Citation
Testing Acquisition of L2 Hindi Voicing and Aspiration Contrasts

Material Information

Title:
Testing Acquisition of L2 Hindi Voicing and Aspiration Contrasts A Comparison from Multiple L1s
Creator:
Aggarwal, Ashima
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (137 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Linguistics
Committee Chair:
WILTSHIRE,CAROLINE R
Committee Co-Chair:
WAYLAND,RATREE
Committee Members:
ROTHMAN,JASON LEONARDO
NARAYANAN,VASUDHA R
Graduation Date:
8/9/2014

Subjects

Subjects / Keywords:
Control groups ( jstor )
Critical periods ( jstor )
Foreign language learning ( jstor )
Language ( jstor )
Learning ( jstor )
Optimality theory ( jstor )
Phonetics ( jstor )
Phonology ( jstor )
Second language acquisition ( jstor )
Syllables ( jstor )
Linguistics -- Dissertations, Academic -- UF
phonology
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Linguistics thesis, Ph.D.

Notes

Abstract:
The available literature on second language phonology provides many unresolved claims about the acquisition of non-native phonology in adulthood. The present study successfully contributes to resolving the problem by providing empirical data that is both generalizable and verifiable. This study examines the acquisition of novel phonological categories by speakers of English and three Indian languages Bodo, Mizo and Tamil. The target phonological categories were present in Hindi language. The speakers of English, Tamil, Bodo and Mizo underwent training that were preceded and followed by pretests and posttests respectively. The training was aimed at helping these speakers perceive the voicing and laryngeal contrasts which are not present in Bodo, Mizo, Tamil and English. The speakers participated in an AX perception experiment that presented CV syllables from Hindi language that differ in voicing, aspiration or both. During training they has to respond to a given pair of syllables as being perceived as same or different. The participant responses were followed by feedback displayed on their laptop screen. The training lasted four sessions followed by a retention cum generalization test that was conducted two weeks after the last session. Following the experiment, the results were analyzed using statistical software SPSS and Microsoft Excel 2007. The mean percent correct Identification and the mean reaction times of all language groups were calculated. The results indicated that all groups take significantly lesser time during the posttest than for the pretest. The experimental groups Bodo, English and Mizo also show greater percentage of correct identification during posttest as compared to the pretest. Exceptionally, Tamil group did not indicate significant improvement in percent correct identification after training. Finally the dissertation also provides a detailed optimality theoretic account of the statistical findings. Implications and shortcomings of the study are presented towards the end of the dissertation, suggestions for future research is also presented. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2014.
Local:
Adviser: WILTSHIRE,CAROLINE R.
Local:
Co-adviser: WAYLAND,RATREE.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-08-31
Statement of Responsibility:
by Ashima Aggarwal.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Embargo Date:
8/31/2015
Resource Identifier:
968785874 ( OCLC )
Classification:
LD1780 2014 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

1 TESTING ACQUISITION OF L2 HINDI VOICING AND ASPIRATION CONTRASTS: A COMPARISON FROM MULTIPLE L1S By ASHIMA AGGARWAL 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 2014

PAGE 2

2 2014 Ashima Aggarwal

PAGE 3

3 To my parents, friends, family and my nieces Kavya and Kriti

PAGE 4

4 ACKNOWLEDGEMENTS Any amount of thanks to all those who supported me throughout the past five years of my life would be meager. Yet I will take this opportunity to thank my committee chair and my idol throughout this challenging period, Dr. Caroline Wiltshire. I am grateful to her for her extremely useful comments and feedback at each stage of my dissertation. She was my c onstant motivator when I slowed down and was always there to encourage me to create and publish quality work. She is a storehouse of knowledge, and with her I discovered that my true interest lies in Phonology. She is a great adviser and a patient person to go back and forth on the n number of drafts I had. I would also like to thank Dr. Ratree Wayland, my committee member and the person who made coming to UF became a possibility for me. I still remember the email she sent me confirming my assistantship. It was like the beginning of a new life for me. Words cannot express what that email meant for me. After I came here, Dr. Wayland has always answered all m y questions, either related to academia or those of a graduate student’s anxiety. Not to forget Kelli, our office manager , who is the most helpful administrative person I have ever seen in my life, to the extent that she remembers my UFID! From fixing the Xerox machine to handling my grants, Kelli has done it all for me with a big smile and a helpful attitude. The most important person I would like to thank is a wonderful friend whom I met a year after I came to UF, during a dire need. She has been a savior and my best friend ever since. We have shared innumerable conversations and travels together , whether about linguistics, our department, UF or about international politics. Heesung has been a great source of inspiration for me, and I learned to stay calm when I was almost about to freak out while working on my PhD. I am extremely grateful to her for being there not only to discuss

PAGE 5

5 linguistics but making me feel home away from home. I am also thankful to my extremely supportive and caring friends Harsh and Divya who have stayed by my side mentally and emotionally in all ups and downs that I faced in the past five years. I would also like to thank Dr. Paula Golombek for being a wonderful supervisor and Dr. Jason Rothman and Dr. Vasudha Narayan for being on my committee and giving useful advice to me. Last but the most important, I thank my parents and grandparents from the bottom of my heart for supporting me in all ways humanly possible. They supported my decision to come to UF even though they missed me . They sacrificed their emotions but never let me feel away from their love and warmth. Above all, I thank God for his kind blessings to make this PhD finally happen!

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGEMENTS ............................................................................................... 4 LIST OF TABLES ............................................................................................................ 9 LIST OF FIGURES ........................................................................................................ 12 ABSTRACT ................................................................................................................... 13 CHAPTER 1 INTRODUCTI ON AND BACKGROUND ................................................................. 15 Introductory Remarks .............................................................................................. 15 Goals and Research Questions .............................................................................. 16 Background of Languages of the Present Study ..................................................... 20 Hindi ................................................................................................................. 20 Tamil ................................................................................................................ 21 Bodo ................................................................................................................. 21 Mizo .................................................................................................................. 22 Organization of the Dissertation .............................................................................. 22 2 LITERATURE REVIEW .......................................................................................... 25 Introductory Remarks .............................................................................................. 25 Critical Period for Phonology .................................................................................. 27 Counter Evidence to Critical Period for Phonology ................................................. 28 Critical Period and Implications for Optimalit y Theory ............................................. 31 Dispersion Theory ................................................................................................... 33 Task for the L2 Learner ........................................................................................... 34 The Gradual Learning Algorithm ...................................................................... 35 Principles and Parameters Approach ............................................................... 36 Cons idering Two Opposing Views .......................................................................... 37 Optimality Theory as the Optimal Theory ................................................................ 40 3 METHODOLOGY ................................................................................................... 47 Introductory R emarks .............................................................................................. 47 Justification for the Choice of L1s ........................................................................... 47 Research Methodology ........................................................................................... 49 Participants ....................................................................................................... 49 Stimuli ............................................................................................................... 50 Stimuli for the pretest ................................................................................. 52 Stimuli for the training ................................................................................ 54 Stimuli for the posttest ............................................................................... 55

PAGE 7

7 Stimuli for retention and generalization test ............................................... 55 Procedure ......................................................................................................... 56 Analysis .................................................................................................................. 57 4 PERCEPTION ACCURACY: PERCENT CORRECTION IDENTIFICATION .......... 59 Main ANOVA .......................................................................................................... 61 Pre Test .................................................................................................................. 61 Effects of Training ................................................................................................... 63 Pretest Vs. Posttest .......................................................................................... 63 Pretest, Posttest and RetentionTest Comparison ........................................... 64 Generalization Test .......................................................................................... 66 Between groups ......................................................................................... 67 Pretest and generalization test comparison ............................................... 67 Posttest and generalization test comparison .............................................. 68 Percent Correct Identification of Contrasts ............................................................. 69 Summary of Results for Percent Correct Identification ........................................... 72 5 REACTION TIME .................................................................................................... 80 Introductory Remarks .............................................................................................. 80 Main ANOVA .......................................................................................................... 81 Pretest .................................................................................................................... 82 Effects of Training ................................................................................................... 83 Pretest vs. Posttest .......................................................................................... 84 Pretest, Posttest and Retention Test Comparison ............................................ 85 Generalization Test .......................................................................................... 88 Be tween Groups ............................................................................................... 88 Pretest and generalization test comparison ............................................... 89 Posttest and generalization test comparison .............................................. 89 Reaction Time of Contrasts .................................................................................... 90 Su mmary of Results for Reaction time .................................................................... 90 6 DISCUSSION AND CONCLUSION ........................................................................ 98 Pretest .................................................................................................................... 98 Pretest and Posttest ............................................................................................... 99 Pretest, Posttest and Retention Test .................................................................... 101 Generalization Test ............................................................................................... 104 Account Using Dispersion Theory ......................................................................... 107 Bodo and English Ranking Using Dispersion Theory ..................................... 109 Mizo Ranking Using Dispersion Theory .......................................................... 109 Target L2 account using Dispersion Theory ................................................... 109 Account Using Laryngeal Contrasts in Optimality Theory ..................................... 110 Optimality Theoretic Account of Bodo and English ......................................... 110 Optimality Theoretic Account of Tamil ............................................................ 111 Optimality Theoretic Account of Mizo ............................................................. 111

PAGE 8

8 Optimality Theoretic Account of Constraint Ranking After Training ................ 112 Concluding Remarks ............................................................................................ 114 Implications of the Study ....................................................................................... 117 Limitations of the Present Study ........................................................................... 118 Future Directions .................................................................................................. 118 APPENDIX A EXPERIMENT INSTRUCTIONS ........................................................................... 125 Instructions Given to Subjects on the First Day of the Experiment ....................... 125 Practice Trials ................................................................................................. 125 Pretest ............................................................................................................ 125 Training .......................................................................................................... 125 Posttest .......................................................................................................... 126 B STIMULI ................................................................................................................ 127 Tokens Used for Pretest, Posttest and Retention Test ......................................... 127 Tokens Used for Generalization Test ................................................................... 128 REFERENCE LIST ...................................................................................................... 130 BIOGRAPHIC AL SKETCH .......................................................................................... 137

PAGE 9

9 LIST OF TABLES Table page 2 1 Example of Constraint ranking ............................................................................ 46 3 1 Laryngeal contrasts in all language groups ........................................................ 58 3 2 Schedule of Training ........................................................................................... 58 4 1 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the pretest .............................................................. 77 4 2 Comparis on of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three ISIs 250ms, 500ms and 1500ms ............. 77 4 3 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at pretest and posttest ................................................ 77 4 4 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the retention test .............................................. 77 4 5 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest, posttest and retention test ....................................................................................................... 78 4 6 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test ..................................... 78 4 7 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the pretest, posttest and generalization test ........... 78 4 8 Summary of significant results for percent correct identification at pretest ......... 78 4 9 Summary of significant results for percent correct identification at pretest vs posttest ............................................................................................................... 79 4 10 Summary of significant results for percent correct identification at pretest vs posttest vs retention test ..................................................................................... 79 4 11 Summary of significant results for percent correct identification at retention Test .................................................................................................................... 79 4 12 Summary of significant results for percent correct identification at generalization Test ............................................................................................. 79 5 1 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the pretest .................................................................................. 91

PAGE 10

10 5 2 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups during pretest at ISIs 250ms, 500ms and 1500ms ......................... 91 5 3 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at pretest and posttest .................................................................... 91 5 4 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups during the retention test ................................................................. 91 5 5 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest , posttest and retention test .............. 92 5 6 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test ......................................................... 92 5 7 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the pretest , posttest and generalization ..................................... 92 5 8 Summary of significant results for reaction times at pretest ............................... 92 5 9 Summary of significant results for reaction times at pretest vs posttest ............. 93 5 10 Summary of significant results for reaction times at ISIs 250ms, 500ms, 1500ms. .............................................................................................................. 93 5 11 Summary of significant results for reaction times at pretest vs posttest vs retention test ....................................................................................................... 93 5 12 Summary of significant results for reaction times at generalization test ............. 93 6 1 Pre training Bodo and English ranking using dispersion theory ....................... 120 6 2 Pre training Mizo ranking using dispersion theory ............................................ 120 6 3 Target L2 ranking using dispersion theory ........................................................ 120 6 4 Bodo and English ranking for /pa/ before training ............................................. 120 6 5 Bodo and English ranking for /ba/ before training ............................................. 121 6 6 Bodo and English ranking for /pha/ before training ........................................... 121 6 7 Bodo and English ranking for /bha/ before training ........................................... 121 6 8 Tamil ranking for /pa/ before training ................................................................ 121 6 9 Tamil ra nking for /ba/ before training ................................................................ 121 6 10 Tamil ranking for /pha/ before training ............................................................... 122

PAGE 11

11 6 11 Tamil ranking for /bha/ before training ............................................................... 122 6 12 Mizo ranking for /pa/ before training ................................................................. 122 6 13 Mizo ranking for /ba/ before training ................................................................. 122 6 14 Mizo ranking for /pha/ before training ................................................................ 123 6 15 Mizo ranking for /bha/ before training ................................................................ 123 6 16 Target L2 constraint ranking for /pa/ ................................................................. 123 6 17 Target L2 constraint ranking for /ba/ ................................................................. 123 6 18 Target L2 constraint ranking for /pha/ ............................................................... 124 6 19 Target L2 constraint ranking for /bha/ ............................................................... 124

PAGE 12

12 LIST OF FIGURES Figure page 2 1 Archangeli’s (1999) schema of OT ..................................................................... 46 4 1 Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the pretest, all ISIs combined ............................................................. 73 4 2 Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three ISIs 250ms, 500ms and 1500ms. ........................................ 73 4 3 Comparison of Mean percent correct identification of H indi, Tamil, Bodo, English and Mizo groups at pretest and posttest ................................................ 74 4 4 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the retention test .............................................. 74 4 5 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest , posttest and retention test ....................................................................................................... 75 4 6 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test ..................................... 7 5 4 7 Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the pretest, posttest and generalization test ........... 76 5 1 Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the pretest . ................................................................................. 94 5 2 Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups during pretest at ISIs 250ms, 500ms and 1500ms. ............................................ 94 5 3 Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups at pretest and posttest . ....................................................................................... 95 5 4 Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups during the retention test. ..................................................................................... 95 5 5 Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest , posttest and retention test. ................................. 96 5 6 Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test. ............................................................................. 96 5 7 Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups at the pretest , posttest and generalization test. .................................................. 97

PAGE 13

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 TESTING ACQUISITION OF L2 HINDI VOICING AND ASPIRATION CONTRASTS: A COMPARISON FROM MULTIPLE L1S By Ashima Aggarwal August 2014 Chair: Caroline Wiltshire Major: Linguistics The available literature on second language phonology provides many unresolved claims about the acquisition of nonnative phonology in adulthood. The present study successfully contributes to resolving the problem by providing empirical data that is both generalizable and verifiable. This study examines the acquisition of novel phonological categories by speakers of English and three Indian languages : Bodo, Mizo and Tamil. The target phonological categories were present in Hindi language. The speakers of English, Tamil, Bodo and Mizo underwent training that was preceded and followed by pret ests and posttests respectively . The training was aimed at helping these speakers perceive the voicing and laryngeal contrasts which are not present in Bodo, Mizo, Tamil and English. The speakers participated in an AX per ception experiment that presented C V syllables from Hindi language that differ in voicing, aspiration or both. During training, they had to respond to a given pair of syllables as being perceived as ‘same’ or ‘different .’ The participant s’ responses were followed by feedback displayed on their laptop screen. The training lasted four sessions , followed by a retention cum generalization test that was conducted two weeks after the last session. Following the

PAGE 14

14 experiment , the results were analyz ed using statistical software, including SPSS and Microsoft Excel 2007. The mean percent correct Identification and the mean reaction times of all language groups were calculated. The results indicated that all groups take significantly less time during the posttest than for the pretest. The experimental groups for Bodo, English and Mizo also show greater percentage of correct identification during posttest as compared to the pretest. Exceptionally, Tamil group did not indicate significant improvement in percent correct identification after training. Finally the dissertation also provides a detailed optimality theoretic account of the statistical findings. Implications and shortcomings of the study are presented towards the end of the dissertation, and suggestions for future research are also presented.

PAGE 15

15 CHAPTER 1 INTRODUCTION AND BACKGROUND Introduct ory Remarks For many years , researchers have taken a keen interest in the nature of second language acquisition (Flege 1991, 1992ab; Escudero 2005) and more recently in a systematic phonological account of the acquisition process as a result of an interaction between a set of innate linguistic universals often referred to as universal grammar (henceforth referred to as, UG) and transfer from first language (See Wiltshire 2005; Escudero and Boersma, 2004; Silke 2009 among others). Clearly, there is a need to pursue a formal theoretical account of second language (henceforth L2) phonological acquisition that acknowledges differences betwe en first language (henceforth L1) and L2, given that, among other factors, L2 acquisition takes place with knowledge of an L1 already in place. Bringing together insights from formal second language acquisition (SLA) theory with articulated phonological theories such as Optimality theory (Prince & Smolensky 2004) helps us to be in the best position to not only describe but also to crucially explain both the development and ultimate attainment of phonological acquisition. While some studies claim that speakers’ L2 phonologies are strongly influenced by the phonological systems of their L1s (e.g., Abramson and Lisker 1970, Werker and Tees 1984, Best et.al. 2001) and that adult L2 learners will have difficulty in achieving nativelike perception of novel second language contrasts which are not used contrastively in their native language (Brown, 1998, Werker et.al. 1981, MacKain et.al., 1980), others suggest that individuals can master nonn ative phonological systems to such an extent that they cannot be reliably distinguished from native speakers (Birdsong

PAGE 16

16 and Molis, 2001). The fact that some individuals attain nativelike proficiency while others do not leaves the question of ultimate attai nment in the domain of L2 phonology/acquisition unanswered (White 2003, Archibald 2000). The learnability of L2 phonological contrasts by post critical period adults can determine how critical the critical period actually is. The critical period is defined as a general biological term as a period of time during which external experience has the most effect on development or learning. If the organism is exposed to the same experience during or after this critical period, it will have little or no effect at all (Newport 2002). This definition bears serious consequences on the teaching and learning of a second language in adulthood. Goals and Research Questions In order to understand the critical period and its effect on phonological acquisition in adulthood, we need to use tools from both SLA theories and phonological theories. In other words, such a theoretical marriage can help us examine the extent to which phonological acquisition in adulthood is truly restricted by a critical period (HancinBhatt 2000, Es cudero and Boersma 2004, Rothman et.al. 2011). In addition, I adopt the position that the acquisition of new phonological contrasts cannot be studied independent ly of how new phonetic information is processed. Thus we need to keep in mind the levels of pr ocessing in any study that seek to understand L2 acquisition better. To this end, the current research will base itself methodologically in the findings of Werker and Logan (1985), who claim that processing takes place phonemically when the inter stimulus interval (ISI) is 1500ms , phonetically when the ISI is 500ms a nd auditory level at 250ms . Tapping the different levels of processing is crucial to making any conclusions about the perceptual capabilities of adult L2 learners.

PAGE 17

17 (A detailed review of Werker and Logan (1985) is presented in the next chapter .) This research will look at the acquisition of nonnative voicing and aspiration contrasts by speakers of various L1s using an OT perspective with ISI as a variable. Despite the fact that there is extensive research about the acquisition of voicing contrasts, to our knowledge no one has so far looked at the acquisition of nonnative voicing and aspiration contrasts using an OT perspective with ISI as a variable. With these goals in mind, this research examines how monolingual adult speakers of English and monolingual speakers of three languages of India, namely Mizo, Bodo and Tamil, perceive voicing and aspiration contrasts in L2 Hindi that are absent in their L1s. By Monolingual, in the context of this study , I mean a person who can speak, read and write only one language. With respect to perception these monolinguals have had some exposure to other languages that are commonly spoken in the geographical area around them. To be specific, 'some' exposure means that this exposure is not sufficient to give them plenty of instances of all voicing and aspiration contrasts of Hindi language. These monolinguals have had no formal education or training in which they have been exposed to the Hindi sounds that are being studied in this research. Apart from examining phonological development, the study will also examine the question of learnability past the critical period by attempting to train them to perceive nonnative laryngeal contrasts from the Hindi language. Under an OT perspective of acquisition, their initial perception of wordinitial voicing or aspiration will depend heavily on access to UG or transfer of these contrasts from their L1s (if they exist in their L1). The success of these adult L2 learners to reta in and generalize novel L2 contrasts will

PAGE 18

18 indicate their access to UG, while failure to do so might provide evidence indicating the inability of post critical adults to access UG. Acquiring the novel Hindi voicing and/or aspiration contrast requires that t he learners also allow all four way voicing and aspiration contrasts (voiced aspirated, voiced unaspirated, voiceless aspirated, voiceless unaspirated) in wordinitial position via constraint reranking (HancinBhatt 2000, 2008). A pilot study by Aggarwal ( 2011) using the AX ‘same or different’ task indicates that monolingual listeners of English, which allows only voiced unaspirated and voiceless aspirated contrasts word initially, do not make the voicing distinctions of Hindi word initially. This is becaus e English L1 speakers rank some markedness constraints above some faithfulness constraints. It will be argued in this research that in order for acquisition to take place, some high ranked L1 constraints must be demoted below the lower ranked L1constraint s (Boersma and Hayes, 2001). The overall goal of this project is to deepen the understanding of L2 acquisition in adults along with encouraging the study of some rare and understudied languages of the world (Bodo and Mizo). The goal is to employ a methodol ogy that will be empirically robust so as to produce broadly generalizable results that can be applied to second language learning and linguistic theories. The choice of Indian languages to test wa s made by keeping in mind the absence of three (Tamil), two (Bodo) or one (Mizo) laryngeal contrasts out of the total four exhibited by the target language, Hindi. A more detailed overview of the consonants in each of these languages is presented in the Methodology section of this dissertation.

PAGE 19

19 No study has ever been conducted on the perception of phonological contrasts by speakers of these rare northeast Indian languages. The study will be first to present a comparative analysis of the phenomenon in IndoAryan, TibetoBurman, IndoEuropean and Dravidian languages . Following Optimality theory (McCarthy and Prince 1993, Prince and Smolensky, 1993) this study will go beyond examining the process of acquisition to explore the interaction of various L1s with the Hindi constraint hierarchy. The extent of reranking const raints by adult L2 learners of the various L1 languages to include novel L2 contrasts in their native perceptual system will give us insights into two unanswered questions: i) Do L2 learners differ from each other in their ultimate attainment ; and ii) Do the end state grammars of L2 speakers differ from that of their native speaker counterparts. Apart from this, previous research by Werker and Tees (1984) also suggests that there are t hree levels of processing: a phonetic level that can be tapped at an i nter stimulus interval of 500ms and a phonological level that can be tapped at an inter stimulus interval of 1500ms and auditory level can be tapped at 250ms. This study aims to see whether the two ISIs have different effects on the performance of the L2 speakers. So, successful retention of the effects of training at 250ms ISI or 500ms ISI would mean that the acquisition of contrasts did not take place at the phonological level, while successful learning and retention at 1500ms ISI would mean that acquisit ion of contrasts has taken place at the phonological level. Specifically, in light of previous OT research on L2 phonological acquisition and learnability after the critical period, with respect to laryngeal contrasts in particular, the following research questions emerge:

PAGE 20

20 1. Can transfer effects (from L1) and/or accessibility to Universal Grammar (UG) explain the ultimate attainment of Hindi laryngeal contrasts by various L1 groups? Sub questions: How do L1 voicing contr asts affect the perceptual attainment of Hindi voiced stops? How do L1 aspiration contrasts affect the perceptual attainment of Hindi voiced stops? 2. Is there a main effect of ISI (250ms, 500ms and 1500ms) on the perceptual accuracy of voicing immediately and two weeks after training? Backgrou nd of Languages of the Present Study Below is a map of India (Figure1. 1) showing the various regions where the languages examined in this study, namely, Bodo, Mizo, Tamil and the target language Hindi, are spoken. Hindi Hindi is spoken by 422,048,642 people in India and in other countries including, but not limited to, Australia, Bangladesh, Bhutan, Botswana, Canada, Djibouti, Equatorial Guinea, Germany, Guyana, Kenya, Nepal, the United Kingdom and the United States. It is the first language of most people living in the north Indian states of Delhi , Uttar Pradesh, Uttarakhand, Rajasthan, Punjab, Madhya Pradesh, Northern Bihar and Himachal Pradesh. Despite the many regional varieties, most phonological and phonetic aspects of Hindi stay constant, and in particular, the contrasts of voicing and aspiration in the stop series appear in all varieties. For this study , Hindi serves as the target language. The stimuli used for the study were recorded by a native speaker of the Delhi variety of Hindi.

PAGE 21

21 Tamil Tami l belongs to the Dravidian language family and occupies a unique position because of its geographical expansion. It is not only the first language of the majority of the population residing in Tamil Nadu in India, but is also the spoken and written language of several millions of Tamils living in Ceylon, Burma, Singapore, Malaysia, Indonesia, South Africa and Mauritius. In India, Tamil is also spoken by minorities in other states such as Karnataka, Andhra Pradesh, Kerala and Maharashtra. When it comes to the sociolinguistic situation with respects to Tamil dialects, a diaglossic situation exists. There are two separate registers that vary by social status and context of use: a high Tamil and a low Tamil (Ferguson, 1959) . Thes e dialects are primarily differentiated from each other in terms of the different phonological changes and sound shifts they have undergone while evolving from Old Tamil. According to Kamil Zvelebil (1970), the Tamil dialects can be segregated on the following 'Centers of Prestige': Madras Tamil , Madurai Tamil , Kongu Tamil , Ne llai Tamil, Kanyakumari Tamil, Thanjavur, Tiruchirappalli Tamil, Jaffna or Yazhpanam Tamil, Triconmalee or Tiruconamalai Tamil, Batticaloa or Mattakkalappu Tamil. For the purpose of this study , I will focus on the variety of Tamil spoken in Chennai (formerly Madras) by lower class people. Bodo The Bodos , often referred to as [ boro] [ bodo] [ b ] , are one of the ethnic and linguistic communities of India. The Bodos were among the early settlers in the largest n orth e ast state of India, namely Assam. The word ‘Bodo’ is used to denote both the language and the ethnic group. Bodo belongs to the Bodo subsection of the BodoNaga section under the Assam Burmese group of the TibetoBurman branch of the Tibeto-

PAGE 22

22 Chinese family. According to the 2001 census1, the number of Bodo speakers in India was 1,350,487. Mizo Unlike the Bodos, Mizo was traditionally the name of the people and not the language they spoke. Due to external influence, the Bengali called them Kuki, the Chakmas called them Hugi , the Burmese called them Chin and the Indians called them Mizo or Lushai . For this dissertation, I will call it the Mizo language since that is the most prevalent name used for this language in India. Mizo is a TibetoBurman language spoken by 675,000 people in India (2001 census) and 687,750 people from all over the world. It is primarily spoken in Mizoram; Assam; Manipur, Churachandpur district; Nagaland and Tripura. Organization of the D issertation The rest of the dissertation is organized as follows : i n C hapter 2, I review the literature relevant to this research that discusses previous studies on s econd l anguage a cquisition and various phonological and phonetic theoretical models related to L2 acquisition. Chapter 3 presents the methodology adopted in this dissertation. It provides the rationale for the use of languages being studied, along with a phonological overview of the consonant inventories of the Tamil, B odo, and Mizo languages. Chapter 4 presents the results drawn from the data. It shows statistical analysis on percent correct identification and main effect of ISI across the four language groups. The results indicate increases in the percent of correct identification of Bodo, Mizo and English groups and shorter reaction times during the posttest for all the experimental groups. 1 The census allows people to self identify, and some may say they are Bodo speakers because they are ethnically Bodo, not because they actually speak it.

PAGE 23

23 Chapter 5 discusses the implications of the results from an optimality theoretic perspective and Universal Grammar perspective; fi nally it suggests scope for further research in the area.

PAGE 24

24 Figure 11. Linguistic areas of India

PAGE 25

25 CHAPTER 2 LITERATURE REVIEW Introduct ory Remarks Although first language acquisition has been shown to be effortless, such is not the case with second language acquisition, especially for adults. In addition to this, second language acquisition differs from first language acquisition in that L2 acquisition might not always be complete (Fathman, 1957, Oyama , 1956) . Both L1 and L2 acquisition are af fected by a number of factors, such as universal developmental order, constraints on possible phonological structures, aptitude, attitude and input . The interaction of these variables may result in differential success in attainment of a language. Especial ly for the domain of L2 phonology, neuromuscular abilities are said to constrain the acquisition of sounds that require different articulatory habits, rendering the fate of L2 phonological acquisition to somewhat less than nativelike (Scovel, 1981, Wode, 1989, 1992). Apart from the issues of muscular dexterity, the influence of L1 phonology has been considered unavoidable when it comes to adult L2 acquisition (Flege 1987a,b, Flege and Munro 1994). Flege claims that agerelated accentedness is perceptually based. He uses Wode’s (1993, 1994) terms continuous mode and categorical mode of perception to elaborate on this claim. Where continuous mode of perception lets one discriminate between the fine gradience in the degree of voicing of, for example /b/, the categorical mode permits the hearer to only perceive those cues that bring about a phonemic contrast between two sounds (e.g., /p/ /b/). Flege goes on to suggest that children begin with a continuous mode of perception, but as they begin to hear the cont rastive phones of their language, they tune their perception to slowly create the phonetic categories

PAGE 26

26 that belong to their native language. These listeners develop an affinity to perceive L2 sounds in terms of their L1 categories. Flege proposed that L2 ph onetic categories that are perceived as similar to L1 categories are merged with them, failing in the construction of new phonetic categories for them. However, this does not rule out the possibility of new sound categories for a new language. It means that the learners will need to reactivate the continuous mode of perception by focusing on minute differences in Voice Onset Time (VOT) between L1 and L2 to form new phonetic categories. The relationship between maturational or psychological constraints, indi cated by muscular dexterity or phonetic categorization of speech perception, and phonological acquisition continues to interest researchers up to the present time. Werker and Logan (1985) made a distinction between different levels of processing based on t heir experiment. They conducted an AX task where contrasting speech syllables were separated by three different ISIs. They concluded that when their experiment replicated the real life communication (longer ISIs) then adults rely on language specific phonemic processing strategy. In other words, they classify the sounds according to the phonological categories that are used to contrast meaning in their native language. Contrary to this , when they used shorter ISIs the participants were sensitive to ac oustic variability, thereby indicating an auditory level of processing. The present study uses this as a background in choosing the three different ISIs in the training of adult monolinguals. Some works have emphasized the role of maturational constraints in phonological acquisition. Long’s (1990) study has shown the domain of phonology to be different from syntax, semantics or morphology, in that it is restricted by a very early

PAGE 27

27 critical period, 56 years of age. It is important to define the critical peri od here as per different researchers. According to Singleton (2005) there is a lot of variation in what is meant by the Critical Period Hypothesis (CPH) based on definitions of various researchers. Our definition of the CPH will affect whether data counts as evidence or counter evidence to the existence of a critical period for phonology (Long, 2005). However there are some similarities in all accounts of critical period: all of them relate to maturational consequences. For instance, for Lenneberg (1967) and Newport (2002), the critical period is biological in nature. The critical period is defined as a time frame during which an external stimulus has the maximum influence on development or acquisition. If an organism is exposed to the same experience during or after this critical period, the experience will have little or no effect at all. The existence of a CPH for any linguistic domain would then imply that “the effects of linguistic experi e nces are (a) cognitively based, (b) thus unavoidable, and (c) sur face in the loss of ability, resulting in a can vs. cannot dichotomy” (Rothman 2008). These implications in turn make possible the empirical verification or falsification of the CPH. Critical Period for Phonology Many researchers have claimed that there e xists a critical period for phonological acquisition, which explains the lim itations on the success of L2 acquisition by adults . Of those, a few are summarized in this section. Fathman’s (1975) study of the order of acquisition of child L2 English suggested that 610 year olds were better at pronunciation than their 1115 year old counterparts, while older children did better on higher order structures such as syntax and semantics, even when both the age groups had equal amounts of exposure to all syntactic , semantic and phonological structures. Oyama (1976) studies whether age of onset (AO) and length of exposure played a role

PAGE 28

28 in phonological production. Based on two tasks, namely, a readaloud and a narration task, 60 ESL learners were rated for foreign ac cent. It was found that AO, not years of exposure, had an effect on the production. However , there might be an effect of the context because it was observed that accent was relatively greater while reading casual stories than while reading formal narratives. In general, L2 phonology studies have looked at effects of exposure and length of residence as variables that affect L2 production, while less attention has been paid to factors such as learners’ concern for accuracy that is of great importance in pronunciation accuracy (Elliott , 1995). Counter Evidence to Critical Period for Phonology There is an emerging literature on the ultimate attainment in adult learners (White 2000, 2003, Swartz and Sprouse 1996) and possible access to UG by adult L2 learners (B irdsong 1992, Young Scholten 1994, Ioup 1989, White & Genesee 1996). In their studies on phonological acquisition in adulthood, Bongaerts, Planken, and Schils (1995) and Bongaerts et.al.(1997) have shown that late learners can achieve nativelike performance. In the study by Bongaerts (1999), he looked at 41 native English speaking subjects and 41 native Dutchspeaking subjects who had not been exposed to English until age 12. Both the groups were given a read aloud task that was judged by nave native spea kers of English for authenticity. Results showed that 5 out of 41 Dutch participants were rated within 1.2 standard deviations of the mean for the native English group. In a similar study on restrictions on French liaison using readingaloud tasks conducted by Birdsong (1997), one fifth of the participants (adults) performed correctly, without any errors, thus exhibiting the complete acquisition of liaison. Jacobvits (1968) rejects the notion of sensitive period based on his argument that not all children l earning

PAGE 29

29 a foreign language in school attain nativelike pronunciation and some adults can acquire nativelike pronunciation and conversational knowledge with intensive training. Neufeld (1977) conducted a study on 20 adult native speakers of English. After receiving 18 hours of intensive instruction in Japanese and Chinese phonology, the 20 adult subjects recorded ten statements in both Chinese and Japanese. The recording was played to three native speaker judges of Chinese and three native speaker judges of Japanese. Two of the twenty participants were rated as native speakers of one of the languages and one subject received native speaker ratings for both languages. These findings led Neufeld to conclude that a second language can be acquired in adulthood without an accent, and hence L2 phonological acquisition is not restricted by a critical period. Klein (1995) also argued that absolute biological barriers for adult perception and production do not exist. What is necessary is massive and continued access to L2 input. He argued that this is possible if the learner has a very high level of motivation which arises from his or her need to sound like a native speaker of that language. Recently more light was shed on the amount and quality of input that is needed for successful L2 acquisition by Rothman and GuijarroFuentes (2010). They claim that L2 acquisition takes place in the classroom and outside with a much smaller amount of input than that required in naturalistic settings. This could be said to contri bute to the nativenonnative differences in acquisition, but the generative paradigm suggests otherwise; since within the generative paradigm linguistic properties are not acquired directly from the input, but from the linguistic universals that are triggered due to the acquisition of properties of the languagespecific lexicon of the environment, equal

PAGE 30

30 quantity of input would not be needed. Rothman and GuijarroFuentes (2010) suggest what is required is ‘ minimally sufficient’ input that provides the triggers for the system to be configured. Schneiderman and Desmarais (1988) reported that their two native English subjects who acquired many languages in adulthood were judged as belonging to Frenchspeaking areas by native Francophones, who listened to short tape recordings of interviews of these subjects. One of the two subjects could also pass as a native speaker of Spanish. In another study by Novoa, Fein and Oblet (1988), a native English speaking subject C.J. , with no exceptional IQ, who acquired French, German, Italian, Moroccan Arabic and Spanish after the age of 15 was judged as lack ing foreign accent by nativespeakers of all these languages (although no formal tests were administered). Finally, Ioup (1989) reported on the success of their two subjects , Julie and Laura (nativeEnglish speakers) , in acquiring Egyptian Arabic as adults. They were described as native speakers by 8 out of 13 judges. The results of such studies would provide a challenge to Scovel’s assertion that there exists a maturational ly constrained time frame for acquisition of accent, ending around 12 years of age, beyond which it is impossible for learners to acquire such a good pronunciation in a nonnative language that they can be judged as native speakers of that language (Scovel 1988). The fact that in studies that show nativelike successful acquisition of linguistic properties by adult learners are constituted by 5 20% of the subject population that is successful cannot be ignored. This shows the possibility of late learning is not just accidental but highly relevant to the study of L2 acquisition as it relates to the critical

PAGE 31

31 period (cf. Bley Vroman, 1989; Selinker, 1972). Thus a great chunk of literature undermines the role of critical period as it relates to phonology. Critic al Period and Implications for Optimality Theory What the critical period entails for phonological acquisition is crucial. If the strongest claims made by researchers regarding the presence of a critical period for phonology hold true, then they would imply that no L2 learning is possible. For an Optimality Theory of phonology (Prince and Smolensky, 1993) , it would mean that there is no possibility of reranking phonological constraints. The advent of Optimality T heory can be traced back to Prince and Smolensky (1993) as a move away from derivational approaches to phonology. Similar to the derivational approach, the underlying mental representation and the surface representation of sounds still exist, but now they are referred to as the input and the output respectively. Instead of rules, there are constraints. Archangeli (1999) provides a description of the way OT works (Figure 21 ). It has components called the GENerator, EVALuator and the CONstraints. The generat or generates a set of candidates or prospective outputs that are evaluated by the evaluator on the basis of the least expensive violation incurred against the universal constraint set, thereby producing the optimal output candidate. In terms of OT, univers al grammar is proposed as a set of universal constraints that are violable. The universality of these constraints themselves provides a means to explain the cross linguistic similarities on one hand and the violability of these constraints explains language variation on the other. OT is economical since it presents a single means to express which constraints are violable through strict domination of constraints, unlike rewrite rules that were so many in number and ordered with respect to one another.

PAGE 32

32 The universal set of constraints consists of either of the two classes: m arkedness constraints or f aithfulness constraints. Markedness constraints ensure well formedness by restricting marked structures in languages while faithfulness constraints allow preservat ion of input structure. Both types of constraints are potentially violable, and the winning candidate is the one that incurs the least serious violations. The seriousness of the violation is determined by how high the constraint is ranked in the hierarchy of that language. In Table 21 the constraint ranking is strictly hierarchical. CONSTRAINT 1>>CONSTRAINT 2>> CONSTRAINT 3. Since CONSTRAINT 1 is highest ranked, any violat ion will not be tolerated. CONSTRAINT 2 is the second most expensive constraint to be violated so candidate 2 is also ruled out if there are more optimal candidates such as candidate 3 here, which violates only the lowest ranked constraint and is therefore the winner. The rankings of constraints are language– specific, and it is claimed that different constraint rankings will give rise to different language typologies. Thus there will be languages that use a different constraint ranking such as CONSTRAINT 2>>CONSTRAINT 1>> CONSTRAINT 3 or CONSTRAINT 1>>CONSTRAINT 3>> CONSTRAINT 2 that will lead to different optimal candidates for those languages. OT assumes that markedness and faithfulness constraints are arranged in a languagespecific hierarchy in the phonologies of speakers, and it is possible for a learner to rerank the constraints in response to the input. Its components , GEN, EVAL and CON, allow infinite outputs to be evaluated based on the languagespecific ranking of universal constraints. A CPH would then place a restriction on the reranking of these

PAGE 33

33 constraints in the face of L2 input whereas a weak CPH would allow reranking of const r aints in a hierarchy different from that of the L1. Beyond accents, a critical period for L2 phonology would indicate that interlanguage phonology would show no development or changes in the ranking of constraints. In terms of principles and parameters approach, learners would have lost their ability to reset the parameters and would be limited to their L1 final ranking (assuming full access). The initial state and the end state of the L2 speaker would then be the same. The lack of reranking would in turn reflect the inaccessibility of Universal Grammar (Chomsky 1995) or accessibility to UG only via the L1. UG proposes that there exists a set of abstract principles that is present in all of the world’s nat ural human languages (Chomsky 1995). In terms of o ptimality theory, a critical period may mean that the L2 learners no more have the ability to rerank UG constraints a nd their interlanguage grammar is not constrained by UG except insofar as L1 is. Movin g beyond the omnipresent L2 accent, YoungScholten and Archibald (2000) in their review of L2 syllable structures seek to answer whether L2 syllables are constrained by allowable L1 syllable structures. They found that not only are sounds of a language transferred, but there is also evidence that learners attempt to maintain their L1 syllable structure; that is, there is transfer from L1. Thus, L2 syllable structure is in part shaped by native language phonology. Finally in terms of ultimate attainment, the existence of a critical period for phonology denies any learnability post puberty and hence never the possibility of nativelike speech perception and production. Dispersion Theory Dispersion theory suggests that there are constraints on the well formedness of phonological contrasts. These contrasts are : i. Maximize the number of contrasts, ii.

PAGE 34

34 m aximize the distinctiveness of contrasts, and iii. m inimize articulatory effort. These constraints entail that the well formedness of a word cannot be evaluated by itself ; instead, it needs to be evaluated with respect to other contrasting forms (Flemming 1996). From a perceptual point of view, the maximization of the distinctiveness between contrasts serves a functional goal of helping listeners perc ei ve the contr asts better. Optimality theory gives an optimal candidate by using a strict hierarchy of constraints , but in d ispersion theory , the complete dominance of any one constraint will not give appropriate results. F or instance, if 'maximizing the contrasts' dominates, then we will get two contrasts that have maximal distinction. There won't be any intermediate finer distinctions. The idea in d ispersion t heory is to balance the constraints and then model them by using strict dominance of subcases. Task for the L2 Learner In this study, the task for the learners from an o ptimality t heory (Prince and Smolensky 1993) perspective would be to arrive at the L2 Hindi ranking for voicing and a spiration in stop consonants in word initial position . Optimality theory assumes that these components are universal. Differences in grammars reflect different rankings of the universal constraint set, CON. Part of language acquisition can then be described as the process of adjusting t he ranking of these constraints. For this they must recognize the underlying forms and the ranking of constraints that lead to correct surface forms for the target language. Within this theoretic al framework, the speakers of various L1s Mizo, Bodo, Tamil are expected to arrive at the Hindi ranking by the use of Gradual Learning Algorithm abbreviated GLA (Boersma 1997; Boersma & Hayes 2001).

PAGE 35

35 The Gradual Learning Algorithm Boersma (1998) proposes stochastic constraint ranking, that is, the ranking of constraints is not absolute or fixed, being instead based on probability. Given such a mechanism , one can explain how the same given input might have different possible outputs even though one output might be more probable than the others. All the constraints in a child’s grammar begin with the arbitrary value of 100. Each constraint is associated with a probability density function whose center lies at its ranking value. The evaluation is nondeterministic, and a strict domination relationship between constraints results if there is space between the constraints once they are ranked. In other words , each constraint has a range of values it can select from at the time of evaluation (production or perception). At this time of evaluation, if the value of two constrai nts selected by the listener overlaps, it will lead to free variation. For a perception grammar , it implies that the listener will alternate between perceiving two phonetic forms. On the other hand, if the values selected for the two constraints are far apart , then the ranking of constraints is relatively strict and the listener will always hear one phonetic form and not the other. This will happen if the input does not have much variation. That is, if the input data is not varied then the constraint values will be far apart and fixed. Whenever a listener hears a target language input, he or she tries to match the ranking values of the target form and his or her form. If they do not match, then the listener increases the values (and thereby the strictness) of all constraints that allow the listener’s form and decreases the values of the constraints that allow the target form. The Gradual Learning Algorithm can also describe the stages of acquisition. Curtin (2001) uses GLA to model the variation in children’ s production and their

PAGE 36

36 stepwise changes in grammar using an orderly demotion of markedness constraints , ther eby resulting in the acquisition of adult like prosody. An advantage of the GLA is that it can learn rates of variation because contradictory options exert contrasting effect on ranking values. This learning also relies on frequency in that the more frequently heard variant gets relevant ranking values of the interacting constraints. Eventually the difference in frequency of conflicting inputs may separate the constraint values so far apart that the ranking will appear fixed. For this study, the various L2 learners should begin with a ranking of their L1 constraints and successfully rerank their constraints in accordance with the L2 input. Principles and Parameters Approach Considering the principles and p arameters (Chomsky and Lasnik 1993) approach, which suggests the existence of abstract linguistic principles and switches or parameters that one needs to set in consonance with the linguistic features present in the target language, the learners of this study could have one of the four possibilities : Full Transfer Full Access, Full T ransfer Partial Access, Partial Transfer Full Access or Partial Transfer Partial Access. A ssuming Full Transfer and Full Access (Schwartz and Sprouse 1996) , which means that the learner has access to all the principles of Universal Grammar beyond his L1 and can transfer all the parameters that he or she set for the L1, and the constraint ranking would appear to have fait hful ness constraints ranked below markedness . The process of L2 acquisition involves subsequent reranking within the markedness constraints, followed by the promotion of faithfulness constraints above markedness once again (HancinBhatt 2008). Boersma and Esc udero (2003) have successfully used phonetically based phonological constraints to portray the perception grammar of Canadian English and Canadian French learners

PAGE 37

37 using the GLA. They make use of ‘optimal perception hypothesis’ that states that an optimal l istener will select auditory properties that consistently distinguish sounds in the production of his or her first language. The optimal hearer is likely to perceive the intended sounds by the speaker. Thus the stimuli used in the training should help the listeners, on the basis of frequency, establish the constraint ranking that allows them to hear all four voicing and aspiration contrasts. Considering Two Opposing V iews Under the principles and parameter approach, the no parameter resetting view suggests that speakers can resort to only those parameter settings in their interlanguage that are present in their L1. This view suggests that any subsequent parameter resetting is not possible in response to L2 input, which means that new parameter values cannot be acquired. No parameter resetting means that nativelike representations like those of speakers of L2 will definitely be unattainable if the L1 and L2 have different parameter values. No parameter resetting then infers that there is full transfer but not full access. An interlanguage grammar based on no parameter resetting would mean that there would be no change in the learner’s representations subsequently. The representations will remain the same, as they were initially, namely L1based (White 2003). T he opposing position to the no parameter resetting position is the ‘parameter setting and resetting’ position. This assumes that the interlanguage grammar of the learner is fully constrained by UG. The learners’ grammar can then draw from universal grammar , even what was not present in their L1. Assuming a Full Transfer/ Full Access view, the initial state would consist of L1 parameter settings with the possibility of their being reset in response to the properties of the L2 input (Schwartz and Sprouse 1994,

PAGE 38

38 1996). Unlike the noparameter resetting hypothesis, the initial and subsequent stages within the interlanguage grammar will be different if the L1 and L2 have different parameter values. Assuming an alternative approach, Full Access No T ransfer support s that L1 is never present in the interlanguage grammar ; that is, it does not form a part of the interlanguage grammar at the initial stage or at subsequent stages. Instead the parameters are set to L2 values right away from the beginning. This is possible because the learners have access to UG that in turn interacts with the incoming input from L2 sans a preceding L1setting stage. This suggests that there is parameter setting, like in the case of L1 acquisition, but no need for resetting (White 2003). Thus the Full T ransfer /Full Access and Full Access without T ransfer agree in that restructuring of parameter values and hence attainment of L2 parameter values is possible; where they differ is the issue, whether L1 settings are ever transferred or not. At the same time, they disagree with the ‘no parameter resetting hypothesis’ that disallows any resetting during the course of L2 development. If it is assumed that noresetting is possible in the case of the L1 learners involved in this study, namely, Mizo, Bodo and Tamil, then it should be found through the perception test that the speakers are incapable of perceiving the contrast(s) missing in their L1. In other words, the Tamils should not be able to perceive anything other than plain, voiceless stops because only this contrast is present in their L1, the Bodos who have only voiceless stops and voiced aspirated stops should not be able to perceive any other contrast and the Mizos should never be able to perceive voiced aspirated

PAGE 39

39 stops from Hindi since the Mizos are missing the voiced aspirated stops in their inventories , irrespective of the type of training or the stimuli used. If it is assumed that resetting of parameters is possible, then it can be claimed that the L2 learners can move beyond the settings of their L1 to adapt to new voicing and aspiration contrasts, even in adulthood. This stance is somewhat more suitable in order to explain learnability. Previous research within the principles and parameters framework suggests that not only are L1 phono logical parameter settings transferred by learners (Archibald 1993), but also that parameters are reset to a new value (Broselow and Finer 1991, YoungScholten 1994). The finding that even though learners may not arrive at the parameter setting of the L2, but that they progress beyond the L1 setting, is evidence that the learner has access to the parameter involved beyond his or her L1.This helps the building of hypotheses for the current project. In consonance with the full Tr ansfer/Full A ccess (Schwartz and Sprouse 1996) and Optimality Theoretic account of perception by Boersma and Escudero (2003) that suggests that reranking in L2 phonology is possible in response to L2 input, I hypothesize that the various L1 speakers at test in this dissertation study should be able to acquire the Hindi ranking, given training with L2 Hindi stimuli. I propose that the initial rankings of each of these learners should reflect their L1 ranking with respect to voicing and aspiration contrasts in onset position. They will t hen, in light of positive evidence from Hindi, rearrange their constraint hierarchy to match that of L2 Hindi. They will have full access to UG even when the L1 lacks the phonological contrast. This means that the end state rankings of L1 and L2 will differ; however, I do not make any claim that the L2 learners will

PAGE 40

40 eventually attain complete knowledge of the L2 contrasts (or that they will not). The interlanguage of the Mizo, Bodo and Tamil learners will show resetting of parameters that will be different from their respective L1s and may be similar to L2 or to some language attested in the world. The results of this study bear great consequences in either supporting or contradicting the existence of the critical period. A reranking of the existing L1 const raint hierarchy (as evidenced by comparison with the pretest) would indicate successful acquisition al capabilities even after puberty, defying the strict CPH. However, the lack of any restructuring of the L1 constraint ranking would be compatible with the existence of a strict critical period for phonological domain. In either case, the implications have serious consequences for L2 pedagogy. Unaccented and ‘unimpaired’ learning, which is the concern of most foreign language learners, depends on the strictne ss of this maturational barrier. The study will conclude whether it is possible to learn novel L2 sounds in adulthood or whether learners need to start early. The question of ‘how early?’ is yet beyond the scope of this project because of the varied defini tions of the critical period proposed by different researchers. Optimality Theory as the Optimal Theory There are three conditions that are proposed by HancinBhatt (2008) to be met in order for a theory to be adequate as a theory of acquisition: f irst, ge neralizability across the levels of phonology ; second, accountability for the range of possible structures that occur in L2 sound patterns ; and third, predictability with respect to how an input will trigger the restructuring of the grammar (also see Grosj ean 1998 and White 2000). This section presents a brief overview of some models that seek to explain L2 acquisition and why they are or are not optimal theories to explain SLA .

PAGE 41

41 Major’s Ontogeny Phylogeny Model (OPM) intends to capture the patterns in inter language grammar and the relationship between L1, L2 and universals. He claims that ‘L2 increases, L1 decreases, and U [universals] increases and then decreases.’ At the initial stage of learning, the learner has only L1, and universals are accessible only through the L1. Although Major’s OPM made predictions about the stylistic variations and linguistic relationships such as similarity and markedness, it leaves a lot of questions unanswered, such as the working of the universals and the L1 and L2 systems. Another model suggested to explain the perception of nonnative contrasts lies in the seminal work by Best (1995) , who proposed the Perceptual Assimilation Model (PAM). PAM suggests that L2 perception is greatly influenced by the L1 phonological system an d that the perception of nonnative segments is performed on the basis of their being sameas or different from the native phonological system (Best, 1995). According to the PAM, a given nonnative phone may be perceptually assimilated to the native phonol ogical sy stem in one of six ways. First is T wo Category assimilation (TC) , where two nonnative phones assimilate to two d ifferent native phones. Second is Single Category assimilation (SC) , when two nonnative phones assimil ate to the same L1 phone. Third is Category Goodness (CG) , which occurs when two nonnative phones get assimilated to the same native phone but one of the nonnative phones seems to fit better than the other. Next, UncategorizedCategorized pair (UC) occurs if one nonnative phone is categorized while the other remains uncategorized. However if both the nonnative phones cannot be categorized, we have a condition called UncategorizedUncategorized pair (UU). Last, the nonnative phones remain NonAssimilable (NA) if

PAGE 42

42 the L2 sounds are per ceived as nonspeech sounds different from any native phonemes. Thus the PAM describes the assimilation of not individual sounds but sound contrasts of nonnative language into that of L1 categories. To this end, PAM is a suitable model to describe the as similation possibilities of nonnative to native phoneme categories, but it might be inadequate to provide an acquisitional account of second language phonemes. In other words , it does not describe how listeners progress as the course of learning proceeds. PAM cannot be criticized as a model of acquisition per se since it meets the requirements of an adequate model as proposed by HancinBhatt, but I suggest that it is does not provide an account of the learning process. It does not shed light on the resetti ng of parameters, speaking in strict UG terms. In the discussion of various acquisition theories , the mention of Flege’s SLM (1995) is important. The SLM claims that the more different the L2 and the nearest L1 sound are, the more likely the learners wil l notice the dissimilarity between the two sounds. In other words, the L2 sounds that remain dissimilated are less challenging than the sounds that are strongly assimilated into L1 sounds/categories. According to Flege’s SLM, the sounds in one’s L1 and L2 interact in two ways. First is “category assimilation , ” which means that when a novel category cannot be instated for the L2 sound because they are always classified as examples of L1 sound, a new category for the L2 sound is not formed; instead, the L1 and the L2 sound ‘merge’ into one phonetic category over time. It is hypothesized that this merged category incorporates the phonetic characteristics of the two L1 and L2 sounds. In terms of voicing, then, an example could be a native speaker of French, wher e /t/ is represented by a short lag

PAGE 43

43 VOT of about 20ms, who learns English as his L2, where /t/ is realized by a longlag VOT value of about 80ms, might develop a merged phonetic category of the two L1 and L2 /t/ sounds producing the new merged /t/ with a V OT value of 60ms (intermediate value of the L1 and L2). This new merged category will now be used by the speaker in both his L1 French and L2 English in the absence of two different categories for each language. The second way in which the L1 and L2 phonetic systems are predicted to interact based on the SLM is called “phonetic category dissimilation , ” resulting in the learner forming a new L2 category. Since bilinguals want to retain a contrast amongst all sounds present in their common L1L2 phonetic are a, it causes the newly formed L2 category to dissimilate or move away from the nearest L1 sound. This is evident from a study conducted by Flege and Eefting (1987) that showed that EnglishSpanish bilinguals used longer VOT values for English voiceless stops and shorter VOT values for Spanish voiceless stops than their monolingual counterparts. This is evidence indicating that L1 Spanish voiceless stop categories somewhat dissimilated from the L2 English voiceless stop categories. Although Flege’s SLM explains the processes going on during acquisition of L2 sounds, it is not clear how it can capture the developmental sequence of the different L1 groups at the different stages of L2 phonological acquisition. The first condition means we should be able to prov ide an account for segments, syllables, feet, words, etc., but PAM (Flege 1985) deals with segments and categories , and it isn’t clear how they’d be extended to other levels. The second condition, accountability, means that a theory should be able to explain the various L2 sound

PAGE 44

44 patterns. OT can describe the existence of aspirated voiced, voiced, voiceless and voiceless aspirated sounds by means of constraint ranking for particular languages , but it is not clear how theories like PAM or Major’s ontogeny phy logeny Model would account for them. The third condition, predictability, means that the theory should be able to forecast how a particular sound will be perceived. SLM and PAM can make predictions about speech perception using phonetic assimilation and dissimilation to L1 sounds but they cannot explain the nativelike perception L1like nonnative sounds. HancinBhatt elaborates further that there are some fundamental issues that a theory of phonological acquisition needs to address : foremost, the theory must define what the assumed phonological features at the initial state are and in subsequent stages ; secondly, what are the inputs and representations that force restructuring in the grammar ; and finally how does the grammar accommodate multiple grammatical representations for a single input ? In the following discussion, I will show how OT fulfills all the criteria of a sound phonological theory as compared to other theories such as those proposed by Best (1995), Flege (1995), Major (2001). Although the task for an adequate theory is by no means simple, nevertheless Optimality Theory has the elements to meet the requirements of a sound acquisitional theory. A brief summary of OT follows, some of which has been mentioned in part in previous sections. These different rankings also delimit the grammars of the world’s languages, unlike the large number of rules within derivational phonology. An advantage of OT over the rule based approaches is its straightforwardness in explanation of the ‘homogeneity of target/heterogeneity of process’ (McCarthy 2001).

PAGE 45

45 Within a rulebased theory, a rule describes the target and the change that applies to it. But there is no motivation for why a number of processes apply to the same target over and over again. There is no explanation for why the same target undergoes recurrent changes. Contrary to this, OT successfully explains the reason: to avoid the target from being marked; that is, the markedness constraints prevent the target from occurring as it is, and therefore it needs to undergo repairs. However, what repairs will take place, resulting in the final output depends on the language specific ranking. Unlike PAM , which cannot explain acquisition at all levels of phonological structure, OT is capable of handling phonological acquisition across levels. Markedness applies not only to segments but to structures also. In other words , markedness at all phonological levels helps explain acquisition at all levels. OT i s capable of not just modeling learning but also describing the reranking during the interlanguage stages of grammar. Universality of the constraints, violability of the constraints and the different ranking of constraints in languages form the foundational pillars of explanatory adequacy in Optimality Theory. These functions and workings of the OT make the case for it to be the best suitable theory to describe acquisition. Where the universal set of constraints characterizes similarities in the world’s gr ammars, violability of constraints allows cross linguistic variation. Last ly, different rankings explain typological variation among languages. This, coupled with the g radual learning algorithm, can efficiently explain the L2 acquisition of novel second la nguage properties or contrasts.

PAGE 46

46 Figure 21. Archangeli’s (1999) schema of OT Table 21. Example of Constraint ranking Input Constraint 1 Constraint 2 Constraint 3 Candidate 1 *! Candidate 2 * Candidate 3 * Input GEN candidate set Output EVAL Constraint hierarchy of the language

PAGE 47

47 CHAPTER 3 METHODOLOGY Introduct ory Remarks This chapter gives a detailed description of the methodology used to collect data, the participants, the stimuli used for pretest , posttest and retention cum generalization test. It also discusses the method used for analyzing the data from the study, and the suitability of the languages chosen for this study. Justification for the C hoice of L1s While the studies on the perception of novel L2 contrasts have focused on many of the more common languages, there has not been any extensive study on perceptual acquisition by speakers of some of the world’s rarest languages. Surveys suggest that there are about 7, 100 languages spoken a round the globe and that approximately 50% of these languages are under threat of extinction by the next century (source: travel.nationalgeographic.com). Two of the languages that were a part of this research project, Mizo and Bodo, are rare languages of northeastern India. These languages are spoken by people with limited accessibility living in remote hills. Bodo belongs to the BodoGaro family of the TibetoBurman languages and Mizo belongs to the Kukish branch of the TibetoBurman family. Today Mizo and Bodo people choose to speak Assamese, an IndoEuropean language, which is the dominant language of the area, or English when they move out of their local area. The number of monolingual speakers of Mizo and Bodo are rapidly decreasing. It is therefore of utmost importance to the linguistic community and to the society as a whole that we preserve the unique phonetic and phonological characteristics of these languages for later

PAGE 48

48 reference. In addition, once these people become bilinguals , it would be impos sible to tap their unique perceptual systems, exclusive of the influence of another language. Another important linguistic factor that makes these languages suitable for testing is that previous studies have shown that Bodo has 6 vowels and 16 consonants. Out of these 16 consonants, the stop series, consisting of bilabial, dental and velar stops, has only a twoway laryngeal contrast, using only aspirated voiceless stops and unaspirated voiced stops, as opposed to Hindi, which has a full four way laryngeal contrast. Mizo, on the other hand, has 20 consonants and 6 vowels and has a threeway voicing and aspiration contrast in stop consonants. Like Bodo, it uses aspirated voiceless stops and unaspirated voiced stops and differs from Bodo in that it also allow s unaspirated voiceless stops. However , Mizo differs from Hindi in the absence of a fourth contrast, namely, the voiced aspirated stop. The third language at hand is Tamil, a Dravidian language of Southern India, which has the least number of contrasts as compared to Hindi. Tamil allows only voiceless , unaspirated stops in word initial position. Although the voiceless unaspirated stop is voiced intervocalically and post nasally, the variation is allophonic not phonemic (Zvelibil 1970, Christdas 1988). The fourth language to be tested is English. Similar to Bodo, English allows voiceless aspirated and voiced unaspirated stops word initially. However, unaspirated voiceless variants may occur as word final or word medial allophones of voiceless aspirated s tops . A pilot study with English monolinguals suggests that the subjects cannot divide the Hindi sounds into nativelike phonetic categories ( see Aggarwal,

PAGE 49

49 2011) , and that they would have to learn to attend to the Voice Onset Time (henceforth, VOT) to disc riminate the sounds into target like phonetic categories. In addition to this, a pretest for all the four L1 groups ensured that speakers of all these languages cannot already discriminate the Hindi target sounds prior to training. The target L2, Hindi, is an IndoEuropean language, spoken in most parts of India. It has 37 consonants and 11 vowels. Hindi exhibits a four way voicing contrast among stop consonants (Dutta, 2007), which is to say that voiced unaspirated, voiced aspirated, voiceless unaspirated and voiceless aspirated stops are phonemic. The fact that Hindi has a consonant inventory representing the full four way contrast of voicing and aspiration in stop consonants makes it a language suitable to be tested against. The laryngeal contrasts in the five L1s and the control group Hindi are summarized in Table 31. Since Mizo, Bodo, Tamil and English contain only a subset of these voicing and aspiration contrasts used in Hindi, they constitute a suitable set of languages to study the perceptual acqui sition of nonnative contrasts. Table 31 shows that in the word initial position Tamil allows only voiceless unaspirated stops, English and Bodo allow voiceless aspirated and voiced stops and Mizo allows voiceless aspirated, voiceless unaspirated and voic ed stops. In terms of the contrasts Tamil has the most to acquire to reach a Hindi like constraint ranking. Research M ethodology Participants The study involves speakers of four different L1s. Four are experimental groups : Tamil speakers, Mizo speakers, Bodo speakers, and English speakers . The fifth group is the control group of Hindi speakers. The control group consists of monolingual native

PAGE 50

50 Hindi speakers (NS) recruited from Jawaharlal Nehru University, New Delhi, India. The Mizo and Bodo speakers were re cruited from the northeast Indian states of Mizoram and Assam respectively. The Tamil speakers were recruited in Chennai, India, and English speakers were recruited from an undergraduate program at the University of Florida in the US. Each group had about 15 participants. To avoid the effect of language experience, all participants participated in a short interview before recruitment to make sure that they are monolinguals or have no prior language experience with any other language. By the term ‘monolinguals , ’ it is not intended that the speaker exists in isolation from other languages of the world, instead it means that the speaker might have been exposed to other languages in his/her surroundings by means of media, etc. , but for Tamil , Mizo, Bodo and English participants of this study, these people are dominant speakers and listeners of only those languages respectively. They have not had formal education or substantial exposure to another language that would be capable of altering the ir perceptual capabilities enough to help them or hinder them in perceiving Hindi laryngeal contrasts. This interview was carried out verbally with the help of an interpreter because all participants of Mizo, Bodo and Tamil were monolinguals and illiterate. The participants’ age range was limited to 1850 years so as to minimize effects of agerelated loss of perceptual abilities. Efforts were made to recruit comparable number of male and female participants. Stimuli This study used naturally produced stim uli so that the stimuli represented natural categories. All the tokens were recorded by a female native speaker of Hindi in a lab setting using a head mounted microphone on a Marantz DAT recorder that used a

PAGE 51

51 compact flash card with a sampling frequency of 48KHz. Three or four variants of each token were used in the experiment. After recording the stimuli, they were transferred from the compact flash card to a Toshiba Laptop. The speaker was 28 years old at the time of recording. The room was quiet and there was no outside noise. All tokens consisted of a CV syllable where the consonant was either voiceless unaspirated stop [p], voiceless aspirated stop [ph] , voiced aspirated stop [gh] or voiced unaspirated stop [g] . The vowel in each token was the same [a]. Care was taken that productions are consistent, including intonational contours. Falling intonation was not used in any of the tokens as it may have lowered the perceptual ability; instead, all the syllables were recorded in a list intonation. For instance, some sample pairings were as follows: [pa] ~ [ba] [ba]~[pa] [kha]~[kha] [ga]~[gha] In the above examples [p1], [p3] are variants of [p] ; [b2],[b3] are variants of [b] ; [kh 1], ]kh 3] are variants of [ kh] and so on. The vowel in all the tokens is kept the same [a] for consistency. The experiment was carried out in quiet rooms with a Toshiba laptop and headmounted microphone. The study consisted of four phases: pretest , Training, posttest , and retention. During training, participants in the experimental groups were trained to perceive Hindi voicing and aspiration contrasts using AX pairs where the subjects were expected to listen and judge the sameness or differentness of the two sounds. Pairs of CV syllables that contrasted in voicing or aspiration of the consonant were used in this study. The study looked at two places of articulation, in particular the bilabial stops and

PAGE 52

52 the velar stops, for the four training sessions and dental stops for the generaliz ation task. In each of the pretest s and posttest s the stimuli were presented with three different inter stimulus intervals, 1500ms 500ms and 250ms. This was done by adding 250ms, 500ms and 1500ms of silence respectively to the syllable duration of the fi rst token in every trial. This was based on the study by Werker and Tees (1984) , who both claimed that with an ISI of 1500ms a phonological level of processing the sounds is activated in the listener, while an ISI of 500ms activates a phonetic level of processing the sounds and an ISI of 250ms activates an auditory level of processing. It is also believed that prior linguistic experience with the language at study affects only the phonological level of processing, not the phonetic level. A phonetic mode of perception is thus believed to be a language general mode of perception (Burnham et al.1992; Werker and Tees 1983). The list below presents the stop consonants used in the stimuli and their average VOT values (in msecs) that were used in this experiment: / p/ /ph/ /b/ /bh/ /t/ /th/ /d/ /dh/ /k/ /kh/ /g/ /gh/ 15 75 90 78 14 73 89 84 16 80 88 85 Stimuli for the pretest The participants were made to listen to pairs of Hindi syllables such as ‘[pa]~[ba] or [pa]~[pha ]’ that differed only in voicing or aspiration or both. The pretest comprised of three blocks of trials, the first block containing stimuli with ISI=1500ms, the second consisting of stimuli with ISI=500ms, and the third block containing stimuli with ISI=25 0ms. Unless the subject pressed the S or K key, the experiment did not move to the next trial. ‘S’ Key stood for true and ‘K’ stood for false. This allowed the participants

PAGE 53

53 to listen and respond at their own pace and for the response time to be recorded for statistical analysis. The order in which these blocks were presented varied within participants and across participants. That is, no two participants were given the same order of ISI, and each participant was given a different ISI in every session. Befor e the pretest , the participants were given instructions (Appendix A) on what to listen for and were presented with eight practice trials involving similar CV syllable pairs but with different contrasts such as ‘[pa]~[na]’, so as not to give away the actual trials. It was made sure that the participants understood the task and correctly responded to at least seven out of the eight practice trials before they were given the pretest . The pretest helped ensure that participants were unable to hear all the L2 v oicing and aspiration contrast correctly/reliably before training. They were only able to perceive contrasts that existed in their respective L1s. The stimuli were alternated with 50% distractors. These distractors consisted of stimuli pairs that contrasted in features not relevant to this study. For instance, pairs like [ba]~[na] that contrasted in manner of articulation, [pa]~[la] that contrast in place and manner of articulation. The reason to alternate the distractors with stimuli was first, to make sure that the participants do not figure out that they are being tested on laryngeal contrasts , and second, the researcher can be sure that they can hear contrasts that are present in their L1. In total , the pretest consisted of three blocks of 126 pairs of stimuli each. The breakdown of the stimuli used in the pretest is given as follows (Appendix B): 1. Practice trials=8stimul us pairs=8 X 2=16 tokens 2. Block one=62 stimul us pairs +64 distractor pairs= 124 tokens +128 tokens=252 tokens. The tokens consis ted of bilabial and velar consonants that contrast in voicing, aspiration and both. Multiple tokens of each stimulus were used to make the stimuli as natural as possible. The stimuli in this block were separated by an ISI of 250ms.

PAGE 54

54 3. Block two=. 62 stimul us pairs +64 distractor pairs= 124 tokens +128 tokens=252 tokens. The tokens consisted of bilabial and velar consonants that contrast in voicing, aspiration and both. Multiple tokens of each stimulus were used. The stimuli in this block were separated by an ISI of 500ms. 4. Block three=62 stimuli pairs +64 distractor pairs= 124 tokens +128 tokens=252 tokens. The tokens consisted of bilabial and velar consonants that contrast in voicing, aspiration and both. Multiple tokens of each stimulus were used. The stimuli in this block were separated by an ISI of 1500ms. The order of the blocks was randomized in order to counter balance across participants. Stimuli for the training The training for each participant was conducted on four consecutive days. In every session, the pretest was followed by the training. The stimuli for training consisted of one block of 126 pairs of syllables. The set of stimuli used for training was same as that used for pretest but the ISI for the training stimuli was 1000ms. The ISI was chosen to be different from the three ISIs used for pretest so that the participants could be trained without any bias or preference of training them at a ‘phonological’, ‘phonetic’ or ‘auditory’ level. The aim of the experiment was to train the listeners to per ceive the laryngeal contrasts but not at any particular ISI. During the training session, the subjects listened to pairs of CV syllables varying in either voicing or aspiration of the consonant. After that, they were expected to respond, as quick l y and accurately as possible, by clicking ‘S’ or ‘K’ keys on the keyboard of the laptop. ‘S’ Key stood for same and ‘K’ stood for different. The choice of keys was made based on them being equidistant from the left and right edges of the keyboard so that t he reaction times should be comparable for a left handed or a right handed person. In this study all the participants were right handed.

PAGE 55

55 After they clicked a key, they received feedback on the screen. The feedback displayed which was the correct key for that particular response. The feedback stayed on the screen for 5 seconds. Stimuli for the posttest In each session, the training block was followed by a posttest . Stimuli for the posttest were the same as those used for the pretest . The posttest consisted of three blocks of 126 trials each, out of which 63 pairs were distractors. The blocks were randomized with respect to the order of ISI (250ms, 500ms, 1500ms). Stimuli for retention and generalization test Two weeks after the four training sessions were ov er, subjects of each language group were administered the retention cum generalization test. This test was done to see if participants remembered the contrasts they learned during the training sessions and whether they could generalize the training to perc eive voicing and aspiration contrast to another place of articulation, namely dental. This test consisted of a warm up session of eight trials, the retention test and a generalization test. The retention test was same as the posttest preceded by warm up tr ials. This delayed posttest will serve to test the retention of contrasts learnt by subjects during the previous sessions. The generalization test followed the posttest, and it consisted of 34 pairs of stimuli. Seventeen of these trials contained stimuli t hat contrasted in voicing and aspiration of dental stops, such as [ta]~[da] , and the remaining 17 trials were distractors. The test consisted of 17 trials so as to include multiple tokens of each laryngeal contrast. (Appendix B). The ISI used for the retention cum generalization test was 1000ms. An ISI of 1000ms was chosen so that the participants do not respond on the basis of their

PAGE 56

56 training on any particular ISI (250ms, 500ms or 1500ms). Their responses need to be unbiased irrespective of any ISI. Procedure The experiment was divided into four training sessions and retention cum generalization test. The training sessions were conducted on four consecutive days. The schedule of training is presented in T able 32. At the beginning of the first session, the participants from the four L1 groups , Mizo, Bodo, Tamil and English, were told that they were going to participate in a research perception experiment and that they will be asked to listen to sounds and click the key ‘ S ’ if they think the two sounds are sam e and ‘ K ’ if they think the two sound they heard were different. They were asked to focus on the actual sound of the syllable rather than extralinguistic features such as tone, volume, talker’s voice, etc. To make sure that these factors do not affect the perception, the same native speaker of Hindi recorded all the stimuli presented. The same talker will also eliminate any effects of talker variability (Logan et al. 1993). In order to familiarize them with the task, they were given practice or warm up trials. This was to ensure that they became accustomed to the kind of stimuli they were going to hear and that they understand the task. After they were able to do most of the practice trials correctly, they were given the actual training session. As elaborated before in this section, the training comprised of the pretest , followed by training, followed by the posttest . The total session lasted about 4550 minutes , and the participants had the option of taking a break at roughly after every 15 minutes. All the participants were tested on all the three ISIs in order to get more generalizable data. Two weeks after the training sessions were over, participants from all the four groups took the retention cum generalization test. Participants of the control group,

PAGE 57

57 Hindi, did not undergo the training sessions. They just took the pretest , posttest and the retention cum generalization test two weeks after the posttest . Analysis The results of the pretest s, posttest s and retenti on cum generalization tests were evaluated statistically using SPSS (version 21) . The various L1 groups , Mizo, Bodo, Tamil and English, were compared with the control group Hindi for percent correct identification and reaction times. Many ANOVAs were conducted to compare the performance of the different groups. Post hoc Bonferroni correction was also done for the main ANOVA. Repeated measures ANOVA was conducted with language as the between subjects factor and ISI as the within subjects factor at the pretes t to see if there is a main effect of ISI. Next, pretest and posttest results for percent correct identification and reaction times were compared for all language groups using repeated measures ANOVA. Results were also analyzed to see if there is an interaction between test time and ISI using repeated measures ANOVA with test time and ISI as withinsubjects factor and groups as the betweensubjects factor. Accuracy was compared for the three test times , pretest , posttest and retention test , using test time as the within group factor and language groups as the betweensubjects factor. Finally, the results of the pretest and the posttest are compared to the generalization test. One way ANOVAs were conducted to compare the performance of the groups at the prete st, retention test and generalization test. Some Ttests and ANOVAs were also conducted following the main ANOVA to compare the groups pairwise in order to find the source of significant difference. Results based on the analysis are reported in C hapter 4 f or percent correct identification and in C hapter 5 for reaction time.

PAGE 58

58 Table 31. Laryngeal contrasts in all language groups Language Number of phonological contrasts Phonemic contrasts (Vl, VlA, Vd, VdA) Examples Allophonic variation Hindi 4 Vl VlA Vd VdA /p/, /p h /, /b/, /b h / /p/ [p], /p h / [p h ], /b/ [b], /b h / [b h ] Tamil 1 Vl /p/ /p/ [p]/#_ [ ~w] /V_V [b]/nasal_ [pp]/word medially English 2 VlA Vd /k h /, /g/ /p/ [p h ]/#_, [p] elsewhere /b/ [b] Bodo 2 VlA Vd /p h /, /b/ /p h / [p h ], /b/ [b] Mizo 3 Vl VlA Vd /p/, /ph/,/b/ /p/ [p 0 ]/_#, [p] elsewhere /ph/ [ph] /b/ [b] Table 32. Schedule of Training Language Day1 Day2 Day3 Day4 Two weeks later experimental groups Mizo, Bodo, Tamil, English pretest , Training1, posttest pretest , Training2, posttest pretest , Training3, posttest pretest , Training4, posttest retention and generalization test control groupHindi pretest posttest retention and generalization test

PAGE 59

59 CHAPTER 4 PERCEPTION ACCURACY: PERCENT CORRECTION IDENTIFICATION The p ercent correct identification of the four Hindi voicing and aspiration contrasts voiceless unaspirated, voiceless aspirated, voiced unaspirated and voiced aspirated were obtained from participants of four different languages : Bodo, Mizo, English and Tamil. This data was obtained before and after training with the help of a pretest and a posttest . The pretest used for analyzing results is the pretest before the first training session and the posttest used for analyzing results is the posttest conducted after the fourth training session. The data was statistically analyzed using the Statistical Package for the Social Sciences (SPSS 21) and Microsoft Excel 2007. The data is tested for the effects of l anguage group (Bodo, Mizo, English, Tamil and control group H indi), t est t ime ( p retest and p os T test ), and i nter stimulus i nterval (250ms, 500ms and 1500ms). In this study, individuals whose scores fell more than two standard deviations (SD) below the group mean were considered outliers and were eliminated from the statistical analyses. A total of two outliers were removed from the data, one from the Bodo group and one from the Tamil group. The Hindi control group did not undergo training, but data for pretest and posttest was collected from them at the same time interval as other groups in order to maintain consistency with experimental groups . The control group pretest and posttest data was used to compare to the pretest and posttests of the experimental groups, respectively. One of the most prevalent hypotheses for the initial state in second language acquisition is that it begins with a full transfer or full manifestation of the L1 grammar. The development beyond the initial state is enabled by full access to the language acquisition device. This is the claim of the Full Transfer/Full Access hypothesis

PAGE 60

60 proposed by Schwartz and Sprouse (1996). In terms of Optimality theory Full Transfer means that the L2 learner begins with full manifestat ion of the hierarchy of constraints of his L1 and Full Access means that learners have access to the set of all constraints and a language acquisition device that helps them in ranking the constraints based on the L2 input. In OT, input is a set of acousti c descriptions provided by the auditory system and the perception grammar is a languagespecific hierarchy of perception constraints that evaluate the candidate set. The winning candidate is the phoneme category that is perceived. Based on this definition of Full Transfer/Full Access, the differences between Hindi and the L1s in this study will have consequences for perception by these L2 learners. For instance, given that Tamils have only one phonemic contrast while Hindi has four, it is hypothesized that Tamil learners will perceive the Hindi contrasts noncategorically. However, it is hypothesized that after training with Hindi contrasts the performance of all the L2 learner groups on the identification task will become closer to that of the Hindi control group. In this chapter, the results of the data analysis are reported. Comparisons are made between the pretest and the posttest mean percent correct identification. Both the pretest and the posttest data for the five language groups (Bodo, Mizo, English , Tamil and control group Hindi) is compared. The analysis also looks at the main effect of the different ISIs (250ms, 500ms and 1500ms) at pretest and posttest. Next, the effect of the training on the perception of participants is reported by comparing pr etest with posttest among all five language groups. Following this, the results of the generalization test comparing all the five groups will be reported. The comparison

PAGE 61

61 between pretest, posttest and retention test that was conducted two weeks after traini ng will also be reported in this chapter. Main ANOVA A repeated measures ANOVA with language groups as the betweensubjects factor and t est t ime ( pretest , posttest , retention and generalization test) and ISI (250ms, 500ms, 1500ms) as the withinsubjects f actor revealed a significant main effect of t est t ime [ F (3,140) = 760.21 , p = .005)], a significant main effect of ISI [ F (2,140) = 130.42 , p = .000], a significant main effect of group [ F (4,70) = 1200.42 , p = .000]. Apart from these, significant interactions were also found between group and ISI [ F (8,140) = 80.02 , p = .000], between group and t est t ime [ F (12,140) = 218.65 , p = .000], between ISI and t est t ime [ F (6,280) = 50.15 , p = .000] and between ISI, test time and group [ F (16,280) = 78.98 , p = .000]. The Main ANOVA was further analyzed by breaking it down into smaller ANOVAs and t t ests. In the following sections , the results have been analyzed by comparing (i) groups at pretest, (ii) groups at pretest vs posttest, (iii) groups at pretest, posttest vs retention and (iv) groups at generalization test. Pre Test The mean percent correct identification of Hindi voicing and aspiration contrasts for the five groups Hindi (n=15), Bodo (n=15), Mizo (n=15), English (n=15) and Tami l (n=15) at three different ISIs: 250ms, 500ms and 1500ms were obtained during pretest. The results are reported in Table 412 and shown in Figure 41. The results show that the participants of the control group Hindi scored highest in comparison to all ex perimental groups on all the three ISIs. It was observed that all the 2 All the tables in chapter 4 and 5 present Language columns in order of the number of contrasts they contain. Tamil, Bodo, English, Mizo and Hindi have 1,2,2,3 and 4 contrasts respectively.

PAGE 62

62 experimental groups and the control group performed best at 250ms, except Tamil who did better at 500ms while the mean correct response was lowest when the ISI was 1500ms. Table 42 and Figure 42 show the mean percent correct identification of all the five l anguage groups at the three different ISIs . A one factor ANOVA with l anguage group (Hindi, Bodo, Mizo , English and Tamil) as the betweensubjects factor was carried out. It gave a significant main effect of group at the pretest [ F (4,70) = 60.99, p = .000]. Following this pair , wise T test s were conducted on Hindi , and each l anguage group to see if each group is significantly different from the control Hindi group before any training i s given to them. The results revealed that each experimental group was significantly different from the control group at the beginning of the experiment, prior to training. Hindi mean = 94% vs Tamil Mean = 65.67% [t (14) = 42.5 , p = .001] Hindi mean = 94% vs Bodo Mean = 78% [t (14) = 4.98 , p = .038] Hindi mean = 94% vs English Mean = 85.33% [t (14) = 9.87 , p = .01] Hindi mean = 94% vs Mizo Mean = 68.67% [t (14) = 28.73 , p = .001] A repeated measures ANOVA was conducted with l anguage groups (Hindi, Bodo, MIzo, English and Tamil) as the betweensubjects factor and ISI (250ms, 500ms and 1500ms) as the withinsubjects factor. On pretest scores , t he main effect of Inter Stimulus Interval was not found to be significant [ F (2, 140) = 2.73, p = .123] but there was a main effect of group [ F (8, 140) = 3.83, p = .000] and no interaction between groups and ISI [ F (4,70) = 2.48, p = .173]. Further, pairwise comparison using paired t tests indicated that the main effect of groups was due to significant difference between the BodoHindi, Tamil Hindi, English Hindi and Mizo Hindi groups

PAGE 63

63 To summarize, the results from the pretest indicate the control group Hindi outperformed all the other e xperimental groups in the study. The languag e groups Tamil, Mizo, Bodo and English performed significantly lower than the Hindi participants. Effects of Tr aining The effectiveness of training was evaluated on the basis of the performance of the language groups on four tests: pretest , posttest , retention test and generalization test. The mean percent correct identification will be compared for all the groups before the training, during pretest and after the training at posttest . Then the pretest , posttest and the retention test (that was conduct ed two weeks after training) will be compared together for all the groups. Next, the results of the generalization test will be compared among the five groups and the generalization test will also be compared with the pretest and posttest. P retest V s . P ost test The mean percent correct identification for the five groups , Hindi (n=15), Tamil (n=15), Bodo (n=15), Mizo (n=15) and English (n=15) , were compared at three different i nter stimulus intervals , 250ms, 500ms and 1500 ms , at two times pretest , pretest and posttest . The results of the pretest and posttest are compared and presented in Table 43 and shown in Figure 43. It can be seen from the comparison that all the four experimental groups (Bodo, Mizo, Tamil and English) performed better on the posttest than on the pretest on all the three ISIs, 250ms, 500ms and 1500ms. During the pretest , the difference between the experimental groups and the control Hindi group was significant, but the posttest comparisons show that the difference between the experimental groups and control group was not significant.

PAGE 64

64 The data from the different groups was submitted to repeated measures ANOVA with L l nguage group as the betweensubject factor and t est t ime ( pretest and posttest ) and ISI (250ms, 500ms and 1500ms) as the w ithin subjects factors. The analysis yielded main effect of ISI [ F (1,70) = 420.34 , p = .000], main effect of test time [ F (1,70) = 1534.25 , p = .000], main effect of group [ F (4, 70) = 2619.49, p = .000], a significant interaction between ISI and group [ F (8, 140) = 137.15, p = .000], significant interaction between test time and group [ F (1, 70) = 260.26, p = .000], significant interaction between ISI and test time [ F (2, 140) = 13.27, p = .000], significant interaction between ISI, test time test time a nd group [ F (8, 140) =112.87, p = .000]. The statistical results show that the accuracy of participants from all the language groups improved significantly from pretest to posttest with most accurate at 250ms and least accurate at 1500ms. P retest , P osttest and R etentionT est Comparison The mean percent correct identification at three different times , before training ( pretest ), immediately after training ( posttest ) and two weeks after training ( r retention test), were compared. The retention test was identical to the posttest . Table 4 4 and Figure 44 show a comparison of mean percent correct scores of all five language groups at the retention test. The retention test was given to see if the perceptual discrimination skills the participants had gained during training were retained at a later time about two weeks from the time of training. The pretest , posttest and retention test scores were compared for the five language groups (Hindi, Tamil, Bodo, Mizo and English). Table 45 and Figure 45 show the comparison of the five groups at three test times ( pretest , posttest and retention test).

PAGE 65

65 As shown in Table 43 and Figure 43 , all experimental groups performed better on posttest than on the pretest . Further , a repeated measures ANOVA with language groups as the betweensubjects factor and t est t ime ( pretest , posttest and retention test) and ISI (250ms, 500ms, 1500ms) as the withinsubjects factor revealed a significant main effect of t est t ime [ F (2,140) = 855.41 , p = .000)], a significant main effect of ISI [F (2,140) = 195.24 , p = .000], a significant main effect of group [ F (4,70) = 1415.98 , p = .000]. Apart from these, significant interactions were also found between gr group and ISI [F (8,140) = 90.67, p = .000], between group and t est t ime [ F (8,140) = 326.16 , p = .000], between ISI and test time [ F (4,280) = 57.05 , p = .000] and between ISI, test time and group [ F (16,280) = 98.99 , p = .000]. To find the exact source of significant difference between the scores at three t est t imes, pair wise comparisons were done between pretest and retention test and posttest and retention test. A repeated measures ANOVA with language groups as the betweensubjects factor and t est t ime ( pretest and retention test) and t est t ime ( posttest and retent ion test) as the withinsubjects factor showed significant main effects of t est t ime [ F (1,140) = 16.11 , p = .005] for pretest vs. retention test and [ F (1,140) = 21.34 , p = .000] for posttest vs. retention test, respectively. The results indicate that t he significant main effect of t est t ime was due to a significant difference between both the pretest and the retention test (mean=78.76% vs. 83.32%) and the posttest and the retention test (mean=88.01% vs. 83.32%). Post hoc pairwise comparison using paired T test between pretest and retention scores of each group revealed a significant difference (p<.05): Tamil mean=65.67% vs

PAGE 66

66 55.17%, Bodo mean= 78% vs. 90.67%, English mean= 85.33% vs 90.63% and Mizo mean= 68.67% vs. 83.5%. Post hoc pair wise comparisons using paired Tt est between posttest and retention of each group showed that the significant difference between the p osttest and retention test scores arise due to two groups, Tamil (t (14)=9.58, p=.005) and Mizo (t (14)=4.61, p=.02). There was no significant difference within the scores of Bodo (t (14)=1.32, p=.16) and within English (t (14)=1.92, p=.10) at the p osttest and the retention test. A comparison of the mean scores of different groups at the retention test at the three ISIs 250ms, 500ms and 1500ms was carried out using an ANOVA. Although the main effect of group was found to be significant [ F (4,70) = 164.6 , p = .000] , there was no main effect of ISI [ F (2,140) = .75 , p = .50]. To sum up, the results from the pretest , posttest and retention test indicate that there was a significant change in all the experimental groups Bodo, Mizo, Tamil and English immediately after the training. All groups performed significantly better on the posttest than the pretest . Two weeks after the posttest, during the retention test, there was no significant improvement in the mean percent correct identification of the experimental groups Bodo, Mizo, Tamil and English. However , the mean scores of Bodo, English and Mizo improv ed significantly from pretest to retention test. G eneralization T est The effectiveness of the training was also determined on the basis of the participants’ performance on a generalization test. All the earlier stimuli during training, pretest , posttest and retention tests presented stimuli from either bilabial or velar place of articulation. The generalization test used stimuli belonging to a new place of

PAGE 67

67 articulation, namely, Dental. It presented all the stimuli at an ISI of 1000ms instead of the three IS Is used earlier for pretest s and posttests, 250ms, 500ms and 1500ms. The generalization test examined whether the participants can extend the effect of training to new stimuli. The g eneralization test was administered two weeks after training. Mean percen t correct identification for all the five groups are analyzed and reported in Table 46 and Figure 46. Between groups The data from the generalization test was submitted to an ANOVA for a single factor with group (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor. The results show that the main effect of group is significant [ F (4,70) = 69.77 , p = .000]. Further post hoc Ttests were done on control group Hindi and each experimental group, pairwise, to see if the difference between each of the pairs is significant or not. The Ttests indicate that the significant main effect of groups arise due to Tamil group only [ t (14)= 9.53, p=.005]. The difference between the mean scores of Hindi ad Bodo (t (14)=1.30, p=.21), Hindi and Mizo (t(14)=.39, p=.38) and Hindi and English (t(14)=.69, p=.28) were not found to be significant. The mean percent correct identification scores of all the groups have been shown in Table 46 and Figure 46 along with their standard deviation and standard err or. P retest and generalization test comparison The mean percent correct identification for the g eneralization test and the pretest were submitted to a repeated measures ANOVA with group (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor and t est t ime ( pretest vs. generalization test) as the withinsubjects factor. The results showed that there is a significant main effect of

PAGE 68

68 group [ F (4,70) = 121.72 , p = .000] and a significant main effect of t est t ime [ F (1,70) = 56.14 , p = .000]. The interaction between group and t est t ime was also found to be significant [ F (4,70) = 46.27 , p = .000]. Upon further pair wise analysis using Ttests, it was found that t he source of interaction were all groups at p retest and generalization (Bodo p=.02, Miz o p=.001, Tamil p=.02) , except English (p=.10). The mean percent correct identification scores have been shown in Table 47 and Figure 47 along with their standard deviation and standard error. P osttest and generalization test comparison The mean percent correct identification for the generalization test and the p osttest were analyzed using repeated measures ANOVA with group (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor and t est t ime ( posttest vs. generalization test) as the withinsubjects factor. The results showed that there is a significant main effect of group [ F (4,70) = 59.54 , p = .000] and a significant main effect of t est t ime [ F (1,70) = 7.12 , p = .024]. There was also a significant interaction between group and t est t ime [ F (4,70) = 20.92 , p = .000]. The results of the ANOVA were further subjected to pairwise comparison using paired Ttests for each l anguage group. The results showed that there is a significant difference in the scores at posttest and generalization test for two experimental groups , Tamil (p=.004) and Bodo (p=.03). The scores were not significantly different for English group (p=.12) , whereas the scores were near significant for Mizo group (p=.06). Furthermore, a repeated measures ANOVA with groups as the betweensubject factor and t est t ime ( pretest , p osttest and generalization ) as the withinsubjects factor showed a significant main effect of group [ F (4,70) = 65.69 , p = .000], a significant main

PAGE 69

69 effect of t est ti me [ F (2,140) = 55.93 , p = .000] and a significant interaction between groups and the three t est t imes [ F (8,140) = 29.73, p = .000]. To sum up the findings of the generalization test, the experimental groups Bodo and Mizo showed significant improvement from pretest to g eneralization tes t (Bodo mean= 78% vs 97% and Mizo mean= 68.67% vs. 90.67%). They were able to generalize the effect of training to new kind of stimuli (belonging to dental place of articulation).No significant improvement was seen in the English (mean=85.33% vs. 89%) and Tamil (mean=65.67% vs. 60%) groups. There was also a significant improvement from p osttest to generalization test for the Bodo group (mean=86.57% vs. 97%) while no significant difference was found between the posttests and g eneralization tests of English (mean= 92.63% vs. 89%) and Mizo (mean= 92.5 vs. 91%). Note that there was no significant difference between the posttest and generalization test scores of the control group Hindi ; also (mean= 94% vs. 90%). English and Bodo groups did equally well on p ostte st and generalization test. However, the scores for Tamil group (mean= 77.27% vs. 60%) saw a significant drop from posttest to g eneralization test that took place two weeks later. This indicates that Tamil was the only group that was unable to generalize t he effects of training to new stimuli. All the other groups Mizo, Bodo and English were able to generalize the discrimination skills they learnt during training to new stimuli. Percent Correct Identification of C ontrasts Apart from comparing the percent correct identification and reaction times across the five different languages at different times and at different ISIs, an analysis of the order of acquisition of contrasts was also done. The participants’ responses to all the stimuli were sorted according t o whether the stimulus pair contrasted in voicing or

PAGE 70

70 aspiration or both. This sorting of stimuli was done for both velar and bilabial stimuli separately. The accuracy of responses for each contrast was recorded at pretest, posttest, retention test and generalization test. The percent correct identification for each contrast was calculated. To begin with, Tamil had only voiceless unaspirated, Bodo and English had voiced unaspirated and voiceless aspirated and Mizo had voiceless unaspirated, voiceless aspirat ed and voiced unaspirated. After training a ll groups, Tamil, Bodo, Mizo and English showed a significant improvement from pretest to posttest (p<.05) in the identification of voicing and aspiration contrasts. The Tamil group improved from 73.7% to 84.2% in identifying the [ ka ] [ kha ] distinction. There was also a significant improvement in identification of [ k ] [ g ] (64% to 80%), [ p ] [ pha ] (75% to 81%) and [ b ] [bha ] (69.4% to 82%) contrasts. The least improvement was seen in identifying the [ ga] [ gha ] contrast (52.6% to 64%). The [ ga] [ gha ] contrast was also the least identifiable by the Tamil participants before training. There was no significant improvement in the perception of [ pha ] [ bha ] , [ p ] [ bha ] and [ b ] [ pha ] distinctions. The Mizo group also showed significant improvement in hearing the contrast between voiced unaspirated and voiced aspirated sounds: 64% to 83.5%for [ b ] [ bha ] contrast and 67% to 80% for [ g ] [ gha ] contrast. However , there was no significant improvement in the perception of any other contrast. This can be explained by the fact that all the other contrasts were already present in Mizo speaker’s L1 and hence they already had nativelike performance during the pretest and posttest on those contrasts. The Bodo speakers’ performance s did not change significantly for [ ka ] [ kha ] and [ p ] [ pha ] contrasts ; however , they showed significant improvement on all other voicing

PAGE 71

71 and aspiration contrasts, the most significant improvement being in the identification of voicing contrasts (70% to 84%). Similar to the Bodo speakers, the English speakers showed significant improvement in their percent correct identification of voicing contrasts [ p ] [ b ] , [ k] [ g ] , [ pha ] [ bha ] and [ kha ] [ gha ] . No significant improvement was seen in the identification of aspiration contrasts [ ga ] [ gha ] (70% to 73%) and [ ka ] [kha ] (74% to 78.2%). It was noted that all groups performed better on posttest than on the pretest when the stimuli contrasted in both voicing and aspiration. However, when the posttest was compared to the retention test and generalization test after two weeks, using repeated measures ANOVA no significant difference was found in the percent correct identification. An exception to that is the Tamil group that showed a significant decline in its performance from posttest to retention and generalization test. All others group did better than the pretest but almost similar to the posttest. The above results indicate that the Tamil group was faster at acquiring the voicing contrast followed by the aspiration contrast, although this acquisition was not long lasting. The Bodo and Engli sh groups were also better at perceiving voicing contrasts before they could perceive aspiration contrasts among stop consonants. Apart from the laryngeal contrasts , a discussion of the performance on the distractors is important. All the stimuli in pretest, posttest, retention test and generalization were alternated with distractors. The distractors had almost an equal number of same and different tokens (8 same and 9 different). These distractors were contrasts that were already present in the participants ' L1s but not the ones that they were going to be trained on (voicing and aspiration contrasts). They contrasted in either

PAGE 72

72 nasality, place of articulation or manner of articulation. All the groups performed well on these distractors with more than 96% ac curacy, indicating (i) they understood what the experiment required them to do and (ii) they were able to hear the contrasts that already existed in their languages. Summary of Results for Percent Correct I dentification The above analyses concluded that al l e xperimental groups Mizo, Bodo, Tamil and English should show a significant improvement with training. They all performed significantly better on the posttest than on the pretest . Further, English and Bodo groups retained the effect of training two weeks after the training. Moreover , all groups except Tamil made a significant improvement from p retest to retention test. Tamil group is different in that it did better on the 500ms ISI in the pretest ; that is, they processed the contrasts at the phonetic level. The Tamils were better at processing at the phonetic level than the other groups. After training, the Tamils performed well on the posttest immediately after training because they continuely process ed the new contrasts phonetically but at the time of retention and generalization test, since a deeper phonological processing was not in place for the Tamils , they performed worse than other language groups. They performed significantly better two weeks after training than they did before training. For the generalization test, all experimental groups except Tamil were able to generalize the effects of training to new stimuli two weeks after training. Table 48 to Table 412 summarize the significant results of percent correct identification from all the tests in this chapter.

PAGE 73

73 Figure 41. Mean percent correct identification of Hindi , Tamil, Bodo, English and Mizo groups at the pretest, all ISIs combined Figure 42. Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three ISIs 250ms, 500ms and 1500ms. 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Hindi Tamil Bodo English Mizo% Correct identification Group Pretest 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Hindi Tamil Bodo English Mizo% Correct Identification Group Pretest at different ISIs 250ms 500ms 1500ms

PAGE 74

74 Figure 43. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at pretest and posttest Figure 44. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the retention test 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Hindi Tamil Bodo English Mizo% Correct Identification Group Pretest vs Posttest Pretest Posttest 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Hindi Tamil Bodo English Mizo% Correct Identification Group Retention Test

PAGE 75

75 Figure 45. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest , posttest and retention test Figure 46. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Hindi Tamil Bodo English Mizo% Correct Identification Group Pretest vs Posttest vs Retention Test Pretest Posttest Retention test 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Hindi Tamil Bodo English Mizo% Correct Identification Group Generalization Test

PAGE 76

76 Figure 47. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the pretest, posttest and generalization test 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Hindi Tamil Bodo English Mizo% Correct Identification Group Pretest vs Posttest vs Generalization test Pretest Posttest Generalization test

PAGE 77

77 Table 41. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the pretest with their Standard deviation and Standard Error. Hindi Tamil Bodo English Miz o Mean 96.1 65.7 78 85.3 68.7 SD 11.0 11.2 15.6 11.5 11.5 SE 6.4 6.4 9.0 6.7 6.7 Table 42. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three ISIs 250ms, 500ms and 1500ms with Standard deviation and Standard Error. Hindi Tamil Bodo 250ms 500ms 1500ms 250ms 500ms 1500ms 250ms 500ms 1500ms Mean 96.4 97.6 95.2 64 66.7 65.1 84.2 78.6 72.2 SD 5 7.3 6 11.5 19.2 14.6 14.2 14.8 18.3 SE 3.5 5.2 4.2 8.1 13.6 10.3 10.0 10.5 12.9 Table 42. Continued English Mizo 250ms 500ms 1500ms 250ms 500ms 1500ms Mean 87 86.1 85.7 70 69 65.1 SD 14.2 14.6 14.6 12 17.3 16.3 SE 10.0 10.3 10.3 8.5 12.2 11.5 Table 43. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at pretest and posttest with Standard deviation and Standard Error. Hindi Tamil Bodo English Mizo Pre Post Pre Post Pre Post Pre Post Pre Post Mean 96.1 94 65.7 77.4 78 85.7 85.3 92.1 68.7 90.8 SD 11.0 12.7 11.2 14.7 15.6 14.8 11.5 11.0 11.5 13.4 SE 6.4 7.3 6.4 8.5 9.0 8.5 6.7 6.4 6.7 7.7 Table 44. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the retention test with Standard deviation and Standard Error. Hindi Tamil Bodo English Mizo Mean 96.6 55.2 90.7 90.6 83.5 SD 11.4 13.3 12.5 12.0 14.3 SE 6.6 7.6 7.2 6.9 8.2

PAGE 78

78 Table 45. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest, posttest and retention test with Standard deviation and Standard Error. Hindi Tamil Bodo Pre Post Ret Pre Post Ret Pre Post Ret Mean 96.1 94 96.6 65.7 77.4 55.2 78 85.7 90.7 SD 11.0 12.7 11.4 11.2 14.7 13.3 15.6 14.8 12.5 SE 6.4 7.3 6.6 6.4 8.5 7.6 9.0 8.5 7.2 Table 45. Continued English Mizo Pre Post Ret Pre Post Ret Mean 85.3 92.1 90.6 68.7 90.8 83.5 SD 11.5 11.0 12.0 11.5 13.4 14.3 SE 6.7 6.4 6.9 6.7 7.7 8.2 Table 46. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test with Standard deviation and Standard Error. Hindi Tamil Bodo English Mizo Mean 92.5 60 96 90.5 91.5 SD 13.5 11.4 11.4 12.12 11.7 SE 9.5 8.1 8.1 8.6 8.3 Table 47. Comparison of Mean percent correct identification of Hindi, Tamil, Bodo, English and Mizo groups at the pretest, posttest and generalization test with Standard deviation and Standard Error. Hindi Tamil Bodo Pre Post Gen Pre Post Gen Pre Post Gen Mean 96.1 94 92.5 65.7 77.4 60 78 85.7 96 SD 11.0 12.7 13.5 11.2 14.7 11.4 15.6 14.8 11.4 SE 6.4 7.3 9.5 6.4 8.5 8.1 9.0 8.5 8.1 Table 47. Continued English Mizo Pre Post Gen Pre Post gen Mean 85.3 92.1 90.5 68.7 90.8 91.5 SD 11.5 11.0 12.12 11.5 13.4 11.7 SE 6.7 6.4 8.6 6.7 7.7 8.3 Table 48. Summary of significant results for percent correct identification at pretest Factors tested F value P value group (4,70) = 60.99 p = .000 group*ISI ISI (2, 140) = 7.042 p = .022

PAGE 79

79 Table 49. Summary of significant results for percent correct identification at pretest vs posttest Factors tested F value P value ISI (1,70) = 420.34 p = .000 test time (1,70) = 1534.25 p = .000 group (4, 70) = 2619.49 p = .000 ISI* group (8, 140) = 137.15 p = .000 test time* group (1, 70) = 260.26 p = .000 test time*ISI (2, 140) = 13.27 p = .000 test time * group (8, 140) =112.87 p = .000 Table 410. Summary of significant results for percent correct identification at pretest vs posttest vs retention test Factors tested F value P value ISI (2,140) = 195.24 p = .000 test time (2,140) = 855.41 p = .000 group (4,70) = 1415.98 p = .000 ISI* group (8,140) = 90.67 p = .000 test time* group (8,140) = 326.16 p = .000 test time*ISI (4,280) = 57.05 p = .000 test time * group (16,280) = 98.99 p = .000 Table 411. Summary of significant results for percent correct identification at retention Test Factors tested F value P value group (4,70) = 164.6 p = .000 group* test time (pretest vs retention) test time (1,140) = 16.11 p = .005 group* test time (posttest vs retention) test time (1,140) = 21.34 p = .000 Table 412. Summary of significant results for percent correct identification at generalization Test Factors tested F value P value group (4,70) = 69.77 p = .000 group X test time (pretest vs generalization) group (4,70) = 121.72 p = .000 test time (1,70) = 56.14 p = .000 group* test time (4,70) = 46.27 p = .000 group X test time (posttest vs generalization) group (4,70) = 59.54 p = .000 test time (1,70) = 7.12 p = .024 group* test time (4,70) = 20.92 p = .000 group X test time (pretest vs posttest vs generalization) group (4,70) = 65.69 p = .000 test time (2,140) = 55.93 p = .000 group* test time (8,140) = 29.73 p = .000

PAGE 80

80 CHAPTER 5 REACTION TIME Introduct ory Remarks Along with the percent correct identification of the four Hindi voicing and aspiration contrasts voiceless unaspirated, voiceless aspirated, voiced unaspirated and voiced aspirated, reaction time data was also obtained from participants of the four differe nt languages : Bodo, Mizo, English and Tamil. This data was obtained before and after training on their performance s on a pretest and a p osttest. The data was statistically analyzed using the Statistical Package for the Social Sciences (SPSS 21). The data w as tested for the effects of l anguage group (Bodo, Mizo, English, Tamil and control group Hindi), t est t ime ( pre test and post test), and Inter Stimulus Interval (250ms, 500ms and 1500ms). Just like the percent correct identification, individuals whose score s fell more than two standard deviations (SD) below the group mean would have been considered outliers; however, there were no outliers in any of the groups for reaction time. Although the Hindi control group did not undergo training, reaction times (in mi lliseconds) for the Hindi group participants were collected for the pretest and posttest at the same time as it was collected for other groups. The control group p retest and p osttest data was used to compare to the p retest and p osttests of the e xperimental groups, respectively, in case reaction time changed only as a result of retesting. In this chapter , the results of the reaction time data analysis are reported. Comparisons are made between the p retest and the p osttest mean reaction times (in millisecond s). Both the p retest and the p osttest data for the five language groups (Bodo, Mizo, English, Tamil and control group Hindi) are compared. The analyses also

PAGE 81

81 examine the main effect of the different ISIs (250ms, 500ms and 1500ms) at p retest and p osttest. Ne xt, the effect of the training on the perception of participants is reported by comparing reaction times for the p retest with p osttest among all five language groups. Following this, the results of the generalization test comparing reaction time for all the five groups will be reported. The comparison among the reaction times during pretest , posttest and retention test that was conducted two weeks after training will also be reported in this chapter. Main ANOVA The reaction times of the participants were analyzed for correct and incorrect responses separately. None of the groups showed any significant difference in their reaction times based on the type of response (correct or incorrect). Following this , a repeated measures ANOVA with group as the betweensubjects factor (Hindi, Bodo, Mizo, English, Tamil) and ISI (250ms, 500ms, 1500ms) and t est t ime ( pretest , posttest , retention test and generalization test) as the within subjects factor was conducted. The results show that there is a significant main effect of t est t ime ( pretest , posttest, retention and generalization test) [ F (3,70) =67.16, p = .009] and a significant main effect of group [ F (4,70) =6.76, p = .005]. However, there was no main effect of ISI. There was also a significant interaction between group and test time [F (12,140) = 5.05 , p = .008]. No significant interaction between group and ISI, ISI and test time and ISI, t est t ime and group were found. The Main ANOVA was further analyzed by breaking it down into smaller ANOVAs and T t ests. The results have been analyzed by comparing groups at pretest, groups at pretest vs posttest, groups at pretest, posttest vs retention and groups at generalization test.

PAGE 82

82 Pret est The reaction time of Hindi voicing and aspiration contrasts for the fiv e groups Hindi (n=15), Bodo (n=15), Mizo (n=15), English (n=15) and Tamil (n=15) during the p retest are reported in Table 5 1 and shown in Figure 51. The reaction times of different language groups at three different ISIs , 250ms, 500ms and 1500ms , were obtained during pretest . The results are reported in Table 52 and shown in Figure 52. Further , the mean reaction times for each Inter Stimulus Interval , 250ms, 500ms and 1500ms , are analyzed separately. The results show that the participants of all the groups take longer to respond to the stimuli when the Inter Stimulus Interval is 1500ms , and they take the least time to respond when the ISI is 250ms. The mean reaction times obtained by all language groups at the three ISIs are reported in Table 52 and shown in F igure 52. It was found that Hindi control group takes longer to respond at 250ms than at 1500ms , but this time difference was not significant (p<.05). Also, all the groups take less time at 500ms than at 1500ms to decide if they heard a particular laryngeal contrast. A one factor ANOVA with l anguage g roup (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor was carried out for reaction times at p retest. It gave a significant main effect of group [ F (4,70) = 10.66, p = .001]. Followi ng this, pair wise T tests were conducted on Hindi and each l anguage group to see if each group is significantly different from the control Hindi group before any training is given to them. The results revealed that each e xperimental group except Bodo (mea n=1511.31msec [t (14) = .067 , p = .49]) was significantly different from the control group at the beginning of the experiment, in terms of reaction time. Hindi mean =

PAGE 83

83 1518.69msec vs Tamil Mean =1055.357msec [t (14) = 3.57 , p = .011]; vs English Mean = 997.97msec [t (14) = 5.70 , p = .005]; vs Mizo Mean = 1065.37msec [t (14) = 3.61 , p = .011] Next, repeated measures ANOVA was conducted with l anguage groups (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor and ISI (250ms, 500ms and 1500ms) as the withinsubjects factor for the p retest reaction times. Although the main effect of Inter Stimulus Interval was found to be significant [ F (2, 140) = 5.54, p = .034] , there was no main effect of g roup [ F (4, 70) = 2.73, p = .150] and there was no significant interaction between the g roups and ISI F (8, 140) = 1.41, p = .353] To summarize the results from the pretest , the analysis indicates that the control group Hindi and Bodo took almost the same amount of time to respond to the stimuli at all th ree Inter Stimulus Intervals , 250ms, 500ms and 1500ms. All the other experimental groups , Tamil, Mizo and English, take significantly less reaction time than control Hindi. Also , there was no significant interaction between groups and ISI during the pretest . Effects of Training Apart from the percent correct accuracy, the effectiveness of training was evaluated on the basis of the reaction time of the language groups on four tests: p retest, p osttest, r etention test and g eneralization test. The reaction time will be compared for all the groups before the training, during p retest and after the training, at p osttest. Then the p retest, p osttest and the r etention test will be compared together for all the groups. Next, the results of generalization test will be compared among the five groups , and the generalization test will also be compared with the pretest and p osttest.

PAGE 84

84 P retest vs. P osttest The reaction time for the five groups , Hindi (n=15), Tamil (n=15), Bodo (n=15), Mizo (n=15) and English (n=15) , were compared at three different Inter stimulus intervals , 250ms, 500ms and 1500ms , at two times , p retest and p osttest. This was done via a repeated measures ANOVA with g roup as the betweensubjects factor (Hindi, Bodo, Mizo, English, Tamil) and ISI (250ms, 500ms, 1500ms) and t est t ime ( p retest vs p osttest) as the within subjects factor. The results show that there is a significant main effect of ISI [F (2,140) =17.76, p = .006] and a significant main effect of t est t ime ( pretest vs posttest ) [F (1,70) = 21.19 , p = .006]. However, there is no significant effect of group [ F (4,70) = 3.537 , p = .099], no significant interaction between ISI and group [ F (8,140) = 1.185 , p = .399], no significant interaction between t est t ime and group [ F (4,70) = .915 , p = .521] , no significant interaction between ISI and test time [ F (2,140) = .826 , p = .438] and no significant interaction between ISI, t est t ime and group [ F (8,140) = 1.154 , p = .419]. Multiple comparisons using Bonferroni post hoc method yielded no main effec t of any group. Table 5 3 and Figure 53 show a comparison of the mean reaction times at pretest and p osttest at the three ISIs. The ANOVA results indicate that all the experimental groups (Bodo, Mizo, Tamil and English) took less time to respond to stimuli on the p osttest than on the pretest on all the three ISIs, 250ms, 500ms and 1500ms. Interestingly, the Hindi control group speakers also reacted faster on the posttest than on the pretest without any training. As seen in Table 51 , there is a significant difference in the reaction times of the control group Hindi and the reaction times of Tamil, Mizo and English during the pretest . The difference between the reaction times of control Hindi and Mizo [t (14) = 1.356, p = .102] was not significant during the p osttest. This shows that the Mizo group became more

PAGE 85

85 like the control group Hindi in their response times after training. For the other experimental groups, the reaction time continued to remain different from that of Hindi even after training. Even though the posttest reaction times were significantly different between groups [ F (4,70) = 4.8 , p = .005], the mean reaction times of the experimental groups were lower than that of Hindi at the posttest . The mean reaction times for all groups at p osttest are presented in Table 53 and F igure 53. Further , the data was submitted to a repeated measures ANOVA with l anguage group as the betweensubject factor and t est t ime ( pretest and posttest ) as the within group factor. The analysis yielded main effect of t est t ime [ F (1,70) = 62.96 , p = .000] and main effect of group [ F (4,70) = 47.17 , p = .000]. There was also a significant interaction between g roup and t est t ime ( p retest and posttest ) F (4,70) = 11.44 , p = .000]. However , ANOVA results for group as the betweensubjects factor and Inter Stimulus Interval as the withinsubjects factor revealed no significant main effect of ISI [ F (2,140) = .51 , p = .61] but a significant main effect of group [ F (4,70) = 7.01 , p = .002]. No significant interaction between group and ISI was found F (8,140) = 8.15 , p = .99] The statistical results show that the reaction time of participants from all the experimental groups decreased significantly after training, irrespective of the Inter S timulus Interval used to present the stimuli. P retest , P osttest and R etention Test C omparison The reaction times at three different times: before training ( p retest), immediately after training ( posttest ) and two weeks after training ( rretention test) were compared. The retention test was identical to the p osttest. Table 5 4 and Figure 54 show a comparison of reaction times of all the five language groups at the retention test.

PAGE 86

86 The reaction time at r etention test was measured to see if the decrease in the time taken to respond to stimuli is retained two weeks after training. The pretest , posttest and retention test scores were compared for the five language groups (Hindi, Tamil, Bodo, Mizo and English). Table 55 and Figure 55 show the comparison of the five groups at three test times ( p retest, posttest and retention test). As shown in Table 53 and Figure 53, all e xperimental groups performed faster on p osttest than on the p retest. Further , a repeated measures ANOVA with language groups as the betweensubjects factor and t est t ime ( pretest , posttest and retention test) as the withinsubjects factor revealed a significant main effect of t est t ime [ F (2,140) = 62.82 , p = .000] and a signi ficant main effect of group [ F (4,70) = 6.58 , p = .000]. The interaction between group and t est t ime was also found to be significant [ F (8,140) = 6.58 , p = .000]. The results were corrected using the Bonferroni method. It indicated that the significant differences were due to significant differences between Hindi and Bodo (p=.045), Hindi and English (p=.000), Hindi and Mizo (p=.005), Hindi and Tamil (p=.000) and Bodo and English (p=.010). Furthermore, two repeated measures ANOVAs, with groups as the bet weensubjects factor, one with test time p retest vs retention test as the within subjects factor and another with test time posttest vs r etention test showed significant main effects of t est t ime [ F (1,70) = 14.52 , p = .003], group [ F (4,70) = 218.047 , p = .000] and a significant interaction between group and t est time [ F (4,70) = 14.493 , p = .000] for pretest vs. retention test. For posttest vs retention there was a significant main effect of group [ F (4,70) = 55.883 , p = .000], main effect of test time [ F (1,70) = 22.425 , p = .001] and significant interaction between test time and group [ F (4,70) = 24.109 , p =

PAGE 87

87 .000]. The results indicate that the significant main effect of test time in the main ANOVA for group* test time ( pretest , posttest and retention test) was due to a significant difference between both the pretest and the retention test and the posttest and the retention test . For posttest and retention test, Post hoc comparison using Bonferroni method revealed a significant difference in t he mean reaction times of Hindi group and Mizo group(p=.034) and a marginally significant difference between Hindi and Bodo group (p=.074). For pretest and retention test, Post hoc Bonferroni results showed that there is a significant difference between each pair of groups (p<.05) , except that there is no difference between the Bodo and English groups (p=.190). A comparison of the mean reaction times of different groups at the retention test at the three ISIs 250ms, 500ms and 1500ms was carried out using an ANOVA with group as the betweensubjects factor and ISI as the within subjects factor. The results indicated a significant main effect of ISI [ F (2,140) = 13.69 , p = .006] and a marginally significant main effect of group [ F (4,70) = 5.85 , p = .040] and a marginally significant interaction between group and ISI [ F (8,140) = 3.561 , p = .062]. This supports the finding that all experimental group participants took more time at ISI=1500ms than at 500ms or 250ms. To conclude, the results from the pretes t , posttest and retention test indicate that there was a significant improvement in all the experimental groups , Bodo, Mizo, Tamil and English, immediately after the training, indicated by the significantly shorter posttest reaction times. All groups took less time to respond to stimuli two weeks after training

PAGE 88

88 than at the pretest . However , there was no significant improvement in the reaction time from posttest to retention test in any group except Tamil group (762.02msec vs. 673.59msec). Generalization T est The effectiveness of the training was also determined on the basis of the participants’ mean reaction time on a generalization test.. The generalization test used stimuli belonging to a new place of articulation, Dental, while the pretest , posttest , training and retention test used bilabial and velar stimuli. All the stimuli in the generalization test was presented at an ISI of 1000ms. The reaction times during the generalization test were studied to see if participants take the same time to react t o new stimuli or not. reaction time for all the five groups are analyzed and reported in Table 56. Between G roups The reaction time data from the generalization test was submitted to an ANOVA for a single factor with group (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor. The results show that the main effect of group is significant [ F (4,70) = 7.60 , p = .023]. Further , post hoc Ttests were done on control group Hindi and each experimental group, pairwise, to see if the difference between each of the pairs is significant or not. The Ttests indicate that the all the experimental groups were significantly different from the Hindi control group (p<.05) , except the Mizo group (p=.078). Figure 56 shows the mean reaction times of all groups along with their standard error.

PAGE 89

89 P retest and generalization test comparison The reaction times for the generalization test and the pretest were submitted to a repeated measures ANOVA with group (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor and test time ( pretest vs. generalization test) as the within subjects factor. The results showed that there is a significant main effect of group [ F (4,70) = 9.68 , p = .014]. There was no main effect of test time [ F (1.70) = .173 , p = .695] and the interaction between group and test time was also not found to be significant [ F (4,70) = .835 , p = .557]. The Post hoc Bonferroni method did not indicate any significant difference between any two groups. Table 57 and Figure 57 show a comparison of the mean reaction scores at three times pretest , posttest and generalization test. P osttest and generalization test comparison The reaction times for posttest and generalization tests were compared and analyzed using a repeated measures ANOVA with group (Hindi, Bodo, Mizo, English and Tamil) as the betweensubjects factor and test time ( posttest vs. generalization test) as the withinsubjects factor. The results showed that there is a significant main effect of group [ F (4,70 ) = 5.89 , p = .039]. There was no significant main effect of test time [ F (1,70) = 4.49 , p = .088] or any significant interaction between test time and groups [ F (4,70) = 1.87 , p = .254]. Bonferroni method showed that none of the groups had any signific ant difference between each other (p>.05). Next, a repeated measures ANOVA with groups as the betweensubject factor and test time ( pretest , posttest and generalization) as the withinsubjects factor showed a marginally significant main effect of group [ F (4,70) = 4.90 , p = .056] and a significant

PAGE 90

90 main effect of test time [ F (2,140) = 5.49 , p = .041]. No significant interaction between group and test time was found [ F (8,140) = 1.86 , p = .213] To summarize the findings of the generalization test, the language groups did not show any significant difference in the mean reaction times from pretest to generalization test or from posttest to generalization test. Reaction Time of C ontrasts The reaction times were also compared for different contrasts at dif ferent times : pretest , posttest , retention test and generalization test. The reaction times for all contrasts were shorter during the posttest than on the pretest ; however , there was no significant difference between the reaction times depending upon contrasts. In other words , there was no main effect of phonological contrast on the time taken to respond to a given pair of stimuli. Summary of R esults for R eaction time The above analyses concluded that all experimental groups , Mizo, Bodo, Tamil , English and the control group Hindi , took significantly less time to react to stimuli on the posttest than on the pretest . Also , all the groups retained the effect of training and took lesser reaction time than they did on the pretest . They performed similar to their performance on the posttest . The mean reaction times for all groups were not significantly different from either the pretest or the posttest . Table 58 to Table 512 summarize the significant results of reaction times from all the tests in this chapter.

PAGE 91

91 Table 51. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the pretest with their Standard deviation and Standard Error. Hindi Tamil Bodo English Mizo Mean 1518 1055 1511 997 1065 SD 48 376 285 124 56 SE 34 265 202 88 40 Table 52. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups during pretest at ISIs 250ms, 500ms and 1500ms with Standard deviation and Standard Error. Hindi Tamil Bodo 250ms 500ms 1500ms 250ms 500ms 1500ms 250ms 500ms 1500ms Mean 1658 1369 1528 904 1018 1242 1435 1443 1654 SD 30 43 71 196 339 594 278 275 305 SE 21 31 50 138 240 420 196 194 216 Table 52. Continued English Mizo 250ms 500ms 1500ms 250ms 500ms 1500ms Mean 1024 926 1044 957 987 1252 SD 58 43 270 35 28 107 SE 41 30 191 25 20 76 Table 53. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at pretest and posttest with Standard deviation and Standard Error. Hindi Tamil Bodo English Mizo Pre Post Pre Post Pre Post Pre Post Pre Post Mean 1518 1019 1055 762 1511 810 997 630 1065 856 SD 48 83 376 55 285 104 124 167 56 318 SE 34 59 265 39 202 73 88 118 40 225 Table 54. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups during the retention test with Standard deviation and Standard Error. Hindi Tamil Bodo English Mizo Mean 1453 674 864 560 1030 SD 23 72 132 163 386 SE 16 51 93 115 272

PAGE 92

92 Table 55. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest , posttest and retention test with Standard deviation and Standard Error. Hindi Tamil Bodo Pre Post Ret Pre Post Ret Pre Post Ret Mean 1518 1019 1453 1055 762 674 1511 810 864 SD 48 83 23 376 55 72 285 104 132 SE 34 59 16 265 39 51 202 73 93 Table 55. Continued English Mizo Pre Post Ret Pre Post Ret Mean 997 630 560 1065 856 1030 SD 124 167 163 56 318 386 SE 88 118 115 40 225 272 Table 56. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test with Standard deviation and Standard Error. Hindi Tamil Bodo English Mizo Mean 1815 1083 1140 756 1174 SD 24 75 155 18 405 SE 17 53 110 13 287 Table 57. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the pretest , posttest and generalization test with Standard deviation and Standard Error. Hindi Tamil Bodo Pre Post Gen Pre Post Gen Pre Post Gen Mean 1518 1019 1815 1055 762 1083 1511 810 1140 SD 48 83 24 376 55 75 285 104 155 SE 34 59 17 265 39 53 202 73 110 Table 57. Continued English Mizo Pre Post Gen Pre Post Gen Mean 997 630 756 1065 856 1174 SD 124 167 18 56 318 405 SE 88 118 13 40 225 287 Table 58. Summary of significant results for reaction times at pretest Factors tested F value P value group (4,70) = 10.66 p = .001 group*ISI ISI (2, 140) = 5.54 p = .034

PAGE 93

93 Table 59. Summary of significant results for reaction times at pretest vs posttest Factors tested F value P value group X ISI X test time ( pretest, posttest) ISI (2,140) = 17.76 p = .006 test time (1,70) = 21.19 p = .006 group X test time (pretest, posttest) test time (1,70) = 62.96 p = .000 group (4,70) = 47.17 p = .000 group* test time (4,70) = 11.44 p = .000 group X ISI group (4,70) = 7.01 p = .002 Table 510. Summary of significant results for reaction times at ISIs 250ms, 500ms, 1500ms. Factors tested F value P value group (4,70) = 5.85 p = .040 ISI (2,140) = 13.69 p = .006 group *ISI (8,140) = 3.561 p = .062 Table 511. Summary of significant results for reaction times at pretest vs posttest vs retention test Factors tested F value P value test time (2,140) = 62.82 p = .000 group (4,70) = 6.58 p = .000 test time* group (8 ,140) = 6.58 p = .000 group* test time (pretest vs retention) test time (1,70) = 14.52 p = .003 group (4,70) = 218.047 p = .000 group* test time (4,70) = 14.493 p = .000 group* test time (posttest vs retention) test time (1,70) = 22.425 p = .001 group (4,70) = 55.883 p = .000 group * test time (4,70)=24.109 p=.000 Table 512. Summary of significant results for reaction times at generalization test Factors tested F value P value group (4,70) = 7.60 p = .023 group X test time (pretest vs generalization) group (4,70) = 9.68 p = .014 group X test time (posttest vs generalization) group (4,70) = 5.89 p = .039 group X test time (pretest vs posttest vs generalization) group (4,70) = 4.90 p = .056 test time (2,140) = 5.49 p = .041

PAGE 94

94 Figure 51. Comparison of reaction time (in msec) of Hindi, Tamil, Bodo, English and Mizo groups at the pretest . Figure 52. Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups during pretes t at ISIs 250ms, 500ms and 1500ms. 0 200 400 600 800 1000 1200 1400 1600 1800 Hindi Tamil Bodo English MizoReaction Time (msec) Group Pretest 0 200 400 600 800 1000 1200 1400 1600 1800 Hindi Tamil Bodo English MizoReaction Time (msec) Group Pretest at different ISIs 250ms 500ms 1500ms

PAGE 95

95 Figure 53. Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups at pretest and posttest . Figure 54. Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups during the retention test. 0 200 400 600 800 1000 1200 1400 1600 1800 Hindi Tamil Bodo English MizoReaction Time (msec) Group Pretest vs Posttest Pretest Posttest 0 200 400 600 800 1000 1200 1400 1600 1800 Hindi Tamil Bodo English MizoReaction Time (msec) Group Retention Test

PAGE 96

96 Figure 55. Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups at the three test time s pretest , posttest and retention test. Figure 56. Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups during the generalization test. 0 200 400 600 800 1000 1200 1400 1600 1800 Hindi Tamil Bodo English MizoReaction Time (msec) Group Pretest vs Posttest vs Retention Test Pretest Posttest Retention test 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Hindi Tamil Bodo English MizoReaction Time (msec) Group Generalization Test

PAGE 97

97 Figure 57. Comparison of reaction time of Hindi, Tamil, Bodo, English and Mizo groups at the pretest , posttest and generalization test. 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Hindi Tamil Bodo English MizoReaction Time (msec) Group Pretest vs Posttest vs Generalization test Pretest Posttest Generalization

PAGE 98

98 CHAPTER 6 DISCUSSION AND CONCLUSION Adult speakers of four different languages , Tamil, Mizo, Bodo and English, were trained to perceive nonnative laryngeal contrasts. The experimental groups differed from the native Hindi group in that the experimental groups have one, two or three of the four laryngeal contrasts that Hindi has. The training was done using a speech perception experiment that presented stimuli on a same or different task on three different ISIs: 250ms, 500ms and 1500ms. During training, feedback was given via the computer screen to ensure that they know the c orrect response to a given pair of sounds. The results of the training were analyzed for m ean p ercent correct identification and the reaction time at four different times : pretest, posttest, retention test and generalization test. P retest The results from the pretest of the four experimental groups , Tamil, Mizo, Bodo, English , and the control group Hindi were compared. The control group Hindi performed the best of all the groups on the pretest. It had the highest mean accuracy percent at all the three ISIs. Also the mean accuracy was highest when the ISI was 250ms and lowest when ISI was 1500ms for all groups. The Bodo and Mizo groups did significantly better at 250ms than at 1500ms. This is in consonance with previous research (Pisoni and Tash 1974) that shows that when two stimuli are presented with a longer ISI, they become difficult to discriminate. On the other hand, when these stimuli are presented with a shorter ISI , the discrimination is better. According to Werker and Tees (1984) , this would be inter preted to mean that adults do better at the auditory level that at the phonological level. However , researchers have argued that better response on 250ms

PAGE 99

99 than on 1500ms is due to memory load. The memory load is more when the ISI is longer. Given a longer ISI, subjects need to remember the auditory trace for A for a longer duration and by the time X is presented, the auditory trace of A may have faded away, given that the lifespan of auditory memory is just 200300ms (Gerrits and Schouten 2004). reaction time data also supports the percent correct identification data. Mean reaction time for all groups is longer when the ISI is 1500ms and is shorter when the ISI is 500ms. Subjects took more time to respond when the time lag between two stimuli was longer, while they responded faster when the time lag was shorter. This indicates an inverse relation between I SI and reactionreaction time. P retest and P osttest The effectiveness of training was analyzed by comparing the percent correct identification and reaction time of all experimental groups during the pretest and the posttest. It was found that the mean react ion time reduced significantly from pretest to posttest. This indicates that after undergoing training, participants had to think less before responding to a pair of syllables on an AX discrimination task. This might be the result of becoming familiar with the discrimination task over the course of training. To make sure that it is not just familiarity with the task that reduces the reaction time during the posttest, we will look at the reaction time during the generalization test later in this chapter. In addition to this, the analysis of mean percent correct identification reveal s that all the experimental groups , Tamil, Mizo, Bodo and English, did significantly better on the posttest that they did on the pretest. Not only did they show significant improv ement, but also they became more similar to the native control group Hindi in their accuracy.

PAGE 100

100 There was a significant difference in the percent correct accuracy of the control group and the experimental group during the pretest. However, there was no signi ficant difference in the performance of control group and the experimental groups during the posttest. The English and Bodo groups showed an improvement of 7% and 8% respectively, Tamil speakers improved in accuracy by approximately 12%, and the most signi ficant improvement can be seen in Mizo group, which improved 22%. It is interesting to note that Mizo has three out if the four laryngeal contrasts that belong to Hindi. Mizo has voiceless aspirated, voiceless unaspirated and voiced unaspirated contrast. T his means that they need to learn only one more laryngeal contrast to attain nativelike perception. Thus the task might be easier for the Mizo group to begin with. This acquisition also supports the theory that adults are capable of acquiring a new contrast if it exists somewhere else in their language, since Mizo adults have the /p/ /ph/ aspiration contrast already in place in the L1 and so they only had to transfer it to the [b]/[bh] contrast. It was also observed that all the groups were more accurate at the 250ms ISI and least accurate at the 1500ms. This further supports the role of memory load in speech perception. The longer the ISI , the more difficult i t becomes for the subject to perceive the contrast. When the ISI is as long as 1500ms , the memory needs to hold the first sound in memory for a longer duration than when the ISI is 250ms only and then retrieve it for comparison upon listening to the next syllable in the pair. The fact that participants perceive better at 250ms than at 1500ms is indicative of continuous mode of perception where the listener is able to perceive the finer distinctions within a

PAGE 101

101 category , rather than the categorical mode of perception where the listener perceives only betweencategory differences. Note that the control Hind i group did not undergo training, although they were administered the pretest and the posttest at the same time interval as the rest of the experimental groups. The Hindi control group did not show any significant difference in its performance on pretest and posttest except in reaction time. To summarize, the results show that the accuracy of participants from all the language groups improved significantly from pretest to posttest both in terms of percent correct identification and reaction time . This impr ovement from pretest to posttest suggests that perceptual training experiments like this one are capable of changing one’s perceptual capabilities, even in adulthood. Thus adults’ perceptual system is malleable (Best 1995; Flege 1995), and they are capable of learning novel sounds that don’t preexist in their phonological system well after the ‘critical period.' The results also support the Speech Learning Model (Flege 1985) that claims that the ability to establish new phonetic categories remains intact throughout one’s lifespan. P retest , P osttest and R etention Test The effect of training was evaluated on the basis of three separate tests: pretest, posttest and a delayed posttest. The delayed posttest was the retention test that the participants were subjec ted to two weeks after the posttest. The results of the retention test for all groups were compared with their pretests to see if adults retain the effect of training. All groups showed significant improvement from pretest to posttest and from pretest to r etention test. But no group showed significant improvement from posttest to retention test. In fact , the percent correct accuracy of all groups except Bodo fell marginally from posttest to retention test. The retention test results were the worst for

PAGE 102

102 the Tamil group, whose accuracy fell 22% after the posttest. Clearly, the Tamil subjects had not been able to establish longterm perceptual categories in their phonological system. The percent correct identification results were supported by the reaction time data. All groups took less time to respond to stimuli two weeks after training than at the pretest. However , there was no significant improvement in the reaction time from posttest to retention test in any group except the Tamil group (762.02msec vs. 673. 59msec). However , this reduction in reaction time from posttest to retention test was accompanied by less accurate identification scores. Although more data is needed to show this, so far the results indicate that for the Tamil group accuracy was compromis ed due to fast response by the subjects. This is a common disadvantage of an AX experiment design since subjects might be inclined to choose speed over accuracy. Overall , the findings suggest that there is improvement in accuracy and reaction time of the Mizo, Bodo and English groups two weeks after training. This indicates that the subjects were able to establish longterm phonological categories for new categories in their perceptual system. In terms of Universal Grammar (Chomsky 1995), this indicates that adults might have access to phonological contrasts beyond their L1s. Bodo subjects had only two laryngeal contrasts, namely, voiced unaspirated and voiceless aspirated in their L1. But the fact that they improved significantly after training and reached a level of 90% accuracy as opposed to 78% accuracy during the pretest indicates efficacy of the training in creating new phonological categories in the subjects’ phonological system. The results seem to support the f ull transfer/ f ull a ccess paradigm

PAGE 103

103 (Schwartz and Sprouse 1994, 1996) since the subjects in the experimental groups were able to draw from the Universal Grammar what was not present in their respective L1s. Considering the results from an Optimality Theory (Prince and Smolensky 1993) point of vi ew, the improvement form pretest to post and from posttest to retention test are indicative of reranking of some constraints. Bodo, Mizo and English subjects that did not perceive voiced aspirated word initial consonants in their L1 seem to now allow them to some extent, thereby leading to significant improvement in percent correct identification. Optimality Theory assumes that the markedness and faithfulness constraints are arranged in a hierarchy in the phonologies of speakers, and it is possible to rerank the constraints in response to the input. The improvement in percent correct identification in the Bodo, Mizo and English groups then indicates that some markedness constraint such as *VOICED ASPIRATED is being lowered below some faithfulness constraint that wants the input and output to be identical such as IDENT LAR (Lombardi 1990). Here *VOICED ASPIRATED means no voiced aspirated consonant, and IDENT LAR means the input and output should be identical with respect to voicing and aspiration. Thus , with t raining , GLA is at work here to allow restructuring of the phonological system so as to allow the new L2 laryngeal contrasts. Especially for the Bodo and English groups that do not have the voiceless unaspirated and voiced aspirated sounds phonemically, the ability to do better with such a short term training means that they are able to access Universal G rammar beyond their L1. In terms of Optimality Theory, it would mean that they have access to Universal Gramma r constraints and the continued flexibility to rerank them. The findings from the study strengthen previous research that has shown that adults are capable of nativelike

PAGE 104

104 performance on phonological acquisition tasks (Bongaerts 1999). It is also important to observe that this ability to improve significantly in acquiring novel L2 contrasts is not just incidental. The study shows that this improvement is true for three out of the four experimental group that participated in this study, which shows that adul ts are capable of phonological acquisition late in life irrespective of their L1. On the other hand, the failure of Tamil group to improve needs to be explored more. First, Tamil had the least number of contrasts, only one (voiceless unaspirated), in their L1, making the task more difficult for them as compared to the other experimental groups. Second, the fact that Tamil did show significant improvement from pretest to posttest but not from posttest to retention test might be indicative that this short ter m training was not sufficient for them. It would seem likely that they needed more time and maybe more training to remember the new contrasts. Nevertheless, their progress from pretest to posttest shows that they were able to rearrange their markedness and faithfulness constraint, just like the other experimental groups , Mizo, Bodo and English. Broadly speaking, all the findings from all the groups suggest that adults have the ability to reorganize their L1 constraints when exposed to novel L2 contrasts eve n in adulthood. Thus, adults are capable of accessing Universal Grammar beyond what is there in their L1, and perceptual training is an effective tool for encouraging them to do so. G eneralization Test The results from the generalization test were analyzed to see how well subjects generalize the effect of training to new stimuli. The analysis revealed that all the experimental groups improved significantly from pretest to posttest. Two weeks after posttest, all the groups were able to generalize the effect of training to a new place of

PAGE 105

105 articulation, namely, dental. All the stimuli used for the pretest, posttest and the retention test belonged to bilabial and velar places of articulation. The subjects were never exposed to Hindi contrasts belonging to dental place of articulation before this generalization test. It was found that Bodo group performed 18% better, the English group did 5% better, and the Mizo group did 22% better on the generalization test than they did on the pretest. The Tamil group showed a s light decline in their performance from pretest to generalization test of about 5%. A comparison of the posttest and the generalization test showed that all the experimental groups except Tamil were able to generalize the effect of perceptual training to dental stimuli. English and Mizo did not show any significant difference in their performance on the posttest and the generalization test. The Bodo group showed a significant improvement of 11% from posttest to generalization test. However, Tamil speakers showed a decline of 17% from posttest to generalization test. Clearly, Tamil subjects were unable to generalize the effect of training to new kinds of stimuli. This is expected of them ; as seen earlier in the chapter, the Tamil subjects were unable to reta in the effects of the training two weeks after the training. They were unable to construct longterm phonological categories of L2 contrasts, which in turn explains why they could not extend the learning to perceive new stimuli. Turning to reaction time data, the ANOVAs did not reveal any significant difference in the reaction times of pretest and generalization test and posttest and generalization test. This indicates that in each of the tests pretest, posttest and generalization test the subjects took sim ilar amount of time to process the information and react to it. The OT analysis treats the contrasts as a matter of Laryngeal identity or

PAGE 106

106 laryngeal markedness, predicting that they should be independent of place; these results are consistent with such an analysis. The results from the retention and generalization tasks of all groups indicate that all the groups except Tamil were able to retain the effects of training two weeks after training. The fact that they were able to generalize the results to a new I SI, 1000ms, and to a new place of articulation, dental, shows that acquisition has taken place at a deeper phonological level and not just the auditory level. This learning has altered the existing phonological systems of the L1 speakers to accomodate more voicing and aspiration contrasts. Thus, the results are not merely indicative of their performance but their competence or capability of acquisition of new phonological contrasts. Although the experiment measures their performance on a perceptual 'same or different' task, their performance suggests a bigger possibility , that adults 'can' acquire new contrasts even after critical period. The fact that the Tamil group improved significantly from pretest to posttest is indicative of an alteration to their phonological systems. However, following the posttest, their failure to retain or generalize the results shows that they still 'can' alter their phonological systems but that they have more learning to do in terms of reranking of phonological constraints. To summarize this section, the ability to generalize the skills acquired from any training is an essential part of acquisition. Improvements in the generalization task supports the interpretation that the learning is in fact longterm and not just a result of rote memory. The fact that the Mizo, Bodo and English groups were successfully able to generalize the effects of training to new stimuli supports the idea that adults have the

PAGE 107

107 capability to access Universal G rammar well beyond their L1 and can rearrange the constraint hierarchy in their L1 in order to acquire novel L2 phonological contrasts. The fact that Tamil group did not do as well needs to be subjected to technical considerations such as length of training in addition to other nonlinguistic considerations such as motivation and interest. Account Using Dispersion Theory The results of the present study can be explained using two approaches: i) an analysis that makes use of the traditional markedness constraints and ii) an analysis that involves a pr eference for maximally distinct contrasts. Functionalist hypotheses that phonological systems are well adapted for communication suggests that humans prefer maximally distinct contrasts. For effective communication, speech sounds must be perceived accurately and quickly. Hearers are more accurate and faster in perceiving the category of a stimulus when it is more distinctive than the contrasting categories (Pisoni and Tash 1974, Podgorny and Garner 1979). An important constraint of speech perception on phonology is the principle of maximization of distinctiveness , the tendency to maximize the distinctiveness of contrasting speech sounds. Different researchers have given different names to this principle. For instance, Flemming (1996) refers to it as ‘Disper sion T heory’ after Lindbom’s ‘Theory of Adaptive Dispersion’ (1990), suggesting that the contrasting categories must be dispersed or separated in perceptual space. Dispersion T heory suggests that there are three main goals of phonological contrasts : i) max imize the distinctiveness of contrasts, ii) minimize articulatory effort and iii) maximize the number of contrasts. The goals of maximizing the number of contrasts and maximizing the

PAGE 108

108 distinctiveness of contrasts inherently conflict with each other because the greater the number of contrasts along a dimension, the closer they will be in auditory space. The experimental groups in this study do not have four way phonemic contrast like Hindi, but they do have some of them present in their language as allophones in different word positions. The D ispersion T heory can explain why then participants would have difficulty in categorizing them as different phonemes before training. Stimuli that are closer together in the auditory space are susceptible to more confusion by listeners (e.g. Shepard 1957, Nosofsky 1992). Under this theory, the results can be analyzed using constraints like MINDIST and MAXIMISE CONTRASTS. MINDIST assigns one violation mark for each pair of contrasting sounds that are separated by less than the minimum distance. Phonemically speaking, we know that distance should be in the order [voiceless unaspiratedvoiceless aspiratedvoiced unaspiratedvoiced aspirated]. On the other hand, MAXIMISE CONTRASTS sees more contrasts as being better; in other w ords , the lesser the distance between contrasts, the better it is. An improvement in accuracy from the pretest to posttest for the language groups of this study is indicative of increasing preference for the constraint MAXIMISE CONTRASTS over MINDIST const raints. After training, the phonological system of the listeners allows them to perceive more contrasts than before. They are trying to maximize the number of contrasts that they can hear even if that means that the minimum distance between contrasts will decrease. The improvement results from an interaction between how much minimum distance can be tolerated and how many contrasts can be tolerated, although more extensive training

PAGE 109

109 will be required to see exactly how much minimum distance is allowed by the s ystems of these L1 learners. Bodo and English Ranking Using Dispersion Theory Table 61 shows the ranking for Bodo and English using dispersion theory constraints MINimum DISTance and MAXIMIZE CONTRASTS. The constraint maximize contrasts is ranked higher t han minimum distance 1 and minimum distance 2 that shows that Bodo and English prefer having more than just one contrast. However , MAXIMIZE CONTRASTS is ranked lower than MIN DIST3, thereby not allowing more than a difference of 3 in terms of voicing and aspiration. The result is a phonological system of stops with three laryngeal contrasts. Table 61shows that Bodo and English prefer contrasts to have at least a minimum distance of 3 even if that means having a smaller number of contrasts. Mizo Ranking Usi ng Dispersion Theory Table 62 shows that although Mizo likes to maximize contrasts , it likes to maintain a minimum distance of at least 2 between the laryngeal contrasts. The result is a system of sounds that have only two laryngeal contrasts. Target L2 account using Dispersion Theory The improvement from pretests to posttest and from pretest to retention is indicative of reranking of constraints to allow a greater number of constraints in Tamil, Bodo, Mizo and English. All the experimental groups are in other words maximizing the number of contrasts that they perceive. This is turn means that the minimum distance between contrasts is reduced. It is then clear that the minimum distance between contrasts is ranked lower than the constraint that prefers maxi mum contrasts. The reranking of all the experimental groups would then be shown as in Table 63.

PAGE 110

110 In Table 63 MINDIST1 and MINDIST2 mean that the candidate must have contrasts that differ by at least 1 and 2 respectively on the laryngeal dimension. Account U sing Laryngeal Contrasts in Optimality Theory A traditional analysis of the results of this study would involve a mapping of input to a set of output candidates using the GENerator and the selection of the best candidate using the EVALuation function. The optimal candidate is chosen based on the ranking of the constraints in a given language. The optimal perceptual output in this study relies on the interaction of markedness constraints and faithfulness constraints. We know that of all the languages in the present study (Tamil, Bodo, Mizo, English and Hindi), Hindi has the most marked segments, the voiced aspirated stops. This means a listener must be able to hear a more marked contrast in order to improve their accuracy on the perception test, which in terms of constraint ranking means violation of a markedness constraint that prevents voiced aspirated sounds such as * VOICED ASPIRATED. This violation will occur to satisfy a higher ranked faithfulness constraint that requires the output to be identical to input in terms of laryngeal features such as IDENT(LAR). This traditional OT analysis thus suggests the lowering of markedness constraints to allow input contrasts to be heard unchanged in the output, by the listener. The pretest shows that the experimental groups Tamil, Bodo, Mizo and English were initially unable to perceive the new Hindi contrasts. The ranking for each group before training is described in more detail below Optimality Theoretic A ccount of Bodo and English Tables 64 to 67 model Bodo and English listeners’ perception before any training. Phonemically speaking, it can be seen that Bodo and English listeners can distinguish between voiceless aspirated and voiced unaspirated, but they cannot

PAGE 111

111 discriminate voiced aspirated and voiceless unaspirated. This indicates that there are some constraints at work that allow the distinction between voiced and voiceless sounds but not aspirated counterparts. The model thus needs to represent constraints that prevent Aspiration (*ASPIRATED). There mus t also be some faithfulness constraint that needs the input and output to resemble each other in their voicing and aspiration (IDENT LAR), ranked high enough to allow voiceless unaspirated and voiced unaspirated but low enough below markedness constraints that prevent aspirated forms of the former. In terms of D ispersion T heory , the constraints rearrange in order to allow more dispersed laryngeal contrasts in the perceptual system of learners , thereby allowing them to perceive another contrast, namely voiced aspirated, which they could not perceive before training. Optimality Theoretic A ccount of Tamil Tables 68 to 611 represent the ranking of markedness constraints with respect to faithfulness constraint. It can be seen that faithfulness is ranked lowest and Tamil does not allow voiced aspirated, voiced unaspirated, or voiceless aspirated sounds in the initial position. However , their better performance on the posttest is indicative of rearrangement of these constraints to allow some Hindi contrasts. This in turn reflects that some markedness must have been demoted below IDENT (LAR) to allow other laryngeal contrasts. In terms of D ispersion T heory , the constraints rearrange in order to allow three more laryngeal contrasts , voiceless aspirated, voiced unaspirated and voiced aspirated, in the perceptual system of learners. Optimality Theoretic A ccount of Mizo Tables 612 to 615 represent the ranking of markedness constraints with respect to faithfulness constraint. It can be seen that Mizo does not allow voi ced aspirated stops

PAGE 112

112 word initially , but it ranks IDENT (LAR) higher than other markedness constraints to allow voiceless unaspirated, voiceless aspirated and voiced contrasts. It is useful to note that *VOICED ASPIRATED and *ASPIRATED initially outrank I DENT [LAR]. However, after the posttest the percent correct identification improves significantly for Mizo listeners which indicates that participants got better in perceiving voiced aspirated and voiceless contrasts. With training the ranking of the const raints was rearranged by the phonological system of the Mizo listener and markedness constraints were demoted below faithfulness constraint IDENT (LAR). In terms of D ispersion T heory , the constraints rearrange to maximize the number of contrasts by allowin g the learner to perceive voiced aspirated stops. Optimality Theoretic Account of Constraint Ranking A fter Training After the training sessions , there was improvement in percent correct identification and reaction time of all language groups. The ranking of constraints of the experimental groups , Tamil, Mizo, Bodo and English, seems to be becoming more like the Hindi constraint ranking as shown in Table 616 to 619. Coming back to the two main research questions of this study , the questions are specifical ly addressed below. 1. Can transfer effects (from L1) and/or accessibility to Universal Grammar (UG) explain the ultimate attainment of Hindi laryngeal contrasts by various L1 groups? Sub questions: How do L1 voicing contrasts affect the perceptual attainment of Hindi voiced stops? How do L1 aspiration contrasts affect the perceptual attainment of Hindi voiced stops? 2. Is there a main effect of ISI (500ms and 1500ms) on the perceptual accuracy of voicing immediately and two weeks after training?

PAGE 113

113 For the first question, yes the transfer effects from L1 and accessibility to Universal Grammar can explain the perception of Hindi laryngeal contrasts by all experimental group subjects , Tamil, Bodo, Mizo and English. The transfer effects of L1 phonological categories is indicated by the pretest results of all groups when none of them had native Hindi like perception, and their mean percent correct identification and reaction times were significantly lower from that of the control group Hindi. Before training, the exper imental groups Tamil, Mizo, Bodo and English could not perceive all four laryngeal contrasts that Hindi has; they could only transfer the contrasts that they had in their respective L1s. The answer to the sub questions is that prior to training, L1 voicin g and aspiration contrasts affected the perception of Hindi contrasts. The participants could access Universal Grammar only via their L1s. This is indicated by the accuracy results before training that indicate Tamils have 65.67%, Bodos have 78%, English s peakers have85.33% and Mizos have 68.67% correct identification prior to any training. The Tamils could not perceive voiceless aspirated, voiced unaspirated and voiced aspirated sounds because they did not have these contrasts in their L1. The Bodos and A merican English subjects could not perceive the contrast between voiced unaspirated and voiced aspirated sounds because this contrast was absent from their L1. The Mizos could not perceive voiced aspirated contrasts since this was missing in their L1. Thes e explain the low mean percent correct responses of all the experimental groups before training . However , after the training sessions, there was a significant improvement in the response accuracy of all the groups. They were now able to perceive a significantly greater number of contrasts than before. This indicates that they

PAGE 114

114 were able to perceive contrasts t hat they could not hear before training and feedback. The participants in the experimental groups could be trained to perceive what was not present in their L1s. These results reflect the ability of these L1 speakers to access Universal Grammar constraints beyond that of their L1 constr aints with the help of training and being able to rerank these constraints differently than their L1. For the second question, throughout the experiment , it was a general observation that all groups performed better when the Inter Stimulus Interval was shorter. All the participants tend to do better when the ISI was 500ms than when it was 1500ms. Not only was the percent correct identification reduced with longer ISI, the reaction times increased with increasing ISIs. There was a statistically significant main effect of ISI on reaction times of the experimental groups immediately and two weeks after training. However, the main effect of ISI was not significant when the percent correct identification of all groups were compared at the posttest and at the retention test. Concluding Remarks To conclude, this study investigated the acquisition of novel L2 contrasts by speakers of four different L1s. The control group was Hindi and the four experimental groups were Bodo, Mizo, Tamil and English. The study investigated the acquisition of four way Hindi laryngeal contrasts by speakers of these experimental groups. Tamil, Bodo, Mizo and English have one way, two way and threeway voicing contrasts , respectively. A speech perception exper iment was designed to train the subjects to perceive novel Hindi contrasts using feedback about the correct response. The data collected was not only analyzed for percent correct identification but also for reaction times. The stimuli were presented at thr ee ISIs 250ms, 500ms and 1500ms in order to see if ISI has a main effect. Tests were administered at different times , before training,

PAGE 115

115 immediately after training, two weeks after training and a fourth time to evaluate the ability to generalize the effects of training to new stimuli varying in its place of articulation. The results of the study show that participants of all groups improve significantly immediately after training. Both the percent correct identification and reaction times support this findi ng. However , two weeks after the training, during the retention test the percent correct accuracy of Tamil group significantly lowers. All other groups were able to retain the effects of training. During the generalization test, all groups except Tamil wer e also able to generalize the discrimination skills to new syllables that contained initial consonant belonging to dental place of articulation. The fact that Bodos, Mizos and English were able to retain the effects of training two weeks later and general ize it to a new place of articulation indicates that acquisition has occurred and that learners of these languages have been able to establish longterm perceptual categories. From an Optimality theoretic perspective, the results show that the learners of Bodo, Mizo and English have full access to the universal set of markedness and faithfulness constraints and that their ability to rerank these universal constraints, according to a new L2, remains intact even in adulthood. The success of Bodo, Mizo and English group can be attributed to the less distance to be covered in learning the contrast, with the direct result that they required less exposure to successfully master it (Boersma and Escudero 2004). On the other hand, the Tamils were not able to retain or generalize the learning. This throws light onto the learning task at hand for the Tamils. The Tamils had more to learn not just in terms of the number of contrasts they had to acquire, but also the

PAGE 116

116 process involved in order to do so. In OT terms , the Tamils had more constraint reranking to do than all the other groups. They had to raise the faithfulness constraint IDENT (LAR) above *VCD ASP, *ASPIRATED and *VOICED. All the other L1 groups need to rerank lesser constraints in order to arrive at the L2 Hindi ranking. This makes the task tougher for Tamil learners than other groups of learners , which might explain their lack of retention or generalization of the effects of training. The learnability of the novel contrasts also support the claims of Flege’s S peech Learning Model (1995) , which suggests that the ability to learn new speech sounds remains malleable across one’s life span, although it is easier to learn L2 sounds at an early age than later. Reflecting back on Optimality Theory, U niversal G rammar i s proposed as a set of universal constraints that are violable. Optimality Theory can model these perceptual outputs that the experimental groups arrive at after training, by means of reranking. Since these constraints are violable and rerankable, suffici ent new input causes the markedness constraint *VOICED ASPIRATED to move lower in the hierarchies of Tamil, Bodo, Mizo and English, thus resulting in optimal candidates that allow Hindi contrasts. Drawing from previous research and current study , it can b e concluded that adults can be trained to establish longterm L2 categories with the help of training well after the ‘critical period’. For a theory of acquisition, this evidence from multiple L1s shows more than just performance; it shows competence. A dis tinction between competence and performance is necessary. According to Chomsky (1995) , competence consists of the mental representation of linguistic rules of grammar. Since the rules of grammar are internalized by the learners, the performance of the L2 l earners is a

PAGE 117

117 window into their competence, although their performance is not their entire competence. In this study, the claim that the L2 learners have acquired native L2like competency in the domain of phonology is based not just on the posttest, conduc ted immediately after training but on the retention and generalization tests. The retention test shows the continued use of newly learned phonological contrasts in a nativelike manner. Furthermore, these L2 learners were able to abstract their learning to a new place of articulation and ISI they had not been trained on. They had not received any input containing dental aspiration and voicing contrasts. Had this been an acquisition at the phonetic or auditory level , this would not have been possible. In addition to this, in this study, the fact that L1 learners we re able to alter their phonological systems in three out of four cases indicates that learners were able to move around the constraints that might have been buried at the bottom, in their L1 ranking, such that they can arrive at the L2 constraint ranking. The improved performance of the L2 learners on the generalization task should thus be understood as evidence of their competence and not just performance. Implications of the S tudy The fact that new phonological contrasts can be successfully learned in adulthood helps to contribute to the question of ultimate attainment (White, 2003, Archibald 2000). This study contributes in providing evidence to the ‘can vs cannot’ dichotomy (Rothman 2008) of second language acquisition after the critical period has passed. It supports the view that adult learners ‘can’ learn new phonological contrasts even after the critical period. This empirical study can be replicated and verified by researchers interested in the domain of second language phonological acquisition. The

PAGE 118

118 fact that phonological production particularly is restricted by muscular dexterity makes it especially interesting to second language acquisitionists. Along with this , it is claimed that phonologica l production is preceded by phonological perception. This study shows changes in perceptual capabilities in post critical adults after training . It implies that speech production might also be capable of altering after the critical period has passed. The study shows that fate of a new language contrast is not dark even though our ears have been trained to perceive sounds a certain way. Sufficient training with feedback has been shown to help develop new phonological categories among the perceptual syst ems of subjects cross linguistically (Neufeld 1977). Limitations of the Present S tudy More studies need to be conducted to ensure that some analyses that yielded numerically but not statistically significantly effects are not actually significant. For thi s, future researchers should use more participants, longer training time, more training sessions with feedback, and more tokens. Due to a lack of time , the current study could employ only a delay of two weeks before the retention test was administered. It is suggested that future studies employ a longer lag between the posttest and the retention test to further test the longterm gain of perceptual categories by participants. Future D irections The generalization test could include more variety of stimuli i nstead of just being varied in terms of place of articulation. For instance, it could use stimuli recorded by both male and female speakers and synthesized tokens instead of natural tokens. Also due to lack of time, this research could only provide four tr aining sessions to each group. It would be good to see how much training is required for each group to achieve near native accuracy. It would be insightful to see how much more training is required

PAGE 119

119 for groups that begin with fewer contrasts, such as Tamil in this study, in order to show significant improvement from the pretest. Further researchers may also explore how speech perception leads to production, for the learners of the given language groups Tamil, Bodo, Mizo and English. It is suggested that future studies take into account metalinguistic factors , such as memory load, motivation and willingness to learn, in addition to ISI. In order to test these meta linguistic factors , the posttest should be conducted at many times after the training, say after 4 weeks, 6 weeks and 8 weeks. The results of the tests at each time will indicate the effect of test time on the effect of training. Moreover , it is suggested the study be conducted with other language groups and the results be compared to control group H indi. Future researchers may also try to use a different training method, such as an AXB forced choice identification instead of the AX same or different task used in this study. A different task could show different results about the effect of task on the accuracy of the subjects (Wayland and Li, 2005). It is thus suggested that researchers continue to explore the area of second language phonological acquisition in more depth. Researchers may study the acquisition of other phonological properties by adults . They should also look at other L1 and L2 combinations while researching these phonological properties.

PAGE 120

120 Table 61. Pre training Bodo and English ranking using dispersion theory MINDIST 3 MAXIMIZE CONTRASTS MINDIST1 MINDIST 2 p p h **! * p b **! * p h b * * p p h b **!**** *** p p h b b h **!******** **** Table 62. Pre training Mizo ranking using dispersion theory MINDIST 2 MAXIMIZE CONTRASTS MINDIST1 MINDIST 3 p p h * ** p b * ** p h b * * p p h b *** ****** p p h b b h ****! ********** Table 63. Target L2 ranking using dispersion theory MAXIMIZE CONTRASTS MINDIST1 MINDIST 2 MINDIST 3 p p h * ** p b * ** p h b * * p p h b *** ****** p p h b b h **** ********** Table 64. Bodo and English ranking for /pa/ before training /pa/ *UNASP VOICELESS *VCD ASP IDENT (LAR) *VOICED *ASPIRATED [ba] * *! [b h a] *! * * [pa] *! [p h a] * *

PAGE 121

121 Table 65. Bodo and English ranking for /ba/ before training /ba/ *UNASP VOICELESS *VCD ASP IDENT (LAR) *VOICED *ASPIRATED [ba] *! [b h a] *! * * * [pa] *! * [p h a] *!* * Table 66. Bodo and English ranking for /pha/ before training / p h a / *UNASP VOICELESS *VCD ASP IDENT (LAR) *VOICED *ASPIRATED [ ba] *!* * [b h a] *! * * * [pa] *! * [p h a] * Table 67. Bodo and English ranking for /bha/ before training / b h a / *UNASP VOICELESS *VCD ASP IDENT (LAR) *VOICED *ASPIRATED [ba] * *! [b h a] *! * * [pa] *! ** [p h a] * * Table 68. Tamil ranking for /pa/ before training /pa/ *VCD ASP *ASPIRATED *VOICED IDENT (LAR) *UNASP VOICELESS [ba] *! * [b h a] *! * * ** [pa] * [p h a] *! * Table 69. Tamil ranking for /ba/ before training /ba/ *VCD ASP *ASPIRATED *VOICED IDENT (LAR) *UNASP VOICELESS [ba] *! [b h a] *! * * * [pa] ** * [p h a] *! *

PAGE 122

122 Table 610. Tamil ranking for /pha/ before training /p h a/ *VCD ASP *ASPIRATED *VOICED IDENT (LAR) *UNASP VOICELESS [ba] *! ** [b h a] *! * * * [pa] *! * [p h a] *! Table 611. Tamil ranking for /bha/ before training /b h a/ *VCD ASP *ASPIRATED *VOICED IDENT (LAR) *UNASP VOICELESS [ba] *! * [b h a] *! * * [pa] ** * [p h a] *! * Table 612. Mizo ranking for /pa/ before training /pa/ *VCD ASP IDENT (LAR) *VOICED *ASPIRATED *UNASP VOICELESS [ba] *! * [b h a] *! ** * * [pa] * [p h a] *! * Table 613. Mizo ranking for /ba/ before training /ba/ *VCD ASP IDENT (LAR) *VOICED *ASPIRATED *UNASP VOICELESS [ba] * [b h a] *! * * * [pa] *! * [p h a] *!* *

PAGE 123

123 Table 614. Mizo ranking for /pha/ before training /p h a/ *VCD ASP IDENT (LAR) *VOICED *ASPIRATED *UNASP VOICELESS [ba] *!* * [b h a] *! * * * [pa] *! * [p h a] * Table 615. Mizo ranking for /bha/ before training /b h a/ *VCD ASP IDENT (LAR) *VOICED *ASPIRATED *UNASP VOICELESS [ba] * *! [b h a] *! * * [pa] **! * [pha] * * Table 616. Target L2 constraint ranking for /pa/ /pa/ IDENT (LAR) *VOICED *ASPIRATED *VCD ASP [ba] *! * [b h a] *!* * * * [pa] [p h a] *! * Table 617. Target L2 constraint ranking for /ba/ /ba/ IDENT (LAR) *VOICED *ASPIRATED *VCD ASP [ba] * [b h a] *! * * * [pa] *! [p h a] *!* *

PAGE 124

124 Table 618. Target L2 constraint ranking for /pha/ /p h a/ IDENT (LAR) *VOICED *ASPIRATED *VCD ASP [ba] *!* * [b h a] *! * * * [pa] *! [p h a] * Table 619. Target L2 constraint ranking for /bha/ /b h a/ IDENT (LAR) *VOICED *ASPIRATED *VCD ASP [ba] *! * [b h a] * * * [pa] *!* [p h a] *! *

PAGE 125

125 APPENDIX A EXPERIMENT INSTRUCTIONS Instructions Given to Subjects on the First Day of the E xperiment (The instructions were given by an interpreter who was a multilingual speaker of Hindi, Mizo and Bodo for data collected from Mizo and Bodo speakers and a bilingual speaker of Hindi and Tamil for data collected from speakers of Tamil language) Welcome here and thank you for showing your interest in participating in this experiment and coming here all the way to contribute to my study. You will be participating in a study of the perceptual training of new sounds from another language. This training study wil l carry on for 4 sessions. Next three days you will take part in one session each day. Each session is expected to last 3040 mins. Today is your first session. In this session, you will have to do one task. Practice Trials In this test, you will liste n to pairs of syllables like [pa]~ [na] or [la]~[ka] separated by a pause. After listening to both these sounds you need to decide whether the two sounds were same or different. Do not focus on intonation or speaker’s voice instead base your judgment on w hether you hear them as instances of same syllable or if you hear them as instances of two different syllables. If you hear them as same press key ‘S’. If you hear the syllables as being different press key ‘K’. After listening to both the syllables try to respond as accurately and quickly as possible. P retest This section is same as the practice trial session except it will be longer. This will take 1015 mins. The pairs of syllables will be separated by a pause. Sometimes this pause will be shorter and s ometime it will be longer. You should wait to listen to both the syllables before responding. After the pretest the participants were asked to take a short 57minute break. Training In this section you will again listen to pairs of syllables like [pa]~ [ba] or [la]~[ka] separated by a pause. After listening to both these sounds you need to decide whether the two sounds were same or different. Respond using ‘S’ if you hear same syllables and ‘K if you hear different syllables. After your response you will see feedback on the screen that displays the correct key for that pair of syllables. This feedback will be displayed for 2 seconds. Pay attention to this feedback and keep it in mind when you listen to similar syllable pairs in the study henceforth. For e.g. if you hear [pa]~[ba] and you press ‘S’ then the screen will display ‘K’. It means that the two syllables are different. Try to keep it in mind that [pa] and [ba] are different syllables and not same

PAGE 126

126 when you hear a similar pair of syllables. This will take 1015mins. After the training the participants were asked to take a short break. P osttest Now you will listen to pairs of syllables and you will need to decide if you hear same syallables or different. Press ‘S’ for same syllables and ‘K’ for different syllables. You will not get any feedback in this section of the experiment. But keep in mind what you just learnt from feedback about the sameness or differentness of different syllable pairs. This will take 1015 mins. Thank you.

PAGE 127

127 APPENDI X B STIMULI Tokens U sed for P retest , P osttest and R etention T est 1. [k2a] [s1a] 2. [p1a] [p2a] 3. [p3a] [n1a] 4. [p2a] [ph 1a] 5. [l1a] [bh 1a] 6. [p3a] [b1a] 7. [m1a] [ph 2a] 8. [p1a] [bh 1a] 9. [kh 1a] [n2a] 10. [b1a] [p1a] 11. [r1a] [b3a] 12. [b2a] [ph 2a] 13. [k2a] [s1a] 14. [b1a] [bh 2a] 15. [k1a] [k2a] 16. [k2a] [g1a] 17. [k3a] [kh 2a] 18. [k1a] [gh 3a] 19. [g1a] [k1a] 20. [g3a] [kh 3a] 21. [g1a] [gh 1a] 22. [ph 1a] [ph 1a] 23. [ph 2a] [bh 2a] 24. [bh 3a] [bh 3a] 25. [k2a] [s1a] 26. [bh 1a] [ph 1a] 27. [kh 2a] [gh 2a] 28. [gh 1a] [gh 3a] 29. [gh 2a] [kh 1a] 30. [p2a] [p2a] 31. [p3a] [ph 3a] 32. [p1a] [b1a] 33. [p2a] [bh 2a] 34. [b2a] [p2a] 35. [b3a] [ph 3a] 36. [b2a] [bh 2a] 37. [k2a] [s1a] 38. [k3a] [g3a] 39. [k1a] [kh 1a] 40. [k2a] [gh 2a] 41. [g2a] [k2a] 42. [g3a] [k3a]

PAGE 128

128 43. [g1a] [kh 1a] 44. [g2a] [gh 2a] 45. [ph 3a] [bh 3a] 46. [bh 2a] [bh 1a] 47. [bh 3a] [ph 2a] 48. [kh 3a] [gh 3a] 49. [k2a] [s1a] 50. [gh 2a] [kh 2a] 51. [p3a] [ph 3a] 52. [p1a] [b1a] 53. [p2a] [bh 2a] 54. [b2a] [p2a] 55. [b3a] [ph 3a] 56. [b2a] [bh 2a] 57. [k2a] [s1a] 58. [k2a] [k2a] 59. [k3a] [g3a] 60. [k1a] [kh 1a] 61. [k2a] [gh 2a] 62. [g2a] [k2a] 63. [g3a] [k3a] 64. [g1a] [kh 1a] 65. [g2a] [gh 2a] 66. [ph 2a] [ph 2a] 67. [ph 3a] [bh 3a] 68. [bh 1a] [bh 1a] 69. [bh 2a] [ph 2a] 70. [kh 2a] [kh 2a] 71. [kh 3a] [gh 3a] 72. [gh 1a] [gh 1a] 73. [k2a] [s1a] 74. [gh 2a] [kh 2a] Tokens Used for G eneralization T est 1. [t1a] [t2a] 2. [t2a] [d3a] 3. [t3a] [th 1a] 4. [t1a] [dh 2a] 5. [d1a] [t2a] 6. [d2a] [d3a] 7. [d3a] [th 1a] 8. [d1a] [dh 2a] 9. [t2a] [t2a] 10. [t3a] [d3a] 11. [t1a] [th 1a]

PAGE 129

129 12. [t2a] [dh 2a] 13. [d2a] [t3a] 14. [d3a] [d2a] 15. [d1a] [th 1a] 16. [d2a] [dh 3a] 17. [t2a] [t3a] 18. [t3a] [d2a] 19. [t1a] [th 1a] 20. [t2a] [dh 3a] 21. [d2a] [t1a] 22. [d3a] [d2a] 23. [d1a] [th 3a] 24. [d2a] [dh 1a] 25. [t3a] [d1a] 26. [t2a] [th 2a] 27. [t1a] [dh 3a] 28. [d3a] [t3a] 29. [d2a] [d2a] 30. [d1a] [th 1a] 31. [d3a] [dh 3a] In all the above tokens subscripts 1,2 and 3 refer to different tokens of that phoneme.

PAGE 130

130 REFERENCE LIST Aggarwal, A. (2011). Acquisition of Hindi Contrasts by English Speakers: An Optimality Theoretic Account. Acta Linguistica Asiatica ,1 (3), 924. Archangeli, D. (1999). Introducing Optimality Theory. Annual Review of Anthropology 28, 531552. Archibald, J. (1993). Language Learnability and L2 Phonology: The Acquisition of Metrical Parameters. Dordrecht: Kluwer Academic Publishers. Archibald, J. (2000). Second Language Acquisition and Linguistic Theory . Oxford, UK: Blackwell. Arunachalam, Elsi Kaiser & Alexander Williams (Eds.), Penn Working Papers in Linguistics 8.1: Proceedings of the 25th Penn Linguistics Colloquium, 7185. Best, C.T. (1995) A direct realist perspective on cross language speech perce ption. In W. Strange (Ed.) Speech perception and Linguistic Experience: Theoretical and Methodological Issues in Cross language Speech Research (pp. 167200). Timonium, MD: York. Best, C.T., McRoberts, G.W. & Goodell, E. (2001). Discrimination of non nati ve consonant contrasts varying in perceptual assimilation to the listener’s native phonological system. Journal of the Acoustical Society of America, 109, 775794. Birdsong, D. (1992). Ultimate attainment in second language acquisition. Language, 68, 7067 55. Birdsong, D. (1997). Constraints on learnability in late L2A. Second Language Research Forum. Birdsong, D., & Molis, M. (2001). On the evidence for maturational effects in second language acquisition. Journal of Memory and Language, 44, 235249. Bley Vroman, R. (1989). What is the logical problem of foreign language learning? In S. Gass & J. Schachter (Eds.), Linguistic Perspectives on Second Language Acquisition (pp. 4168). New York: Cambridge University Press. Boersma, P. (1997). How we learn variat ion, optionality, and probability. Proceedings of the Institute of Phonetic Sciences of the University of Amsterdam, 21,43– 58. Boersma, P. (1998). Functional Phonology. Formalizing the interactions between articulatory and perceptual drives . Ph.D. Disserta tion. University of Amsterdam. The Hague: Holland Academic Graphics. Boersma, P. & Bruce, H. (2001). Empirical tests of the Gradual Learning Algorithm. Linguistic Inquiry , 32(1), 4586.

PAGE 131

131 Boersma, P. and Escudero, P. (2003): Modelling the perceptual developm ent of phonological contrasts with Optimality Theory and the Gradual Learning Algorithm. In Arunachalam, Sudha, Elsi Kaiser and Alexander Williams (eds.) Penn Working Papers in Linguistics 8.1: Proceedings of the 25th Penn Linguistics Colloquium, 7185. B ongaerts, T. (1999). Ultimate Attainment in L2 pronunciation: the case of very advanced late learners. In D. Birdsong (Ed.), Second Language Acquisition and the Critical Period Hypothesis (pp. 133159) . Mahwah, New Jersey: Lawrence Erlbaum Associates. Bon gaerts, T., Planken, B. & Schils, E. (1995). Can late learners attain a native accent in a foreign language? A test of the critical period hypothesis. In D. Singleton & Z. Lengyel (Eds.), The age factor in second language acquisition (pp. 3050). Clevedon: Multilingual Matters. Bongaerts, T., van Summeren, C., Planken, B., &, Schils, E. (1997). Age and ultimate attainment in the pronunciation of a foreign language. Studies in Second Language Acquisition, 19, 447465Jakobovits, L. A. (1968). Implications of recent psycholinguistic developments for the teaching of a second language. Language Learning, 18, 89109. Broselow, E. & Finer, D. (1991). Parameter setting in second language phonology and syntax. Second language research, 7, 3559. Brown, C. (1998). The role of the L1 grammar in the L2 acquisition segmental structure. Second Language Research, 14, 136193 Burnham, D., Kitamura, C., and Vollmer Conna, U. (2002). What’s new, pussycat? On talking to babies and animals. Science 296, 1435. Chomsky (1995). T he Minimalist Program . Cambridge, MA: The MIT Press. Chomsky, N. and Lasnik, H. (1993) Principles and Parameters Theory, in Syntax: An International Handbook of Contemporary Research, Berlin: de Gruyter. Christdas, P. (1988). The Phonology and Morphology of Tamil. Ph.D. Dissertation, Cornell University. Curtin, Suzanne (2001). Children’s Early Representations: Evidence from Production and Perception. Proceedings of the Holland Institute of Linguistics Phonology Conference (HILP 5), Potsdam, Germany. Dutta, I. (2007). Four way stop contrastsin Hindi: An acoustic study of voicing, fundamental frequency and spectral tilt. Ph.D. Dissertation, University of Illinois, UrbanaChampaign. Elliott, A. (1995). Field independence/dependence, hemispheric specialization, and attitude in relation to pronunciation accuracy in Spanish as a foreign language. Modern Language Journal , 79, 356– 371.

PAGE 132

132 Escudero, P. (2005), Linguistic Perception and Second Language Acquisition. Explaining the attainment of optimal phonological categorization. Ph.D. Dissertation. Utrecht University. Escudero, P., & Boersma, P. (2004). Bridging the gap between L2 speech perception research and phonological theory. Studies in Second Language Acquisition, 26, 551585. Escudero, P., & Paul, B. (2003). Modelling the perceptual development of phonological contrasts with Optimality Theory and the Gradual Learning Algorithm. In Sudha Fathman, A. (1975). The relationship between age and second language productive ability. Language Learning, 25, 245– 253. Fer guson, C. (1959). Diglossia. Word,15, 325– 340. Flege, J. (1987a). A critical period for learning to pronounce foreign languages? Applied Linguistics , 8, 162– 177. Flege, J. (1987b). Effects of equivalence classification on the production of foreign language speech sounds. In A. James & J. Leather (Eds.), Sound patterns in second language acquisition (pp. 9– 40). Dordrecht: Foris. Flege, J. (1991). Age of learning affects the authenticity of voice onset time (VOT) in stop consonants produced in a second lang uage. Journal of the Acoustical Society of America, 89, 395--411 Flege, J. (1992a). The intelligibility of English vowels spoken by British and Dutch talkers. In R. Kent (Ed.), Intelligibility in speech disorders: Theory, measurement, and management (pp. 157232). Amsterdam: Benjamins. Flege, J. (1992b). Speech learning in a second language. In C. Ferguson, L. Menn, and C. StoeI Gammon (Eds.), Phonological development, models, research, and applications , (pp. 565604). MD: York, Parkton. Flege, J. (1995). Second language speech learning: Theory, findings and problems. In W. Strange (Ed.), Speech perception and linguistic experience: Theoretical and methodological issues (pp. 233277). Baltimore: York press. Flege, J., & Munro, M. (1994). The word unit in s econd language speech perception and production. Studies in Second Language Acquisition, 16, 381– 411. Flemming, E. (1996). Evidence for constraints on contrast: The dispersion theory of contrast. Chai Shune K. Hsu (Ed.) UCLA Working Papers in Phonology 1, 86106. Gerrits, E., & Schouten, M. E. (2004). Categorical perception depends on the discrimination task . Attention, Perception, & Psychophysics , 66(3), 363376.

PAGE 133

133 GoodinMayeda, L., Renaud, J. & Rothman, J. (2011). Optimality Theoretic L2 reranking and the Constraint Fluctuation Hypothesis: coda nasals in the L2 English of L1 Spanish speakers. In Perez Leroux, A.T. et al. (Eds.), Selected Proceedings of GALANA IV (Generative Approaches to Language Acquisition North America) (pp. 6677). Somerville, MA: Cas cadilla Press. Grosjean, F. (1998). Transfer and language mode. Bilingualism: Language and Cognition, 1(3), 175176. HancinBhatt, B. (2000). Optimality in second language phonology: Codas in Thai ESL. Second Language Research, 16(3), 201232. HancinBhatt, B. (2008). Second language phonology in Optimality Theory. In J.G.Hansen & M. Zampini (Eds.), Phonology and second language acquisition (pp. 117146). Philadelphia/Amsterdam: Benjamins Ioup, G. (1989). Immigrant children who have failed to acquire native English. In S.Gass, C. Madden, D. Preston & L. Selinker (Eds.), Variation in second language acquisition: Psycholinguistic issues (pp. 160– 175). Clevedon, UK: Multilingual Matters. Klein, W. (1995). Language acquisition at different ages. In D. Magnusson (Ed.), The lifespan development of individuals: Behavioral, neurobiological and psychosocial perspectives. A synthesis (pp. 244264). New york: Cambridge University Press. Lenneberg, E. (1967). The biological foundations of language. New York: Wiley Lindblom, B. (1990). Explaining phonetic variation: a sketch of the H&H theory. In W.J. Hardcastle & Marchal (Eds.), A Speech production and speech modelling (pp. 403439). Dordrecht: Kluwer Lively, S., Logan, J., Pisoni, D. (1993). Training Japanese lis teners to identify English /r/ and /l/: II. The role of phonetic environment and talker variability in new perceptual categories. Journal of the Acoustical Society of America, 94(3), 1242– 1255. Lombardi, L. (1999). Positional faithfulness and voicing assi milation in Optimality Theory. Natural Language and Linguistic Theory, 17, 267– 302. Long, M. H. (1990) Maturational constraints on language development. Studies in Second Language Acquisition, 12, 251285. Long, M. H. (2005). Problems with supposed counter evidence to the critical period hypothesis. IRAL, International Review of Applied Linguistics in Language Teaching, 43(4), 287317. MacKain, K. W., Best, C. T., & Strange, W. (1980). Native language effects on the perception of liquids. Journal of the Acoustical society of America, 27. Major, R. (2001). Foreign accent: The ontogeny and phylogeny of second language phonology . Mahwah, NJ: Erlbaum

PAGE 134

134 McCarthy, J. (2001). A Thematic Guide to Optimality Theory . Cambridge: Cambridge University Press. McCarthy, J., & Prince, A. (1993). Prosodic Morphology: Constraint Interaction and Satisfaction. Rutgers University Center for Cognitive Science Technical Report 3. Novoa, L., Fein, D. and Obler, L. (1988). Talent in foreign languages: A case study. In L. Obler & D.Fein (Eds.), The exceptional brain. Neuropsychology of talent and special abilities (pp. 294302). New York: Guilford Press. Neufeld, G. (1977). Language learning ability in adults. A study on the acquisition of prosodic and articulatory features, Working Papers on Bilingualism 12 , 4660. Newport, E.L. (2002). Critical periods in language development. In L. Nadel (Ed.), Encyclopedia of Cognitive Science. London: Macmillan Publishers Ltd./Nature Publishing group. Nosofsky, R.M. (1992) Similarity scaling and cognitive process models. Annual review of psychology, 43, 2553. Oyama, S. (1976). A sensitive period for the acquisition of a nonnative phonological system. Journal of Psycholinguistic Research 5(3), 261 – 283. Oyama, S. (1976). A sensitive period for the acquisition of a nonnative phonological system. Journal of Psycholinguistic Research 5(3), 261 – 283. Pisoni, D. B., & Tash, J. (1974). reaction times to comparisons within and across phonetic categories. Perception & Psychophysics , 15, 285290. Podgorny, P., & Garner, W. (1979). reaction time as a measure of inter and intraobject visual similarity: Letters of the alphabet. Perception & Psychophysics , 26, 3752. Prince, A. and Smolensky, P. (1993). Optimality Theory: Constraint Interaction in Generative Gramm ar . Technical Report 2, Center for Cognitive Science, Rutgers University. Cambridge, MA: MIT Press. Prince, A. and Smolensky, P. (2004). Optimality Theory: Constraint Interaction in Generative Grammar. Blackwell Publishing. Rothman, J. (2008). Why Not All Counter Evidence to the Critical Period Hypothesis Is Equal or Problematic: Implications for SLA. Language and Linguistics Compass , 2(6), 1063– 1088. Schneiderman, E. I., and Desmarais, C. (1988). The talented language learner:Some preliminary findings. S econd Language Research 4, 91109 Schwartz, B. D. & Sprouse, R. (1994). Word order and nominative case in nonnative language acquisition: a longitudinal study of (L1 Turkish) German interlanguage. In T. Hoekstra and B. D. Schwartz (eds.), Language acquisit ion studies in generative grammar (pp. 317368). Amsterdam: John Benjamins.

PAGE 135

135 Schwartz, B. D. & Sprouse, R. (1996). L2 cognitive states and the full transfer/full access model. Second Language Research , 12, 4072. Scovel, T. (1981). The effects of neurological age on nonprimary language acquisition. In R. Andersen (Ed.), New dimensions in second language acquisition research (pp. 33– 42). Rowley, MA: Newbury House. Scovel, T. (1988). A Time to Speak: A Psycholinguistic Inquiry into the Critical Period for Human Language. Rowley, MA: Newbury House. Selinker, L. (1972). Interlanguage . International Review of Applied Linguistics , 10, 209231. Shepard, R. N. (1957). Stimulus and response generalization: A stochastic model relating generaliz ation to distance in psychological space. Psychometrika , 22, 325– 345Rothman, J. & Guijarro Fuentes, P. (2010). Input Quality Matters: Some Comments on Input Type and Age Effects in Adult SLA. Applied Linguistics , 31(2), 301306. Silke, H. (2009). Variatio n in the perception of an L2 contrast: A combined phonetic and phonological account. In: F. Kgler, C. Fery & R. v. d. Vijver (Eds.) Variation and Gradience in Phonetics and Phonology (pp. 79– 105). Berlin: Mouton de Gruyter. Singleton, D. (2005). The criti cal period hypothesis: A coat of many colours. International Review of Applied Linguistics , 43, 269286. Tesar, B. and Smolensky, P. (2000). Learnability in Optimality Theory . Cambridge, MA: MIT Press. Wayland, R., & Li, B. (2005). Effects of two training procedures on cross language perception of tones. In Proceeding of ISCA Workshop on Plasticity in Speech Perception, London, UK. Werker, J.F., Gilbert, J. V. H., Humphrey, K., & Tees, R, C, (1981). Developmental aspects of cross language speech perception. Child Development, 52, 349355 Werker, J. &Logan, J. (1985). Cross language evidence for three factors in speech perception. Perception & Psychophysics , 37, 3544. Werker, J. & Tees, R. (1983). Developmental changes across childhood in the perception of nonnative sounds. Canadian Journal of Psychology , 37, 3544. Werker, J. & Tees, R. (1984). Phonemic and phonetic factors in adult cross language speech perception. Journal of the Acoustical Society of America, 75, 18661878 White, L. (2000). Second language acquisition from initial to final state. In J. Archibald (Ed.), Second language acquisition and linguistic theory (pp. 130155). Oxford: Blackwell Publishers. White, L. (2003). Second language acquisition and Universal Grammar. New York, NY: Cambridge University Press.

PAGE 136

136 White, L. & Genesee, F. (1996). How native is near native? The issue of ultimate attainment in adult second language acquisition. Second Language Research 12, 233265. Wode, H. (1989). Maturational changes of language acquisitional abili ties. In S. Gass, C. Madden, D. Preston, & L. Selinker (Eds.), Variation in second language acquisition, Vol. II: Psycholinguistic issues (pp. 176– 188). Clevedon, UK: Multilingual Matters. Wode, H. (1992). Categorical perception and segmental coding in the ontogeny of sound systems. In C. Ferguson, L. Menn, & C. Stoehl Gammon (Eds.), Phonological development: Models, research, implications (pp. 605– 631). Timonium, MD: York Press. Wode, H. (1993). The development of phonological abilities. In K. Hyltenst am & A. Viberg (Eds.). Progression and regression in language: Sociocultural, neurophysical and linguistic perpectives (pp. 415438). New York: Cambridge University Press. Wode, H. (1994). Nature, nurture, and age in language acquisition: The case of speech perception. Studies in Second Language Acquisition, 16, 325345. Wiltshire, C. (2005). The “Indian English” of TibetoBurman language speakers. English World Wide, 26 (3), 275300. YoungScholten, M. (1994). Positive evidence and ultimate attainment in L2 phonology.' Second Language Research, 10, 193214. YoungScholten, M. & J. Archibald (2000). Second language syllable structure. In J. Archibald (Ed.) Second Language Acquisition and Linguistic Theory ( pp.64101). Oxford: Blackwell. Zvelebil, K. (1970) . Introduction to the Historical Grammar of the Tamil Language. Spolu s Jaroslavem Vackem. Academia, Prague.

PAGE 137

137 BIOGRAPHICAL SKETCH Ashima Aggarwal was born in New Delhi, India. She received her b achelor’s degree in e lementary e ducation in 2007 from Delhi University, India. After that, she completed her Master’s in Linguistics in 2009 from Jawaharlal Nehru University, India. Her m aster’s thesis focused on reduplication in a NorthEast Indian Language, Tagin. She enrolled in the PhD program in Linguist ics at the University of Florida in f all 2009. During her time at the University of Florida, she worked as a T eaching A ssistant in the Department of Linguistics. Her doctoral degree was supported by the Language Learning Dissertation Grant from University of Michigan and the Hazen E. Nutter Scholarship from the University of Florida. One of the articles she published during her PhD was funded by Open Access Publishing Fund, University of Florida. Since January 2014, she is working as Speech Linguistic Project Manager at Google.