadado La respuesta correcta es B Marque la opcin correcta en la hoja. 21. No gana un gran sueldo pero se las arregla para llegar a fin de mes. a) ahorra b) hace lo posible c) hace chapuzas Gramtica y Vocabulario. Parte 3. Complete las frases siguientes con el trmino adecuado de los dos o cuatro que se le ofrecen. Marque la opcin correcta en la hoja.
273 36. Si insisti en que __________ es porque os necesitaba de verdad. a) fuisteis b) fuerais 45. Te recomiendo que te busques otra forma de divertirte, ya que nad a __________ hacemos te gusta. a) con lo que b) de lo cual c) lo que d) de lo que Gramtica y Vocabulario. Parte 4. A continuacin le presentamos dos textos. En ellos, debe usted detectar un total de cinco errores. Estos errores se han distribuido al azar, de manera que puede haber, por ejemplo, 4 en el primer texto y 1 en el segundo; o 2 en el primero y 3 en el Segundo. Cada error corresponde a uno de los segmentos en los que se han dividido los dos textos. As, por ejemplo: FRASE: La dije que viniera a verme esta tarde. La dije que viniera a verme esta tarde 1 2 3 4 5 6 7 8 El error estaba en el primer segmento de la frase, por lo que debe marcar el nmero 1: La dije que viniera a verme esta tarde 1 2 3 4 5 6 7 8 Marque los nmeros correspondientes a cada uno de los errores en la hoja. Texto 1: Si usted desea negociar algn asunto dentro de su empresa es necesario elegir el momento adecuado para hacer la peticin. Si una empresa atraviesa por malos momentos, probablemente sus peticiones no lleguen en buen puerto. Otro buen consejo es lo de ser sincero consigo mismo. Si va a solicitar a sus superiores un aumento de sueldo, asegrese de que ellos saben exactamente cules han sido sus logros por la compaa, su implicacin hacia el xito y mejora de la empresa. Comprensin de Lectura. Parte 1. En la parte 1 encontrar usted tres textos y una serie de preguntas relativas a cada uno de ellos. Seleccione la respuesta correcta entre las tres opciones que se le ofrecen: a) b) c) En la parte 2 encontrar una entrevis ta en la que debe relacionar cada pregunta de la columna A con su respuesta de la columna B. Marque la opcin correcta en la hoja.
274 TEXTO 1: LA REVOLUCIN DE LAS FLORES Hasta hace menos de 20 aos, las rosas colombianas, con sus clsicas variedades rojas, eran la envidia en las ferias internacionales. Y pese a que siguen siendo reconocidas, los floricultores ecuatorianos se estn haciendo notar despus de desarrollar variedades de rosas que estn acaparando todas las miradas. Una de las principales produc toras ecuatorianas de flores logr desarrollar nuevas variedades que le permitieron obtener especies propias de su pas y que, a manera de comercializacin, resultan atractivas y exclusivas para grandes mercados como el norteamericano y el europeo. De est a forma, y ya convertida en una propiedad multinacional, le est dando la vuelta al negocio transformndola en una industria en la que floricultores locales compran el derecho a cultivar variedades a hibridadores especializados en pases como Holanda, Fran cia o Alemania, que les cobran regalas que pueden llegar a representar cerca del 10% de los costos operativos de las empresas. Los cruces de esta empresa han dado resultado y, con alrededor de siete nuevas especies, ha contribuido a que Ecuador se ubique en los primeros lugares entre los productores y exportadores de flores en el mundo. Su estrategia est tan bien aceptada que les est vendiendo algunas variedades a los productores colombianos.  (638 words) (Adaptado de El espectador Colombia) Pregu ntas: 1. Segn el texto, los floricultores ecuatorianos a) han conseguido especies de rosas autctonas. b) llevan mucho tiempo siendo el centro de atencin de las ferias internacionales. c) compran especies en diferentes pases de Europa. Comprensin de Lectura. Parte 2. A continuacin encontrar una entrevista realizada a un alpinista ecuatoriano. Aqu se le ofrecen en la COLUMNA A, por orden, las intervenciones del entrevistador, y en la COLUMNA B, las respuestas que dio el enfrentado. Lo que usted debe hacer es relacionar cada pregunta de la COLUMNA A con su respuesta de la COLUMNA B. As, por ejemplo al nmero 7 de la COLUMNA A, le corresponde la letra D de la columna B. Utilice el recuadro al final de la entrevista para escribir sus respuestas. COLUMNA A COLUMNA B 16  A Cuando uno recin inicia es la parte ms difcil de sortear. En el 2005, la Concentracin Deportiva de Pichincha me nombr el mejor andinista del ao; eso me abri muchas puertas con empresas importantes del pas que confiar on en mi proyecto. En un inicio tuve que costearme solo, pero luego me dieron la ayuda necesaria.
275 7 Cmo fue esa ascensin mxima al Everest? B La formacin como andinista se adquiere con la experiencia. Es un proceso de formacin, el cuerpo tiene que ac ostumbrarse al fro, se necesita tener espritu y tenacidad, ya que arriba las situaciones son adversas. Todo esto es un proceso de formacin, es un trabajo que se da con el tiempo, en el que uno tiene que educar el cuerpo, la mente y el espritu. 810  C J  1. 2. 3. 4. 5. 6. 7. 8. 9. 10. D
276 APPENDIX I INFORMED CONSENT FOR M Protocol Title: The role of age of initial exposure and proficiency in the mental representation, access, and cognitive strategies for Spanish inflectional m orphology by foreign language learners. Please read this consent document carefully before you decide to participate in this study. Purpose of the research study: This study investigates the mental representation, access, and production of Spanish inflec tional morphology by adult beginner, intermediate, and advanced learners of Spanish as a foreign language in a formal instruction context. What you will be asked to do in the study: If you agree to participate in this research project, you will be asked to attend to two data collection sessions. The first session will take place at the Language and Cognition Lab located in Turlington Hall B 127. After reading and signing this consent form, you will be asked to sit in front of a PC screen, wearing a set o f headphones with a microphone attached to it. First, you will be asked to read some Spanish words presented over the screen and say to the microphone if it was masculine or femenine (for nouns), or produce the past form (for verbs). This activity will las t approximately 1215 minutes. Second, you will be asked to listen to some Spanish words presented over the headphones. Right after the word has been presented, a word will appear on the computer screen and you will be asked to press one of two bottoms as fast as possible to decide if both words were the same or not. This activity will last approximately 12 15 minutes. Third, after sitting on a chair in front of the principal investigator, you will be asked to pick a picture story sequence from several ones and you will be asked to narrate it to the researcher in Spanish using the verbs below each drawing. Your voice will be recorded with a digital recorder. You will have about 6 8 minutes to complete the narration. Fourth, you will be asked to select a new picture story sequence from several ones and you will be asked to narrate it to the researcher in Spanish, this time using the verbs of your choice. Again, your voice will be recorded digitally and you will have about 68 minutes to complete the narration. At the end of the session you will be asked to fill out a language background information form. The second data collection session will take place on campus at the University of Florida. You will be asked to complete some exercises in Spanish similar to the ones used at the beginning and intermediate Spanish courses at UF such as: reading comprehension activities, listening comprehension activities, and vocabulary and grammar activities. This data collection session will last about 60 minutes. Time Required: Two sessions of about 60 minutes will be required to complete the study. Risks, Benefits, and Compensation: Theres no risk to you. Theres no direct benefit for you for participation. If you agree to participate and complete the sessions accor ding to the instructions given, you will receive a $10 cash bonus per session for a total of $20 in cash. Confidentiality: Your identity will be kept confidential to the extent provided by law. All materials collected from you will be identified by a code number. All information including results of your performance will be kept in a locked file which only the investigator (Juan P.
277 Rodrguez) and his dissertation committee members (Dr. Theresa A. Antes, Dr. Joaquim Camps, Dr. Gillian Lord, and Dr. Wind Cow les) will have access to. Your name will not be used on any written or verbal report when presenting the results of the study. The audio recordings will not be attached to your name in any way in order to ensure confidentiality. Your name will not be used in any report. Voluntary Participation and right to withdraw from the study: Participation in this study is completely voluntary. There is no penalty for not participating. You have the right to withdraw from the study at anytime without consequence. Wh om to contact if you have questions about the study: Principal Investigator: Juan P. Rodrguez. Graduate Student, Program in Hispanic Linguistics. Department of Romance Languages and Literatures. P.O. Box 117405. 170 Dauer Hall, University of Florida at Gainesville. Gainesville, FL 326117405. 3929222, FAX (352) 3925679. firstname.lastname@example.org, email@example.com Supervisor: Theresa A. Antes, Ph.D., Associate Professor in French and Linguistics. Department of Romance Languages and Literatures. P.O. Bo x 117405. 170 Dauer Hall. University of Florida at Gainesville. Gainesville, FL 326117405. (352) 3922016 ext. 236, FAX (352) 3925679. firstname.lastname@example.org Whom to contact about your rights as a research participant in the study: UFIRB office, P.O. Box 11 2250, University of Florida, Gainesville, FL 326112250, Ph. (352) 3920433. email@example.com Agreement: I have read the entire procedure described above. I voluntarily agree to participate in the study, and I also consent to the audio recordings as outlined above. I have received a copy of this description. _______________ ___________ ________________ ___________ Participant Date Principal Investigator Date
278 APPENDIX J TAXONOMY USED FOR CODING ERROR TYPES Error 0: None. The participant produced t he verb form with accuracy and very fast. For example: dijo (she/he said) for the test item decir (to say). Error 1: Overregularization. Participants produced an irregular verb form incorrectly because they treated it as a regular verb form by attachi ng a regular suffix to the verb root of the infinitive. For example: reduci ( she /he reduced, nonexistent verb form) for the test item reduc i r (to reduce) instead of the corresponding irregular verb form redujo. Error 2: No vowel change but correct s uffix. For vowel stem change verbs which required a vowel change but were produced without that change. For example: senti ( she /he felt, incorrect verb form) for the test item sentir ( to feel ) whose correct preterite form required the vowel stem chan ge from /e/ to /i/, i.e. sinti. Err or 3: Nonfinite form or a noun. Used when the participant produced a noun with a similar stem to the test item, or when the response was a nonfinite form such as an infinitive or a past participle. For example: para re galar ( in order to give a present ) for the test item reg alar ( to give a present ). Sometimes an infinitive was preceded by another verb in the preterite as a cognitive strategy to avoid the preterite form of the test item while including the past refere nce required by the demands of the task. For example: decidi medir ( she /he decided to measure ) for the test item medir ( to measure ) whose preterite form undergoes a vowel stem change: midi ( she /he measured ). Err or 4: Hypercorrection. Right after t he participant provided the correct response, she/he changed it for an incorrect one as her/his final answer. For example: puse ( I put ) for the test item poner ( to put ), right after saying puso ( she/he put ). Err or 5: Buying time. Three different ob served behaviors were included under this cognitive strategy: (a) lengthened syllable, usually the first one, (b) small pause before suffix addition, and (c) repetition of the response. Participants used these strategies to get extra time before making sur e that they accessed the correct form while making use of their monitor. All strategies led to correct responses with this code For example: puuuuuuudo ( she/he was able to ) for the test item poder ( to be able to ). Err or 6: Incorrect P N marker The r esponse was in the preterite but the participant used the incorrect Person / Number suffix, usually the first person singular suffix instead of the third person singular. For example: llam ( I called ) for the test item llamar ( to call ), instead of llam (she /he called ). Error 7: No response. Sometimes participants skipped some test items without even trying. Instead of risking an error, they simply moved on to the next item. Error 8: Incorrect stress. The verb form was the elicited one, but the partic ipant placed the stress on a different syllable and/or segment. The response sounds strange for a native speaker but the verb form is completely correct from a morphological point of view. For example:
279 extrajo is the elicited preterite form for the verb ex traer (to extract), but some participants placed the stress on the last syllable extraj when it should be on the penultimate. Error 9: There is a combination of no vowel change and a TAM suffix from a different verb class. The elicited verb form for t he verb verter (to pour) is virti which suffers the internal vowel change e > i and takes the TAM suffix i The form produced by some participants was vert in which the vowel of the root is kept unchanged and the incorrect TAM suffix is attached to the stem Error 10: Different tense. The verb was conjugated into a tense other than the preterite tense. The response is a finite form but not in the tense they were instructed to use. For example, the present tense sigo (I continue) was provided i nstead of the appropriate preterite form sigui (s/he continued). Error 11: A combination of overregularization and buying time. Perhaps the participant was buying time to look for the appropriate link between an infinitive and a memorized past form and in the absence of correct form retrieval, decided to attach the regular suffix: *veeeeni instead of vino (s/he came). Error 12: Monitor. Right after the participant produced an oral response, and after listening to it, s/he realized the mistake and c hanged it to the correct response, as in menti which was followed by minti (s/he lied), the correct answer. Error 13: Wrong root. In this case the correct suffix was attached to the stem of the verb but the root was incorrect. It does not illustrate a failed trial at providing a vowel stem change or anything similar. It is usually confusion in the syllable structure of the root, or even a mispronunciation not monitored by the participant. For example, instead of providing escap ( s /he escaped) one participant said *esquep Error 14: A combination of monitor and incorrect PN marker. This error type is very similar to behavior 12, in that the participant realized her/his mistake and tried to fix it with a new verb form. However, this time the new response is still wrong. The root was changed to the appropriate preterite root but the PN suffix is different from the one elicited because an incorrect PN suffix was attached to the root. For example, after saying pongo (I put) the participant produced the correct preterite root in puse (I put) but with the incorrect PN suffix. The target form was puso (s/he put). Error 15: A combination of buying time and incorrect PN marker. The participant provided a verb form close to the elicited one except fo r the PN suffix. S/he also bought some time before producing her/his response as the final version: *puuuude (I could) instead of pudo. Error 16: TAM suffix of another verb class. The elicited TAM suffix for regular verbs is (for verb class I) and i (for verb classes II and III). Sometimes the participant attached the incorrect TAM suffix, even to irregular roots, as in : *vini (s/he came) instead of vino
280 The interesting case in this example is that i would have been the correct TAM suffix if t he elicited form were regular. This example is not considered overregularization because the participant provided the correct irregular root [ vin ]. Error 17: Nonexistent form or a verb form from another verb. The verb form provided by the participant doe s not exist in Spanish and there is no way to analyze its internal structure to reveal any behavior. For example: hugo (nonexistent word) as the preterite for hacer ( to do ) instead of the elicited verb form hizo ( s/he did ). Sometimes the verb form pr ovided is nonexistent for the test item but is a verb form from another verb, like se puso (s/he put on) as the preterite verb form for poder (to be able to) whose correct preterite form is pudo. Error 18: Combination of monitor and overregularizatio n. The behavior here is that the first response provided was monitored by the participant and, although the next response was somehow improved, the remaining error is overregularization. For example after providing the verb form *andi as the preterite fo r andar ( to walk ), it was ch anged so that the TAM suffix matched the verb class of the verb (i.e., from i to ). However, the new response and is still incorrect because the elicited verb form is irregular : anduvo ( s/he walked ). Error 19: A combi nation of buying time and wrong stress. The only difference with error 8 is that this time the participant took some time before producing the response, usually by lengthening a syllable of the verb, as in *repiiiiiiiito for the elicited verb form repiti (s/he repeated) Again, the verb form was the elicited one, but the participant placed the stress on a different syllable and/or segment. Error 20: A combination of no vowel change and wrong stress. For example: corrego instead of corrigi (s/he cor rected). Error 21: A combination of overregularization and wrong stress. For example: reduco instead of redujo ( s /he reduced). Error 22: A combination of no vowel change and buying time, as in meeeedi instead of midi (s/he measured).
281 APPENDI X K LIST OF TEST ITEMS A ND THEIR ELICITED PR ETERITE FORM Regular Verbs High Frequency Low Frequency Infinitive Target form Infinitive Target form Abrir Abri ( s/he opened ) Besar Bes (s /he kissed ) Empezar Empez ( s/he started ) Cantar Cant (s/ he sang ) Entrar Entr ( s/he entered ) Escapar Escap (s/he escaped ) Llamar Llam ( s/he phoned ) Esperar Esper (s/he waited ) Llevar Llev ( s/he carried ) Inventar Invent (s/he invented ) Mirar Mir ( s/he looked ) Olvidar Olvid (s/he forg ot ) Nacer Naci ( s/he was born ) Pagar Pag (s/he paid ) Preguntar Pregunt ( s/he asked ) Regalar Regal ( s/he gave away ) Salir Sali ( s/he went out ) Vender Vendi (s/he sold ) Tomar Tom ( s/he took ) Viajar Viaj (s/he travelled ) Irr egula r Verbs High Frequency Low Frequency Infinitive Target form Infinitive Target form Decir Dijo (s /he said ) Andar Anduvo ( s/he walked ) Estar Estuvo ( s/he was ) Atraer Atrajo ( s/he attracted ) Haber Hubo ( there was ) Caber Cupo ( s/he fit ) Hacer Hizo ( s/he did ) Conducir Condujo ( s/he drove ) Ir Fue ( s/he went ) Contraer Contrajo ( s/he contracted ) Poder Pudo ( s/he could ) Extraer Extrajo ( s/he extracted ) Poner Puso ( s/he put ) Introducir Introdujo ( s/he introduced ) Producir Produjo ( s/he produced ) Reducir Redujo ( s/he reduced ) Tener Tuvo ( s/he had ) Traducir Tradujo ( s/he translated ) Venir Vino ( s/he came ) Traer Trajo ( s/he brought ) Stem Change Verbs High Frequency Low Frequency Infinitive Target form Infinitive Tar get form Advertir Advirti ( s/he warned ) Corregir Corrigi (s/he corrected) Conseguir Consigui ( s/he obtained ) Despedirse Se despidi (s/he said bye) Convertir Convirti ( s/he converted ) Divertirse Se divirti (s/he enjoyed) Morir Muri ( s/he died ) Dormir Durmi (s/he slept) Pedir Pidi ( s/he asked ) Herir Hiri (s/he hurt) Repetir Repiti ( s/he repeated ) Medir Midi (s/he measured) Seguir Sigui ( s/he continued ) Mentir Minti (s/he lied) Sentir Sinti ( s/he felt ) Rers e Se ri (s/he laughed) Servir Sirvi ( s/he served ) Verter Virti (s/he poured) Sonrer Sonri ( s/he smiled ) Vestirse Se visti (s/he dressed)
282 LIST OF REFERENCES Aguirre, C., & Dressler, W. (2006). On Spanish verb inflection. Folia Linguis tica, 40(12) 75 96. Albright, A., & Hayes, B. (2003). Rules vs. analogy in English past tenses: A computational/experimental study. Cognition, 90(2) 119 161. Almela, R. (2005). Frecuencias del espaol: Diccionario y estudios lxicos y morfolgicos Ma drid: Universitas. Altmann, G. (1997). The ascent of Babel: An exploration of language, mind, and understanding. Oxford: Oxford University Press. Andersen, R. (1991). Developmental sequences: The emergence of aspect marking in second language acquisitio n. In T. Huebner, & C. Ferguson (Eds.), Crosscurrents in second language acquisition and linguistic theories (pp. 305324). Amsterdam: John Benjamins Publishing Company. Anderson, J. (1983). The architecture of cognition Cambridge, MA: Harvard Universit y Press. Bauer, L. (2003). Introducing linguistic morphology Washington, D.C.: Georgetown University Press. Beck, M. (1997). Regular verbs, past tense and frequency: tracking down a potential source of NS/NNS competence differences. Second Language Re search, 13(2) 93115. Birdsong, D. (1999). Introduction: Whys and why nots of the critical period hypothesis for second language acquisition. In D. Birdsong (Ed.), Second language acquisition and the critical period hypothesis (pp. 122). Mahwah, NJ: Law rence Erlbaum Associates. Birdsong, D. (2004). Second language acquisition and ultimate attainment. In A. Davies, & C. Elder (Eds.), The handbook of applied linguistics (pp. 82105). Malden, MA: Blackwell. Birdsong, D., & Flege, J. (2001). Regular irre gular dissociations in the acquisition of English as a second language. In A. Do, L. Domnguez, & A. Johansen (Eds.), Proceedings of the 25th annual Boston University conference on language development (Vol. 1, pp. 123132). Somerville, MA: Cascadilla Pres s. Blackwell, A., & Broeder, P. (1992). Interference and facilitation in SLA: A connectionist perspective. Presented at the seminar Parallel Distributed Processing and Natural Language Processing, San Diego, CA. Brovetto, C., & Ullman, M. (2001). First vs. second language: A differential reliance on grammatical computations and lexical memory Presented at the 14th Annual CUNY Conference on Human Sentence Processing, Philadelphia, PA.
283 Brovetto, C., & Ullman, M. (2005). The mental representation and pro cessing of Spanish verbal morphology. In D. Eddington (Ed.), Selected proceedings of the 7th Hispanic linguistics symposium (pp. 98105). Somerville, MA: Cascadilla Press. Bruhn de Garavito, J. (2003). Knowledge of tense in the mental representation of L2 Spanish. In P. Kempchinsky, & C. Pieros (Eds.), Theory, practice, and acquisition. Papers from the 6th Hispanic linguistics symposium and the 5th conference on the acquisition of Spanish and Portuguese (pp. 284298). Somerville, MA: Cascadilla Press. B ybee, J. (1985). Morphology. A study of the relation between meaning and form Amsterdam: John Benjamins Publishing Company. Bybee, J. (1995). Regular morphology and the lexicon. Language and Cognitive Processes, 10(5) 425455. Bybee, J., & Moder, C. (1983). Morphological classes as natural categories. Language, 59(2) 251270. Camps, J. (2005). The emergence of the imperfect in Spanish as a foreign language: The association between imperfective morphology and state verbs. International Review of Applied Linguistics in Language Teaching, 43(3), 163192. Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press. Clahsen, H. (2006). Dual mechanism morphology. In K. Brown (Ed.), Encyclopedia of language and linguistics (Vol. 4, pp. 15). Elsevier: Oxford. Clahsen, H., Aveledo, F., & Roca, I. (2002). The development of regular and irregular verb inflection in Spanish child language. Journal of Child Language, 29(3) 591 622. Davies, M. (2002present). Corpus del espa ol. Retrieved J anuary, 2008, from http :// www.corpusdelespanol.org. DeKeyser, R. (1997). Beyond explicit rule learning: Automatizing second language morphosyntax. Studies in Second Language Acquisition, 19(2) 195221. DeK eyser, R. (2000). The robustness of critical period effects in second language acquisition. Studies in Second Language Acquisition, 22(4) 493 533. Doughty, C., & Long, M. (2003). The scope of inquiry and goals of SLA. In C. Doughty, & M. Long (Eds.), The handbook of second language acquisition (pp. 316). Malden, MA: Blackwell. Ellis, N. (2003). Constructions, chunking, and connectionism: The emergence of second language structure. In C. Doughty, & M. Long (Eds.), The handbook of second language acquisit ion (pp. 63103). Malden, MA: Blackwell.
284 Ellis, N., & Schmidt, R. (1998). Rules or associations in the acquisition of morphology? The frequency by regularity interaction in human and PDP learning of morphosyntax. Language and Cognitive Processes, 13(23) 307 336. Ellis, R. (1985). Sources of variability in interlanguage. Applied Linguistics, 6(2) 118131. Ellis, R. (1987). Interlanguage variability in narrative discourse: Style shifting in the use of the past tense. Studies in Second Language Acquisit ion, 9(1) 1 20. Elman, J. (1990). Finding structure in time. Cognitive Science, 14 (2) 179 211. Elman, J. (1991). Distributed representations, simple recurrent networks, and grammatical structure. Machine Learning, 7(2 3) 195 225. Elman, J. (1998). G eneralization, simple recurrent networks, and the emergence of structure. In M Gernsbacher, & S. Derry (Eds.), Proceedings of the 20th annual conference of the cognitive science society. Mahway, NJ : Lawrence Erlbaum Associates Elman, J. (2004). An alter native view of the mental lexicon. Trends in Cognitive Sciences, 8(7) 301306. Flege, J., Yeni Komshian, G., & Liu, S. (1999). Age constraints on secondlanguage acquisition. Journal of Memory and Language, 41(1) 78 104. Fodor, J. (1983). The modularit y of mind: An essay on faculty psychology Cambridge, MA: MIT Press. Gasser, M. (1990). Connectionism and universals of second language acquisition. Studies in Second Language Acquisition, 12(2) 179199. Gordon, P., & Alegre, M. (1999). Is there a dual system for regular inflections? Brain and Language, 68(12) 212217. Hahne, A. (2001). Whats different in second language processing? Evidence from event related brain potentials. Journal of Psycholinguistic Research, 30(3) 251266. Hahne, A., & Frie derici, A. (2001). Processing a second language: Late learners comprehension mechanisms as revealed by event related brain potentials. Bilingualism: Language and Cognition, 4(2) 123141. Hahne, A., Mueller, J., & Clahsen, H. (2006). Morphological proces sing in a second language: Behavioral and event related brain potential evidence for storage and decomposition. Journal of Cognitive Neuroscience, 18(1) 121134.
285 Housen, A. (2002). The development of tense aspect in English as a second language and the v ariable influence of inherent aspect. In R. Salaberry, & Y. Shirai (Eds.), The L2 acquisition of tense aspect morphology (pp. 155197). Amsterdam: John Benjamins Publishing Company. Johnson, J., & Newport, E. (1989). Critical period effects in second language learning: The influence of maturational state on the acquisition of English as a second language. Cognitive Psychology, 21(1) 6099. Kaan, E. (2007). Event related potentials and language processing: A brief overview. Language and Linguistics Compass, 1(6) 571591. Kempe, V., & MacWhinney, B. (1998). The acquisition of case marking by adult learners of Russian and German. Studies in Second Language Acquisition, 20(4) 543587. Kuc era, H., & Francis, W. (1967). Computational analysis of present day American English. Providence: Brown University Press. Kutas, M., & Hillyard, S. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207(4427) 203205. Lalleman, J., van Santen, A., & van Heuven, V. (1997). L2 processing Dutch regular and irregular verbs. Review of Applied Linguistics, 115/116, 1 26. Lee, S., & Huang, H. (2008). Visual input enhancement and grammar learning. Studies in Second Language Acquisition, 30(3) 307331. Lenneberg, E. (1967). Biological foundations of language New York: John Wiley. Leung, Y. (2006). Verb morphology in second language versus third language acquisition: The representation of regular and irregular past participles in English Spanish and Chinese English Spanish interla nguages. EUROSLA, 6 27 56. Lightbown, P. (1985). Great expectations: Second language acquisition research and classroom teaching. Applied Linguistics, 6(2) 173189. Long, M. (1991). Focus on form: A design feature in language teaching methodology. In K De Bot, R. Gingsberg, & C. Kramsch (Eds.), Foreign language research in cross cultural perspective (pp. 39 52). Amsterdam: John Benjamins Publishing Company Matthews, P. (1997). The concise Oxford dictionary of linguistics Oxford: Oxford University Pr ess. McClelland, J., & Patterson, K. (2002). Rules or connections in past tense inflections: What does the evidence rule out? Trends in Cognitive Sciences, 6(11) 465 474.
286 McLaughlin, B. (1990). Restructuring. Applied Linguistics, 11(2) 113128. McLau ghlin, J., Osterhout, L., & Kim, A. (2004). Neural correlates of second language word learning: Minimal instruction produces rapid change. Nature Neuroscience, 7(7) 703704. Montrul, S. (2004). The acquisition of Spanish. Morphosyntactic development i n monolingual and bilingual L1 acquisition and adult L2 acquisition. Amsterdam: John Benjamins Publishing Company. Murphy, V. (2004). Dissociable systems in second language inflectional morphology. Studies in Second Language Acquisition, 26(3) 433459. Opitz, B., & Friederici, A. (2003). Interactions of the hippocampal system and the prefrontal cortex in learning languagelike rules. Neuroimage, 19(4) 17301737. Opitz, B., & Friederici, A. (2004). Brain correlates of language learning: The neuronal di ssociation of rule based versus similarity based learning. Journal of Neuroscience, 24(39) 84368440. Orsolini, M., & Marslen Wilson, W. (1997). Universals in morphological representation: Evidence from Italian. Language and Cognitive Processes, 12(1) 147. Osterhout, L., & McLaughlin, J. (2000). What brain activity can tell us about secondlanguage learning. Presented at the 13th Annual CUNY Conference on Human Sentence Processing, San Diego, CA Penfield, W., & Roberts, L. (1959). Speech and brain m echanisms New York: Athenaeum. Pienemann, M. (1987). Psychological constraints on the teachability of languages. In C. Pfaff (Ed.), First and second language acquisition processes (pp. 143 168). Cambridge, MA: Newbury House Publishers. Pinker, S. (1991) Rules of language. Science, 253(5019) 530 535. Pinker, S. (1998). Words and rules. Lingua, 106(14) 219 242. Pinker, S. (1999). Words and rules: The ingredients of language New York: Harper Perennial. Pinker, S. (2004). Clarifying the logical probl em of language acquisition. Journal of Child Language, 31(4) 949953. Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a parallel distributed processing model of language acquisition. Cognition, 28(12) 73193.
287 Pinker, S., & Prince, A. (1994). Regular and irregular morphology and the psychological status of rules of grammar. In S. Lima, L. Corrigan, & G. Iverson (Eds.), The reality of linguistic rules (pp. 321 351). Amsterdam: John Benjamins Publishing Company. Pinker, S., & Ullman, M. (2002). The past and future of the past tense. Trends in Cognitive Sciences, 6(11) 456465. Prasada, S., & Pinker, S. (1993). Generalisation of regular and irregular morphological patterns. Language and Cognitive Processes, 8(1) 1 56. Real A cademia Espa ola. Corpus de referencia del espaol actual (CREA). Retrieved January 2008, from http://corpus.rae.es/creanet.html Rodr guez, J. (in press). Acquisitional patterns of the Spanish copular verbs ser and estar : Data from L2 beginning learners in favor of the Declarative/Procedural model. Revista Espaola de Lingstica Aplicada. Rossi, S., Gugler, M., Friederici, A., & Hahne, A. (2006). The impact of proficiency on syntactic second language pr ocessing of German and Italian: Evidence from event related potentials. Journal of Cognitive Neuroscience, 18(12) 2030 2048. Rumelhart, D., & McClelland, J. (1986). On learning the past tenses of English verbs. In D. Rumelhart, & J. McClelland (Eds.), Pa rallel distributed processing: Explorations in the microstructure of cognition (pp. 216271). Cambridge, MA: MIT Press. Rschemeyer, S., Fiebach, C., Kempe, V., & Friederici, A. (2005). Processing lexical semantic and syntactic information in first and s econd language: FMRI evidence from German and Russian. Human Brain Mapping, 25(2) 266286. Sakai, K., Miura, K., Narafu, N., & Muraishi, Y. (2004). Correlated functional changes of the prefrontal cortex in twins induced by classroom education of second l anguage. Cerebral Cortex, 14(11) 12331239. Salaberry, R. (1999). The development of past tense verbal morphology in classroom L2 Spanish. Applied Linguistics, 20(2) 151178. Salaberry, R. (2000). The acquisition of English past tense in an instructio nal setting. System, 28(1) 135152. Schmidt, R. (1990). The role of consciousness in second language learning. Applied Linguistics, 11(2) 1746. Schwartz, B. (1993). On explicit and negative data effecting and affecting competence and linguistic behavi or. Studies in Second Language Acquisition 15(2) 147163.
288 Sharwood Smith, M. (1993). Input enhancement in instructed SLA: Theoretical bases. Studies in Second Language Acquisition, 15(2) 165179. Seidenberg, M., & Gonnerman, L. (2000). Explaining deriv ational morphology as the convergence of codes. Trends in Cognitive Sciences, 4(9) 353 361. Singleton, D. (1995). Introduction: A critical look at the critical period hypothesis in second language acquisition research. In D. Singleton, & Z. Lengyel (Eds. ), The age factor in second language acquisition. A critical look at the critical period hypothesis (pp. 129). Clevedon, England: Multilingual Matters. Skehan, P. (1998). A cognitive approach to language learning. New York: Oxford University Press. So anes, C., & Hawker, S. (2005). Compact Oxford English dictionary of current English (3rd ed.). Retrieved March 21, 2009, from http://www.askoxford.com/dictionaries/ compact_oed/?view=uk. Sokolik, M., & Smith, M. (1992). Assignment of gender to French nouns in primary and secondary language: A connectionist model. Second Language Research, 8(1) 3958. Squire, L. (1994). Declarative and nondeclarative memory: Multiple brain systems s upporting learning and memory. In D. Schacter, & E. Tulving (Eds.), Memory systems 1994 (pp. 203231). Cambridge, MA: MIT Press. Swain, M. (1995). Three functions of output in second language language learning. In G. Cook, & B. Seidlhofer (Eds.), Principl e and practice in applied linguistics: Studies in honour of H. G. Widdowson (pp. 125 144). Oxford: Oxford University Press. Swain, M. (1998). Focus on form through conscious reflection. In C. Doughty, & J. Williams (Eds.), Focus on form in classroom second language acquisition (pp. 6481). Cambridge: Cambridge University Press. Taraban, R., & Kempe, V. (1999). Gender processing in native (L1) and non native (L2) Russian speakers. Applied Psycholinguistics, 20(1) 119148. Tatsuno, Y., & Sakai, K. (2005). Languagerelated activations in the left prefrontal regions are differentially modulated by age, proficiency, and task demands. Journal of Neuroscience, 25(7) 16371644. Tokowicz, N., & MacWhinney, B. (2005). Implicit and explicit measures of sensitivit y to violations in second language grammar. Studies in Second Language Acquisition, 27(2) 173204. Ullman, M. (2001a). The declarative/procedural model of lexicon and grammar. Journal of Psycholinguistic Research, 30(1) 3769.
289 Ullman, M. (2001b). The n eural basis of lexicon and grammar in first and second language: The declarative/procedural model. Bilingualism: Language and Cognition, 4(1) 105122. Ullman, M. (2001c). A neurocognitive perspective on language: The declarative/procedural model. Nature Reviews. Neuroscience, 2(10) 717726. Ullman, M. (2004). Contributions of memory circuits to language: The declarative/procedural model. Cognition, 92(1 2) 231270. Ullman, M. (2005). A cognitive neuroscience perspective on second language acquisition: The declarative/procedural model. In C. Sanz (Ed.), Mind and context in adult second language acquisition: Methods, theory, and practice (pp. 141 178). Washington, DC: Georgetown University Press. Ullman, M., Corkin, S., Coppola, M., Hickok, G., Growdon, J., Koroshetz, W., & Pinker, S. (1997). A neural dissociation within language: Evidence that the mental dictionary is part of declarative memory, and that grammatical rules are processed by the procedural system. Journal of Cognitive Neuroscience, 9(2) 2 66276. VanPatten, B. (1996). Input processing and grammar instruction. New York: Ablex. Walenski, M., & Ullman, M. (2005). The science of language. The Linguistic Review, 22(24) 327346. Weber Fox, C., & Neville, H. (1996). Maturational constraints o n fu n ctional specializations for language processing: ERP and behavioral evidence in bilingual speakers. Journal of Cognitive Neuroscience, 8(3) 231256. Wolfram, W. (1985). Variability in tense marking: A case for the obvious. Language Learning, 35(2) 229253. Yaden, B. (2003). Mental representations of Spanish morphology: Rules or analogy? In P. Kempchinsky, & C. Pieros (Eds.), Theory, practice, and acquisition. Papers from the 6th Hispanic linguistics symposium and the 5th conference on the acquisit ion of Spanish and Portuguese (pp. 299 312). Somerville, MA: Cascadilla Press. Yaden, B. (2004). Mental representations of Spanish morphology: Rules or analogy? Dissertation Abstracts International, A: The Humanities and Social Sciences, 64(11), 4033A. Yaden, B. (2007). The processing and representation of verbal inflection: Data from L1 and L2 Spanish. Hispania, 90(4) 795808.
290 BIOGRAPHICAL SKETCH Juan Pablo Rodr guez Prieto was born in Palencia, Spain. His passion for the English language was instil led by one of his teachers at the Official Sc hool of Languages in Palencia. In 2002, he earned a Bachelor of Arts in English Philology from the University of Valladolid. During his undergraduate studies, he spent one summer in Galway, Ire land. In 2004, he obtained a Master of A rts in Teaching English to Speakers of Other Languages from West Virginia University, where he taught beginning and intermediate Spanish language courses. His teaching experience in Spanish helped him decide to pursue a Doctor of Phil osophy ( Ph.D. ) in Hispanic l inguistics at the University of Florida with an emphasis on second language acquisition. He taught a variety of Spanish language courses at the University of Florida, including Introduction to Hispanic Linguistics and the F oreig n L anguage A cross the C urriculum course entitled Spain and the European Union and he also worked for two years as assistant to the coordinators of the beginning Spanish program. The McLaughlin Dissertation Fellowship awarded by the University of Florida h elped him for the data collection of the present study. Juan Pablo has been recognized for his excellence in teaching and for his innovative teaching materials. In 2006 he obtained the Award for Cooperative Leadership in Teaching (first y ear Spanish ) at t he University of Florida. I n 2008, he was finalist in the Third Premio Cristbal de Villaln from the University of Valladolid, and he was also awarded a Graduate Student Teaching Award for the 20082009 academic year at the University of Florida. He has published in Revista Espaola de Lingustica Revista Espaola de Lingstica Aplicada Revista Iberoamericana de Educacin and MarcoELE : Revista de Didctica ELE He received his Ph.D. from the University of Florida in the summer of 2009. H e will work as assistant professor of Spanish at Ball State University in Muncie, Indiana. His main areas of interest are Spanish second language acquisition, pedagogy and technology education.