<%BANNER%>

First-Language Transfer and Universal Markedness in Second-Language Production and Perception of Word-Final Obstruents a...

Permanent Link: http://ufdc.ufl.edu/UFE0022664/00001

Material Information

Title: First-Language Transfer and Universal Markedness in Second-Language Production and Perception of Word-Final Obstruents and Obstruent Clusters
Physical Description: 1 online resource (55 p.)
Language: english
Creator: Chappell, Milla
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: esl, language, mandarin, obstruents, syllable
Linguistics -- Dissertations, Academic -- UF
Genre: Linguistics thesis, M.A.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: While all adult second language learners are influenced by their first language and by universal principles of markedness, Chinese learners of English have a unique challenge as they move from a relatively unmarked language to one with highly complex syllable structures and marked segments. Our study examined the role of first-language influence and universal markedness in adult Chinese learners of English. More specifically, it describes the production of word-final consonant clusters; also, it examines the perception of word-final obstruents and obstruent clusters by listeners from different language backgrounds, Mandarin, Spanish, and English, in order to explore the effect of first-language influence on perceptual cues. Overall findings of this study show that Chinese learners of English have difficulty producing syllable structures that are disallowed in their own language; additionally, it seems that listeners are attuned to different perceptual cues, as all three listening groups performed differently on the perception task, with the Chinese group having the greatest level of understanding and the Spanish group having the lowest. Findings also suggest that the place of articulation of the key segment and its position in the phrase also predicts success in production and perception.
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.
Statement of Responsibility: by Milla Chappell.
Thesis: Thesis (M.A.)--University of Florida, 2008.
Local: Adviser: Wiltshire, Caroline R.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022664:00001

Permanent Link: http://ufdc.ufl.edu/UFE0022664/00001

Material Information

Title: First-Language Transfer and Universal Markedness in Second-Language Production and Perception of Word-Final Obstruents and Obstruent Clusters
Physical Description: 1 online resource (55 p.)
Language: english
Creator: Chappell, Milla
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: esl, language, mandarin, obstruents, syllable
Linguistics -- Dissertations, Academic -- UF
Genre: Linguistics thesis, M.A.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: While all adult second language learners are influenced by their first language and by universal principles of markedness, Chinese learners of English have a unique challenge as they move from a relatively unmarked language to one with highly complex syllable structures and marked segments. Our study examined the role of first-language influence and universal markedness in adult Chinese learners of English. More specifically, it describes the production of word-final consonant clusters; also, it examines the perception of word-final obstruents and obstruent clusters by listeners from different language backgrounds, Mandarin, Spanish, and English, in order to explore the effect of first-language influence on perceptual cues. Overall findings of this study show that Chinese learners of English have difficulty producing syllable structures that are disallowed in their own language; additionally, it seems that listeners are attuned to different perceptual cues, as all three listening groups performed differently on the perception task, with the Chinese group having the greatest level of understanding and the Spanish group having the lowest. Findings also suggest that the place of articulation of the key segment and its position in the phrase also predicts success in production and perception.
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.
Statement of Responsibility: by Milla Chappell.
Thesis: Thesis (M.A.)--University of Florida, 2008.
Local: Adviser: Wiltshire, Caroline R.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022664:00001


This item has the following downloads:


Full Text

PAGE 1

FIRST-LANGUAGE TRANSFER AND UN IVERSAL MARKEDNESS IN SECONDLANGUAGE PRODUCTION AN D PERCEPTION OF WORD-F INAL OBSTRUENTS AND OBSTRUENT CLUSTERS By MILLA CHAPPELL A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS UNIVERSITY OF FLORIDA 2008

PAGE 2

2008 Milla Chappell

PAGE 3

TABLE OF CONTENTS page LIST OF TABLES ................................................................................................................ .......... 4 LIST OF FIGURES ............................................................................................................... ......... 5 A BSTRACT. ..................................................................................................................... .............. 6 1 BACKGROUND AND METHODS ....................................................................................... 7 Introduction ............................................................................................................... ............... 7 Background ................................................................................................................. ............. 7 Second Language Acquisition ......................................................................................... 7 Chinese Learners of English .......................................................................................... 10 Rationale for this Study ................................................................................................... ...... 16 Methods ................................................................................................................................. 18 Production Task ............................................................................................................ 18 Perception Task ............................................................................................................ .. 21 Expectations ............................................................................................................... ............ 22 2 RESULTS ................................................................................................................. ............. 25 Introduction ............................................................................................................... ............. 25 Production Results ......................................................................................................... ........ 25 Perception Results ......................................................................................................... ......... 27 Group Effects .............................................................................................................. ........... 28 3 ANALYSIS AND CONCLUSIONS ..................................................................................... 35 Introduction ............................................................................................................... ............. 35 Analysis ................................................................................................................................. 35 Optimality Theory .......................................................................................................... 35 Modeling Learning in the Gradual Learning Algorithm ................................................ 38 Conclusions ............................................................................................................................ 42 LIST OF REFERENCES ..51 B IOGRAPHICAL SKETCH

PAGE 4

LIST OF TABLES Table page 1-1 Phoneme inventory: Mandarin, English, and Spanish. ......................................................24 2-1 Overall production results ................................................................................................ ..31 2-2 Phrase position results........................................................................................................32 2-3 Effect of place of articulation on perception scores: percent correct ................................32 2-4 Effect of sentence position on pe rception scores: percent correct ....................................32 2-5 Perception of codas by group and coda type: percent correct ..........................................32

PAGE 5

LIST OF FIGURES Figure page 2-2 Effect of place of articulat ion on perception scores: percent correct ..............................33 2-3 Effect of sentence position on perception scores: percent correct ...................................34 2-4 Perception of codas by group a nd coda type: percent correct .........................................34 3-1 Obstruent devoicing as the preferred strategy ...................................................................46 3-2 Consonant deletion as the preferred strategy .....................................................................46 3-3 Three-consonant coda simplification .................................................................................47 3-4 Deletion and devoicing .................................................................................................... ..47 3-5 Example of ranked constraints in stochastic OT (Boersma, 2003) ...................................47 3-6 Possible output of ranking in stochastic OT ......................................................................48 3-7 Ranking values and tableau for th e initial state Mandarin Chinese ................................48 3-8 Ranking values and tableau at 300 trials ............................................................................49 3-9 Ranking values and tableau at 600 trials ............................................................................49 3-10 Ranking values and tableau for output [bulb]. ...................................................................50 3-11 Initial state gram mar: input /markt/ ...................................................................................50 3-12 GLA after 600 trials ...........................................................................................................50 3-13 GLA after 1200 trials .........................................................................................................51

PAGE 6

Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Arts FIRST-LANGUAGE TRANSFER AND UN IVERSAL MARKEDNESS IN SECONDLANGUAGE PRODUCTION AN D PERCEPTION OF WORD-F INAL OBSTRUENTS AND OBSTRUENT CLUSTERS By Milla Chappell August 2008 Chair: Caroline Wiltshire Major: Linguistics While all adult second language learners are influenced by their first language and by universal principles of markedne ss, Chinese learners of English have a unique challenge as they move from a relatively unmarked language to one with highly complex syllable structures and marked segments. Our study examined the role of first-language influence and universal markedness in adult Chinese learners of English. More specifically, it describes the production of word-final consonant clusters ; also, it examines the percepti on of word-final obstruents and obstruent clusters by listeners from different language backgrounds, Mandarin, Spanish, and English, in order to explore the effect of first-language influen ce on perceptual cues. Overall findings of this study show that Chinese learners of English have difficulty producing syllable structures that are disallowe d in their own language; additionally, it seems that listeners are attuned to di fferent perceptual cues, as all three listening groups performed differently on the perception task, with the Chinese group having the greatest level of understanding and the Spanish group having the lowe st. Findings also suggest that the place of articulation of the key segment a nd its position in the phrase also predicts success in production and perception.

PAGE 7

CHAPTER 1 BACKGROUND AND METHODS Introduction While all adult second language learners are influenced by their first language and by universal principles of markedne ss, Chinese learners of English have a unique challenge as they move from a relatively unmarked language to one with highly complex syllable structures and marked segments. This study seeks to examine the role of first-language influence and universal markedness in adult Chinese learners of English. More specifically, it describes the production of word-final consonant clusters ; also, it examines the percepti on of word-final obstruents and obstruent clusters by listeners from different language backgrounds, Mandarin, Spanish, and English, in order to explore the effect of fi rst-language influence on perceptual cues. What follows in this chapter is an overvi ew of the relevant literature as well as a description of the methodology and the expectati ons of the study. Overall findings of this study show that Chinese learners of English have difficulty producing syllabl e structures that are disallowed in their own language; additionally, it seems that listeners are attuned to different perceptual cues, as all three listening groups performed differently on th e perception task, with the Chinese group having the greatest level of understanding and the Spanish group having the lowest. Findings also suggest that the place of articulation of the key segment and its position in the phrase also predicts success in production and perception. Background Second Language Acquisition When learning a language, it is generally argued th at earlier contact is better, as learners who are exposed to a new language at a young age te nd to become more proficient in a shorter amount of time than those who begin learning in adulthood. The critical period hypothesis for language learning states that t here is a limited developmental period during which it is possible

PAGE 8

to acquire a language, be it L1 or L2, to normal, nativelike levels. Once this window of opportunity is passedthe ability to learn language declines (Birdsong, 1992, p. 1). Firstlanguage research has for some time proposed that as humans mature neurologically, the ability to learn the L1 decreases (Le nneberg, 1969), and examples of la te learners of L1s show the negative effects of withhol ding language until later in life (Curtiss 1977, 1989). Discussions of the critical period hypothesis have also b een applied to second-language learning. When learning a second language, it is clear that child ren are more successful than adults (Birdsong, 1992, p. 4), and researchers such as Hurford (1991) and Pinker (1994) have attributed this to physiological changes that acco mpany puberty, such as loss of neural plasticity and loss of the language learning mechanism. Ho wever, research has shown that adult second language learners can be quite successful in la nguage learning; in many documented cases, latelearners of a second language are able to atta in native-like proficiency. Birdsong (1992) found native-like competence in American learners of French who had moved to France at a mean age of 28.5 years; Mayberry (1993) shows that late learners of American Sign Language attained native-like proficiency; and White & Genesee (1996) studied acquisiti on of English by native speakers of French and found that many native-lik e speakers had first studied English after age 12. These studies are just examples from a body of literature that has shown that adult learners of a second language are often able to attain native-like competence, even if the onset of language learning occurr ed after puberty. So, although learning a second language may be more difficult for adults than for children, it is certainly not impossible. And, success in second language learning depends on other factors besides age of acquisition, such as similarity between the L1 and L2 and learner adaptability. For example, the Speech Lear ning Model (SLM) developed by Flege and his

PAGE 9

colleagues (1995) states that L1 phonetic categories can be adap ted later in life as a second language is learned, and that learners can percep tually relate positional allophones in the L2 to the closest positionally defined allophone (or sound) in the L1 (p. 238). Other models of speech learning state that a learners ability to perceive and produce nonnative contrasts depends on universal markedness of language segments, and articulatory and acoustic similarity of the L1 and L2. Bests perceptual assimilation model (PAM) hypothesizes that a speakers ability to pe rceive and produce non-native contrasts will be largely dependent upon the articulatory properties of the sounds in the speakers first language(s). Best (1995) says that when learning an L1, childre n discover correspondences betwee n higher-level invariants of relations among gesturesand lingu istically functioning elements (p. 178) and create gestural categories into which all sounds are placed. First-language sounds and second-language sounds that resemble those in the fi rst language are more easily cate gorized; however, second-language learners may have difficulty knowing how to classify sounds that dont fit pre-existing categories. Another model by Kuhl proposes that we comp are heard sounds to stored prototypes. Kuhl et al. (1992) describes prot otypes as ideal repr esentatives of a give n phonetic category (p. 606) as identified by adult speakers of that langua ge. For learners of an L2, the L1 prototype could make L2 sounds more difficult to understa nd and therefore produce, as the L1 categories and prototypes could pull cl ose L2 sounds into the wrong per ceptual space, producing what Kuhl calls the magnet effect. While these models focus on the similarity of the L1 and L2, the Markedness Differential Hypothesis (Eckman, 1977) addresse s universal markedness, as it states that certain language structures are more difficult to produce, and ar e therefore more marked cross-linguistically.

PAGE 10

And, the hypothesis makes claims about language learning that are based on this universal principle of markedness, namely, that language le arners who have more marked structures in their L1s should have an easier ti me learning languages which have less marked structures while on the other hand, learners with unmarked structures in the L1 will have a difficult time acquiring the marked structure of many L2s. In general, these models predict that a ll individuals acquiring the phonology of a second language after childhood will face such challenges as a decline in neural plasticity, interference from a first language, and the markedness of a second language. Adult Chinese learners of English are certainly no exception, and as the segmental phonology of their first language is less marked than the learned language of Englis h, they are faced with unique difficulties in pronunciation and perception. Chinese Learners of English Studies have identified that for adult Chines e learners of English, syllable-final obstruents and obstruent clusters are difficult to produce. According to these theoretical models, this difficulty should be a result of the difference be tween the L1 (Mandarin) and the L2 (English), for the Mandarin language is more specific than English with respect to the types of consonants that are allowed, and in which sy llable position they may be f ound. The difficulty should also be attributed to universal marke dness, for the English language contains structures that are highly marked compared to the stru ctures found in Mandarin. Mandarin Chinese prefers a simple (CV) syllable structure and does not allow any obstruents or consonant clusters in word-final position; however, the target language of English allows single voiced obstruents /b,d,g/ and voicele ss obstruents /p,t,k/ in word-final position as well as complex codas such as CC (eg. learn, soft) and CCC (eg. worl d, first) and CCCC (eg.

PAGE 11

twelfths) (Cheng, 1991). English also differs from Mandarin in its use of marked structures; in general, English allows segments that are high ly marked and are therefore difficult to acquire cross-linguistically. For example, research has shown that voi ced obstruents are more marked in coda position and are th erefore more difficult for second language learners to acquire than voiceless obstruent codas, regard less of whether the learners first language has this voicing distinction, (Eckman, 1981; Fl ege & Davidian, 1984; Wang, 1995; Broselow et al., 2004). While English allows all types of voiced obstrue nts in coda position, Mandarin Chinese does not have such a voicing distinction, a nd this creates difficulty for Chin ese learners of English. According to Broselows Syllable Structure Transfer Hypothesis, the differences between syllable structures will cause a problem for Ma ndarin learners of English. Broselow (1984) states, when the target language permits syllable structures which are not permitted in the native language, learners will make errors which involve altering these structures to those which would be permitted in the native language (p. 263). Studi es have shown that Broselows assertion is true for Chinese speakers of English, for in gene ral, when faced with word-final obstruents, many native speakers of Mandarin epenthesize a schw a after the final consonant or simply delete the final consonant. Additionally, participants may devoice the final obstruent in an effort to both retain the original form while making the form less marked and therefore easier to articulate (Anderson, 1983; Yin, 1984; Eckman, 1981; We inberger, 1988; Wang, 1995). Though Mandarin treatment of word-final obstruent clusters has not been well documented, segment deletion and obstruent devoicing ma y be common simplification strate gies to ease articulation as well. Studies have shown that these simplificati on strategies are comm on cross-linguistically since open (CV) syllables are pref erred cross-linguistically to closed syllables (CVC), and when

PAGE 12

words do contain obstruent codas, voiceless obs truents are easier to articulate than voiced obstruents. Since English cont ains both voiced and voiceless obs truents in word-final position, language learners from various ba ckgrounds have difficulty with the complex syllable structure, and they use similar strate gies to ease production. An early study by Eckman (1981) used word lists to examine word-final obstruent production by two native speakers of Mandarin. Eckman found that early learners of English used epenthesis to offset the CVC structure, and he attributed this to inte rlanguage transfer and to the markedness hypothesis. According to Eckman (1977), those areas of the target language which differ from the native language and are mo re marked than the native language will be difficult (p. 321). In this case, the English structure CVC is marked, so participants in his study pronounced /rob/ as [rab ] and /started/ as [statid ], therefore creating a fi nal syllable structure of CV which is most common in Mandarin Chin ese and is least marked universally. Eckman found no examples of deletion or devoicing in his study; however, the number of participants was small, and these learners were highly profic ient, high-intermediate/a dvanced students in a University intensive English program, which could have contributed to their choice to use a this form. Studies by Yin (1984) and Wang (1995) replicate the findings of Eckman. Yin examined production of word-final stops, word-final nasals, and word-final consonant clusters by native speakers of Mandarin, Taiwanese, and Hakka. Ten Mandarin speakers participated in the study by producing free speech on a given topic. Yin analyzed the word-final stops, nasals, and consonant clusters for changes, and like Eckman, she found that participan ts often produced an epenthetic schwa as well as other vowels. The study also documents deletion of final stop consonants, though this change wa s less frequent than epenthesis Wangs study provides further

PAGE 13

support for the markedness hypothesis, for alt hough word-final stops are not allowed in Mandarin in either voiced or voiceless form, when native Mandarin speakers read words with word-final voiced and voiceless stops aloud, th ey made more errors producing voiced stops, which are universally more marked than the voiceless counterpart. The practice of devoicing final voiced stops has also been explained by the notion of recoverability. According to this idea, language learners may c hoose to devoice final obstruents to partially preserve the underlying form, inst ead of deleting the final stop completely. Weinberger (1987) refers to the impact of context on recoverabilit y, stating that when words are read in isolated contexts such as a word lists or short frame sentences, participants will try to retain the recoverable form and will therefore ep enthesize or devoice more often than delete. However, when words are found in spontaneous, natural speech in which the context may serve as a cue for meaning, native speakers of Mandarin will be more inclined to delete the final obstruent. Weinbergers (1988) study found that Mandarin lear ners often devoiced final voiced obstruents in formal context-fr ee situations, and Eckman (1981) reported similar findings for native speakers of Cantonese. Though complete deletion of a final stop c onsonant is found much less often in the literature, there are two studies by Tarone (1980) and Anderson (1983) that document this type of deletion by native speakers of Cantonese and Mandarin, and it seems that the task could have influenced the decision to delete rather than epenthesize or devoice. Tarone asked Cantonese speakers to describe a short cart oon in their own words, and Anders on asked participants to talk about their own cultural festivals and holidays. The fact that speakers were producing target words in an informal, even personal context coul d have contributed to the speech being produced

PAGE 14

less carefully and with less phonolog ically recoverable ta rget words, as the context provided the information necessary for unde rstanding to occur. While studies have not specifically documented Chinese production of obstruent clusters in word-final position, it could be assumed based on the previous findings that these clusters will also cause difficulty. Mandarin prefers a simple (CV) syllable structure and does not allow consonant clusters in word-final position; however, the target language of English allows complex codas such as CC (eg. learn, soft) a nd CCC (eg. world, first) and CCCC (eg. twelfths) (Cheng, 1991). Most commonly, English final cluste rs are made up of two consonants; however, it is possible to have as many as four consonants in a word-final cluster. Any consonant may be found word-finally except h, r, w, and j, and wh ere two-consonant cluste rs are found, there is either a consonant preceded by a pre-final cons onant, or a consonant followed by a post-final consonant (Roach, 2000, p. 73). Pre-final conson ants, or those that precede consonants in clusters, are m, n, l, r and s as seen in bump [bamp] bent [b nt] bank[bk] belt[belt] ask [sk], and burp [b rp] (Roach, 2000, p. 73). Post-final consonants, or those that follow other consonants in cluste rs are s, z, t, p, d, and as seen in bets [b ts] beds [b dz] backed [bkt] bagged [bgd] bum p [bamp], and eighth [eIt ] (Roach, 2000, p. 73). In addition to two-consonant clusters, there are two types of th ree-consonant clusters: pre-final + final + postfinal, as seen in helped [helpt] and banks [ban ks], or more than one pos t-final cluster, as seen in fifths [fif s] and next [nekst] (Roach, 2000, p. 73) Most four-consonant clusters are analyzed as a pre-final + final + post-fina l + post-final, as in twelfths [twelf s] and prompts [prompts]. While research on Chinese simplification stra tegies has described single-consonant codas or complex onsets (Tarone, 1980; Eckma n, 1981; Anderson, 1983; Weinburger, 1988; Yin,

PAGE 15

1984; Flege et al., 1996; Wang, 1995), cross-linguistic research on first and second-language learning has shown that that similar simplifica tion strategies take pl ace when learners are acquiring consonant clusters in coda position (Roberts et. al., 1990; Ohala, 1999; Jongstra, 2003; Pater & Barlow, 2003; Yoo, 2004; Wiltshire, 2006). Young children and beginning secondlanguage learners often delete one segment of the cluster or epenthesize a vowel between the cluster consonants while older children and more advanced learners often produce clusters with less marked features. When deletion occurs, it is common that language learners will retain the least sonorous segment, and when epenthesis occurs, it is usually an unmarked schwa which is inserted. Research on first-language acquisition has s hown that the younger children are, the more likely they are to simplify consonant clusters by deleting one of the se gments (Roberts et. al., 1990; Jongstra, 2003). According to Roberts et. al. (1990), there is a si gnificant correlation between the age of the child and the rate of cluster reduction, and it is most often the most sonorous segment which is deleted (Ohala, 1999; Pater & Barlow, 2003). For all children, twosegment clusters are easier to acquire than three-segment clusters (McLeod, 2001), and by the age of two, children are able to produce consonant clusters; howeve r, the clusters are typically simplified in some way and are therefore less marked than those produced by adults (McLeod, 2001). Second-language research shows simila r trends. A longitudi nal study by Yoo (2004) examined Korean learners of English, finding that complex codas were acquired after simple ones and that the most common cluster simplifi cation strategy was consonant deletion. These cross-linguistic findings seem to suggest that universal markedness governs consonant cluster production. However, there is also evidence that first-language background

PAGE 16

influences acquisition of second-language cluste rs. Wiltshire (2006)s study of Indian speakers of English shows that speaker s of Indian English simplify complex codas by deleting one consonant, and she also shows that rate of de letion is contingent upon first-language background, for speakers of languages that did not allow clusters more frequently deleted a segment of the cluster than did speakers of la nguages which allowed clusters. Rationale for this Study While many studies have clearly described the production of English word-final stops by Mandarin learners, a detailed description of Chinese production of English word-final consonant clusters is missing in the liter ature. And, in addition to de scribing production patterns of Mandarin speakers, it is also im portant to examine whether th ese speakers are understood by the people around them when producin g these particular sounds. Previous perception studies have yielded some information about how segmental and syllable features influence cross-cultural unde rstanding. Magen (1998) examined native speakers of Spanish, showing that altered syllable structure (such as vowel epenthesis) and consonant factors, such as /t /substitution and /s/ deletion (p. 392) co ntributed to higher accent ratings by native speakers of American English. When ex amining intelligibility in native speakers of English, Bradlow et. al. (1996) found that phonol ogical changes such as vowel space reduction, consonant deletion, and incorrect at tachment of segments to syllables were related to decreased intelligibility. However, these studies have focused on native-speaker ratings, and since many Chinese learners of English liv e and work in multicultural en vironments such as national universities or international businesses, it is important to examine not only native speaker judgments of Mandarin-accented speech but to also elicit judgments from other language groups.

PAGE 17

Worldwide, Mandarin Chinese is spoken by the largest number of people; 873 million people speak Chinese as a first language. Englis h is the second most popular language with 340 million native speakers, and Spanish is the fi fth most common language with 322 million native speakers (Gordon, 2005). In the United States, Ch inese is the second most common foreign language and is spoken by .78% of the population acco rding to the U.S. Census Bureau (2000). The most common foreign language is Spanis h, as it is spoken by 10.71% of the population according to U.S. Census Bureau (2000). The number of Spanish speakers in the United States outnumbers all other foreign language groups by more than 20 million speakers. These two language groups comprise a sign ificant percent of the population; therefore, examining the interaction between these two la nguage groups in an English-speak ing environment is important. Another reason for studying perceptions of Mandarin and Spanish groups as well as native speakers of English is the difference between the syllables structure and phoneme inventories of the three language s. When specifically examining voiced and voiceless obstruents, Mandarin is more restrictive than either Spanis h or English (see Table 1-1). As mentioned previously, Mandarin allows voice less obstruents in word-medial position, but it doesnt allow voiced obstruents at all; additionall y, voiceless obstruents are not allowed in word-final position. Spanish also allows both voiced and voiceless obst ruents; however, only th e voiced obstruent /d/ is allowed in word-final pos ition (Bedore, 1999). The Englis h language allows voiced and voiceless obstruents in word-final position. With regard to syllable stru cture, the Mandarin language is ag ain the most restrictive, for Mandarin prefers a CV syllable structure and does not allow syllable-final consonant clusters. Spanish allows two-consonant clusters syllable -finally, and English allows as many as fourconsonants in a syllab le-final cluster.

PAGE 18

The differing phoneme inventories and syllable structures are important, for if L1 phonetics influences L2 perception and production, then it would be possible to see perception differences between language groups, and it ma y be possible to describe speech cues that influence understanding. Therefore, the aim of this study is to answer the following questions: do adult Chinese learners of English simplify comp lex codas in the same way that research has documented simplification of simple obstruent codas? And, does listener group play an important role in understanding? For example, are Chinese listeners able to better understand Mandarin-accented speech than English or Spanish lis teners? And if so, wh at are the cues that listeners are attuned to? Methods This experiment consists of two tasks: a production task and a per ception task. Eleven Chinese learners of English were recorded performing multiple speaking tasks, and native speakers of Mandarin, English, and Spanish listened to the speech of two speakers in order to judge the accuracy of the speech. The speaki ng data was analyzed for trends in acoustic production, and the perception data was analyzed us ing simple t-tests and multiple comparisons. Production Task Recorded materials were collected from eleven Chinese learners of English (8f/3m). Participants recorded three types of speech: a two-minute segment of free speech describing life in the United States, a read lis t of forty-six longer conversationa l sentences, and a read list of sixty short frame sentences with embedded target words. Speech was recorded in a soundattenuated room with a Mara ntz PMD660 solid state recorder and head-mounted Shure SM10A microphone that was situated 1 inch from the participants mouth.

PAGE 19

The present study examines only the fifty-si x short frame sentences; however, data was collected using three different me thodologies in order to elicit da ta produced in three different environments with varying levels of naturalness and control. Usi ng lists of words or short frame sentences to elicit speech from participants is somewhat unnatural; however, this method for collecting data allows for key words to be pr oduced in an extremely controlled environment without intonation or personal word choice. L onger sentences or paragraphs allow for more natural intonation and pronunciation while still stipulating the word s that should be read. And finally, free speech is the most natural and unhindered; however, the researcher has little control over the words that could be produced, and the speaker has the ability to avoid words and structures that are difficult to pronounce. Previous literature examining word-final consonants and consonant clusters most commonly uses word lists and free speech to elic it data, and differences between the results suggest that methodology can infl uence production strate gies of Chinese learners of English. Tarone (1980), Anderson (1983), and Yin (1984; in Wang, 1995) use free speech to elicit data, and they find less instances of epenthesis a nd deletion than Eckman (1981) and Weinberger (1983), who both use a combination of word lis ts, paragraph reading, and free speech. This could be attributed to recovera bility, as the meaning of words produced in free speech can be recovered based on context. Based on these findings the goal of this larg er study is to analyze data collected in all three environments; howev er, the present paper examines only the frame sentences in order to examine the key words pr oduced in controlled environments in both sentence medial and se ntence final position. Thirty-two of these short frame sentences cont ained word-final obstruents; twenty-four of the frame sentences contained words ending in co nsonant clusters. The words containing word-

PAGE 20

final obstruents were divided into eight sets of minimal pairs that were read in two different frame sentences, and each pair differed only in the voicing of the word-final obstruent. The obstruents examined in this experiment were bilabial /b/-/p/, alveolar /d/-/t/, and velar /g/-/k/. The minimal pairs were /mob/-/mop/ and /cub/-/cup/ for the /b/-/p/ contra st; /mad/-/mat/, /god//got/, /bid/-/bit/ for the /d/-/t/ co ntrast; and /pig/-/pick/, /bag/-/back /, and /tag/-/tack/ for the /g/-/k/ contrast. Twenty-four of these short frame sentences included word-final clusters made up of two or three consonants and contai ning at least one voiced or voiceless obstruent. The clusters examined in this experiment were /ld/, /lb/, /lpt/, /rp/, /rb/, /rd/, /mp/, /rkt /, /rk/, /rst/, /nk/, and /vd/. The read words, in standard English orthography are gold, bulb, sculpt, burp, curb, bird, lamp, marked, park, burst, bank, and moved. Each word was read twice, once in the sent ence-final context of Repeat __. and once in the sentence-medial context of Repeat __ agai n. The two types of frame sentences were ordered randomly in the list, and the target wo rds were spaced so that minimal pairs did not occur in back-to-back sentences. In order to elicit the most natural speech as possible, participants were given time to practi ce pronouncing the sentences while wearing the microphone before recording bega n; additionally, many partic ipants asked for pronunciation advice from the researcher. Since read speech is not as natural as that produced in real-world situations, these measures were taken to ensure maximal comfort and naturalness from the participants. All speaking participants were intermediate or high-intermediate learners of English enrolled in graduate school at the University of Florida. Proficienc y level was measured by TOEFL-IBT or SPEAK test scores; TOEFL-IBT scores of 24 or SPEAK test scores of 45 were accepted as evidence of intermed iate standing while TOEFL-IBT scores of 26 or SPEAK test

PAGE 21

scores of 50 were accepted of evidence of highintermediate standing. In addition, a selfreported proficiency questionnaire was completed by each participant. This form asked for age of beginning study in English, time spent in an English-speaking county, time spent speaking English per day, and the language spoken at home, as well as a 5-point likert scale rating of proficiency in reading, writing, speaking, and list ening in English. All participants started studying English after the ag e of nine and had been living in the United States for at least one month and at most four years. Additionally, all participants gave their speaking ability a likert scale rating of at least three, which was also accepted as ev idence of intermediate standing. All participants were paid for their time. After all speaking data was collected, the data was transferred to a computer using a USB 2.0 cable and was analyzed using Praat ve rsion 4.5.19 by Boersma & Weenink (1992-2007). Additionally, the key words were specifically an alyzed for vowel length and aspiration of the obstruent in order to determine the production pa tterns among speakers. Aspiration intensity was measured by highlighting the main burst in Praat and choosing get intensity from the intensity pull-down menu. Vowel length was measur ed from the onset of vowel voicing to the offset. The perception study examines all fifty-six frame sentences while the production study examines only the cluster data. Perception Task After the recorded material was collected, a perception task was completed. In this experiment, listeners from three different language backgrounds listened to the speech of two speakers and completed a judgment task and a rating task. Listen ers were native speakers of English (8), native speak ers of Spanish (6), and native speak ers of Mandarin Chinese (12). All English listeners were graduate st udents enrolled at the University of Florida. All Mandarin and

PAGE 22

Spanish listeners were intermediate to high-interm ediate learners of English who were enrolled in graduate school at the Univer sity of Florida. The proficie ncy standing of listeners were decided by the same method descri bed for speaking participants. In the first part of the perception task, each listener was seated in a sound-attenuated booth while 108 sentences, 54 from each each of two speakers, were presented through headphones with five-second interv als between each sentence. After each sentence, listeners were asked to first write the key word in English, as best you can. If you dont understand the word, just write down what part you could unders tand and then to rate how easy each word was to understand on a seven-point scale, with 1 being difficult to und erstand and 7 being easy to understand. Each participant was given a form with blanks for the key word and a blank likert scale. In the second part of the perception task, each listener was seated in a sound-attenuated booth and was presented with the same 108 sentences in a different order, this time with only a two-second interval between each se ntence. After each sentence, pa rticipants were asked only to write the key word in English, as best you can. If you dont understand the word, just write down what part you could understand. This section did not contai n a rating task. Since it could be assumed that less proficie nt speakers would be more likely to produce simplified forms, the two speakers used for the per ception task were the two participants with the lowest test scores and the least time spent in the United States. After the perception task was completed, multiple comparisons were completed for both perception data and rating data. Expectations When developing this study, I expected that the differences between syllable structures and phoneme inventories would provide difficultie s for intelligibility between language groups.

PAGE 23

Mandarin only allows the least-marked syllable structures of CV or CVC(sonorant); however, this study examines the most marked syllabl e structure, CVC(stop) and CVCC(C)(C). Therefore, I expected that Mandarin learners would have difficulty producing final consonants and consonant clusters and might try to correct this problem with various simplification strategies. Since the proficiency of speakers in this study is high-intermediate, it might be expected (based on the notion of recoverability ) that participants will epenthesize or devoice more than delete, in order to retain the original form to the greatest extent possible. It was also expected that Chinese speaking and listening participants may use aspiration as a cue to voicing. According to Wang (1995), in the Mandarin language, aspiration in obstruents is phonemic; so, aspirated stops are differentiated from unasperated stops. Therefore, it is very common among Manda rin speakers to differentiate between English voiced and voiceless stops by means of aspiration. Consequently/b/ is often perceived and produced as unaspirated /p/, and /p/ as /ph/ regardless of its context (39). Therefore, I expect to find that participants may produce /p/ with burst of aspiration a nd /b/ with none. And, I would also expect that Mandarin listeners would be aware of this vo icing cue and could theref ore have a higher rate of understanding than Spanish or English listeners. Another expectation is that Spanish listeners would have a more difficult time determining wo rd-final [p] than would native speakers of English, as the voicing distinction does not exist word-finally in Spanish. Whether or not these cues are interpreted correctly by native speakers of English or by native speakers of Spanish is not documented in the literature. Finally, I expect to find that Mandarin liste ners rate the speakers as more easy to understand than Spanish or English listeners do. It is possible that Spanish listeners will rate the speech as more difficult to understand relati ve to English listeners based on the fact that

PAGE 24

English allows more complex structures and a word-final voicing distinction. Additionally, English speakers will be more familiar with the key words and could therefore have an easier time understanding the speech. Table 1-1. Phoneme inventory: Mandarin, English, and Spanish. Bilabial Alveolar Velar Mandarin p ph t th t th Spanish p b t d t d English p b t d t d

PAGE 25

CHAPTER 2 RESULTS Introduction Overall production results show that interm ediate Chinese speakers of English have learned to produce some word-final clusters but still have difficulty producing clusters containing voiced obstruents in particular. Additionally, these speakers are understood to varying degrees by people from different langu age backgrounds, for Mandarin listeners were able to understand key words better than Englis h or Spanish listeners. This could be due to differing voicing cues between Mandarin Chinese a nd English. Additionally, findings show that other factors such as the key words position in a sentence and the place of articulation of the word-final obstruent play a significant role in production and in understanding. Production Results Cluster production data show that Chinese sp eakers have difficulty producing word-final consonant clusters containing voi ced obstruents and that these cl usters are usually simplified by either obstruent devoicing or consonant deleti on. Word-final consonant clusters containing voiceless obstruents undergo less ch ange, however (Table 2-1). When simplification occurs, there is a pref erence for deleting one consonant of threeconsonant cluster and for devoicing or deleting on e consonant of a two-consonant cluster, and when deletion takes place, learne rs prefer to delete the most sonorous member of a cluster. When devoicing occurs, it is the obstruent th at is devoiced, often along with the preceding consonant. What follows are some examples of simplification strategies found in the data. As seen in (1), the three-consonant cluste rs are usually simplified by deletion, and the deletion usually takes the most sonorous member of the 3-consonant cluster. When two consonants are deleted, the fina l consonant is retained.

PAGE 26

(1) /markt/ [makt] /sk lpt/ [sk pt] [sk t] In example (2), the CC clusters often undergo de voicing and deletion. For example, the cluster /vd/ is sometimes devoiced and produced [ft] wh ile sometimes the more sonorant [v] is deleted while the obstruent [d] is retained. In ot her CC clusters, the obstruent is devoiced. (2) /muvd/ [muft] [mud] /burd/ [burth] /curb/ [crph] Voiceless obstruent clusters are produced with less difficulty; however, some simplification strategies do take place. Exam ple (3) shows that /bank/ undergoes deletion and weakening while /p rk/ just undergoes deletion. (3) /benk/ [bk] [brs] /park/ [pak] Sentence position is an important factor in production of wo rd-final obstruent clusters, for clusters in phrase-medial position, before th e word again, are simplified less often than those in phrase-final position. In particular, wh en clusters with voiced obstruents are positioned at the end of a phrase, far more mistakes occur than when clusters are phrase-medial. The effect of sentence position is negligible when the clusters contain voiceless obstruents. Numerical results are shown in Table 2-2 while overall percentages are give n in Figure 2-1. The difference in production patterns between clusters containing voiced and voiceless obstruents is striking. Production results show that as Chinese sp eakers move from an unmarked,

PAGE 27

native syllable structure to the non-native structure of Eng lish, they produce many complex clusters correctly while still simplifying highly marked structur es such as voiced obstruents. Perception Results The perception study examined listener understa nding of key words with two types of codas: complex obstruent codas and simple obstruent codas. Since previous research shows that simple obstruent codas are frequently devoiced a nd deleted by Chinese learners of English, and since the results of the first experiment show that devoici ng and deletion are also common simplification strategies for complex coda structur es, it was expected that listeners would have difficulty identifying key words with these coda type s. It was also expected that there might be differences between listener groups with regards to understanding. In general, perception results shows that al l three listener groups of Mandarin, English, and Spanish understand Chinese production of word-f inal voiceless stops a nd clusters more than word-final voiced stops and clus ters; additionally, there is a significant difference in perception between the groups, as Mandarin listeners be st understand the Mandarin-accented speech, followed by English listeners and then Spanish li steners. Both voiced and voiceless stops in sentence-medial position are identified as voiced more often than those in sentence-final position. Additionally, place of articulation and vow el duration seem to affect perceptions of word-final stops and word-final clusters. Ther e was no significant effect of speaker (p=.3); therefore, data was analyzed using both speakers throughout. Results show that regardless of speaker, the most significant predictor of listener understanding is the intended voicing of the key wo rd. For all listener groups, key words ending in final voiceless obstruents were perceived corr ectly more often than those ending in final voiced obstruents (p<.01); additionally, there is an effect of place of articul ation, for the alveolar

PAGE 28

obstruent [d] was understood more often than the bilabial obstruent [b] (p =.001) and the velar obstruent [g] is understood more often than al veolar (p = .03) Wh en misunderstandings occurred, the segment was most of ten heard as a voiceless stop in the same place of articulation; however, in some cases, the listeners perceived a li quid or glide in the same place of articulation. There was no significant effect of place of articulation for voiceless obstruents; however, descriptive data shows that bilabial [p] was understood correctly more often than alveolar [t] or velar [k]. Results are given in Table 2-3 and Figure 2. There is also a simple main effect of sent ence position (p<01). Fo r all groups, words in phrase-final position are understood more often than those follo wed by a vowel-initial word (Table 2-4 and Figure 2-3). This effect is most pronounced for voiceless obstruents, as phrasefinal position allows for greater aspiration, which could be a cue to voicing for Chinese learners of English. This will be disc ussed later in the paper. Group Effects Another significant predicto r of understanding is group. Results show that Mandarin listeners are able to understand all key words be tter than English list eners (p=.005) and better than Spanish listeners (p<.01). Additionally, E nglish listeners perceive key words better than Spanish listeners (p=.0002). There is also a signi ficant effect of syllable type, for simple C codas were understood more than complex clusters overall (p<.01). Surprisingly, although the Mandarin language does not allow word-final obstru ents or complex clusters while the English language allows both, Mandarin listeners underst ood all cluster types more accurately than English and Spanish listeners (Table 2-5 and Figure 2-4). Even though English and Spanish listeners dont differentiate well between voiced and voiceless obstruents produced by Chinese speakers, it is apparent that the Chinese listeners are

PAGE 29

more aware of the distinction. As Wang (1995) predicted, Chinese speakers seem to use aspiration to cue listeners to the quality of th e final stop consonant, and Chinese listeners could have more accurate perception scores because they are aware of this cue. The word-final [p] segments that were released with a strong aspi rated burst (average in tensity of 47.9 db) were always judged as [p] by native English listeners. However, those produced with a less intense burst (avg. 44.4 db) are judged incorrectly by nati ve listeners. Though Chinese listeners do not always judge these words correctly, the perception sc ores suggest that they may be aware of this cue to voicing. While aspiration appears to be an important cue for Chinese speakers and listeners, an important voicing cue for native speakers of E nglish is vowel length, for when the vowel length is greater than .25s, the following stop consonant is often interpre ted as voiced, even when the intention of the speaker was to produce a voiceless consonant. When the vowel length differed by .081s, and both words were interpreted correctly by all listener groups. However, when the vowel length is similar, native speaker s interpreted /mob/ as [mop]. This finding is corroborated by the findings of Wardrip-Fluin (1982) and Flege et al. (1992) who report that vowel length is an impor tant cue for native speakers of English when determining the voicing quality of a word-final ob struent. Flege et. al.s (1992) study reports that native speakers produce vowels before a word -final voiced obstruent with a longer vowel duration than those produced before a word-f inal voiceless obstruent. Additionally, when judging the /t/-/d/ contrast in sp eakers of foreign languages, native speakers used vowel length as a significant cue. According to Flege et al., l isteners may tailor their perceptual processing of stops to the range of cues that are readily available (p. 141). So, in the case of this study, where

PAGE 30

there was limited contextual info rmation and the target words were only one syllable in duration, vowel length was a readily available cue. While we have evidence for voicing cues for Mandarin listeners and English listeners, it is unclear what Spanish listeners use to determin e voicing in foreign-accent ed speech. No clear patterns exist for aspiration or vowel length, as with Mandarin a nd English, so since Spanish listeners performed more poorly than even Englis h listeners, it may be that simple familiarity with English words allowed English speakers to identify key words more accurately. Important findings include the production patte rns of Mandarin learners of English, as there are clear patterns in si mplification strategies depending upon the size and voicing of the obstruent clusters. Additiona lly, perception results show that understanding of different language groups can be influenced by the first language; this may be a result of first-language influence on perceptual cues to voicing.

PAGE 31

Table 2-1. Overall production results Target cluster English Word Percent Correct Actual Production [rp] burp 95 (21/22) [b] (1/22) [rst] burst 86 (19/22) [rd] (1/22) [r t] (1/22) [r ] (1/22) [nk] bank 81 (18/22) [k] (3/22) [rs] (1/22) [mp] lamp 81 (19/22) [m] (2/22) [p] (1/22) [lpt] sculpt 45 (10/22) [pt] (5/22) [t] (2/22) [rpt] (3/22) [pt ] (2/22) [rkt] marked 27 (6/22) [makt] (13/22) [d ] (1/22) [kIt] (2/22) [rk] park 40 (9/22) [k] (12/22) [rb] curb 45 (10/22) [rp] (12/22) [rd] bird 45 (10/22) [rt] (10/22) [r] (1/22) [rd ] (1/22) [vd] moved 18 (4/22) [vt]/[ft] (9/22) [d] (4/22) [v d] (3/22) [vd ] (3/22) [ld] gold 4 (2/22) [d] (8/22) [lt] (4/22) [ld ] (4/22) [t] (4/22) [lb] bulb 0 (0/22) [b] (7/22) [p] (7/22) [mp] (5/22) [lp] (1/22) [rb] (1/22) [rp] (1/22)

PAGE 32

Table 2-2. Phrase position results Target Phrase Medial Percent Co rrect Phrase Final Phrase Final [rp] burp again 100 (11/11) burp 90 (10/11) [rst] burst again 100 ( 11/11) burst 90 (10/11) [rkt] marked again 27 (3/11) marked 27 (3/11) [nk] bank again 81 (9/11) bank 81 (9/11) [mp] lamp again 72 (8/11) lamp 90 (10/11) [lpt] sculpt again 45 (5/11) sculpt 45 (5/11) [rk] park again 36 (4/11) park 45 (5/11) [rd] bird again 72 (8/11) bird 18 (2/11) [rb] curb again 63 (7/11) curb 11 (1/11) [vd] moved again 36 (4/11) moved 0 (0/11) [ld] gold again 18 (2/11) gold 0 (0/11) [lb] bulb again 0 (0/11) bulb 0 (0/11) Table 2-3. Effect of place of articulati on on perception scores : percent correct Bilabial Alveolar Velar Voiced 14% 28% 36% Voiceless 85% 74% 82% Table 2-4. Effect of sentence position on perception scores: percent correct Voiceless Voiced Overall 74% 28% Sentence Medial 67% 27% Sentence Final 79% 29% Table 2-5. Perception of codas by gro up and coda type: percent correct All Key Words Clusters Simple C Mandarin 54% 52% 55% English 48% 45% 50% Spanish 39% 33% 44%

PAGE 33

Figure 2-1. Production results percent correct Figure 2-2. Effect of place of articulati on on perception scores : percent correct

PAGE 34

Figure 2-3. Effect of sent ence position on perception sc ores: percent correct Figure 2-4. Perception of codas by group and coda type: percent correct

PAGE 35

CHAPTER 3 ANALYSIS AND CONCLUSIONS Introduction In this chapter, the production data is analyzed within the framework of Optimality Theory (OT) in order to explain production as a dynamic interaction between the first language of Mandarin and the second language of English; additionally, a model of language learning by Chinese learners of English is provided within the gradual learning al gorithm. Finally, major conclusions from production and perception data are discussed, and limitations of the study as well as ideas for future research are given. Analysis Optimality Theory Production data shows that Chinese learners of English simplify complex codas in various ways, and Optimality Theory (OT) provides a framework within which we may analyze the data by discussing how the interaction betw een language transfer a nd universal markedness predict production strategies. OT can be used to analyze secondlanguage data as multiple sets of ranked constraints, showing how speakers move from total faithfulness to the native language ranking to accuracy in the second language. Theref ore, the OT constraints needed to account for the Chinese production data will address both fa ithfulness to the native language of Mandarin and universal markedness (Table 3-1). According to Broselow et al. (2004), Manda rin Chinese has a constraint ranking which favors markedness constraints and therefore produces unmarked st ructures. In the following ranking, markedness constraints will keep codas from surfacing at all: NoCoda, NoVoicedObsCoda, NoObsCoda, *ComplexCoda >> Faithfulness The ranking of NoCoda above faithfulness is all that is needed for th e Mandarin language to avoid codas of any kind, including obstruent codas, complex codas, or voiced obstruent codas, as

PAGE 36

the other constraints describe. English, on th e other hand, ranks faithfulness above NoCoda, therefore producing complex syllable structures and marked segments: Faithfulness >> NoCoda, NoVoicedObsCoda, NoObsCoda, *ComplexCoda As Chinese learners of Eng lish become more proficient, they must learn to demote markedness constraints below faithfulness constrai nts. According to Broselow et al. (1998), the reranking process can be attribut ed to the interaction between language transfer and universal markedness, for learners begin with the ranking of the native language and then, under pressure from interlanguage data, begi n to construct an interlanguage grammar in which the rankings of constraints may diffe r from the native-language ranking. In this case, markedness effects that are not visible in either the native or the target language may become visible in the interlanguage data (279). Therefore, it was expected in this experiment a nd also found that intermediate Chinese learners of English have learned to demote certain marke dness constraints, but not all of them, and that intermediate learners ther efore produce some complex codas while also relying upon simplification strategies to avoi d highly marked structures. In general, intermediate learners of English simplify complex CC clusters in one of three ways: consonant deletion, voiced obstruent devo icing, or both, and they simplify complex CCC clusters by consonant deletion. Optimality Theory should be able to account for all of these changes. In this section, we will examine each of these simplif ication strategies separately, giving examples and appropria te constraint rankings. One common strategy used for correcting voiced obstruent clus ters is obstruent devoicing. For example, /burd/ [burt] 80% of the time and /kurb/ [kurp] 100% of the time that changes occurred. In order to account for this process within OT, we need the following ranking: NoVoicedObsCoda >> MaxI-O >> IdentI-O[ObsVce]

PAGE 37

Additionally, since complex codas remain, *ComplexCoda and NoCoda must be ranked low (see figure 3-1): NoVoicedObsCoda, MaxI-O >> IdentI-O[ObsVce] >> *ComplexCoda, NoCoda Another common strategy is consonant deletion. For example, /park/ [pak] 84% of the time and /bulb/ [bub] or [bup] 77% of the time that ch anges occur. The following ranking is needed to allow for consonant deletion: *ComplexCoda >> MaxIO When deletion occurs, it is usually the non-obstruent that is deleted. In order to account for this phenomenon, the OT constraint must desc ribe which consonant is deleted. *Complex Coda, MaxC[-son] >> MaxI-O Additionally, epenthesis is a very uncommon strategy and th erefore DepI-O should be ranked higher than the preferred strategy of MaxI-O violations (figure 3.2). *ComplexCoda, MaxC[-son], DepI-O >> MaxI-O This ranking isnt complete, however, as it should also be able to account for three-consonant coda simplification such as in /rkt/ [kt]. In this example, we need to introduce the constraint *CCC, and it should be ranked above *ComplexCoda to keep three-consonant clusters from emerging on the surface. Additionally, *ComplexC oda must be demoted, as a complex coda is left on the surface (figure 3-3). In some cases, both obstruent devoicing and consonant deletion occur. For example, /bulb/ [bap] and /gold/ [got]. In this case, the constr aints must be ranked to allow both deletion and devoicing to occur (figure 3-4). NoVoicedObsCoda, MaxC[-son], DepI-O >> *ComplexCoda >> MaxI-O, IdentIO[ObsVce]

PAGE 38

With this simple ranking, we can capture the general trends of the data; however, there is great variability in the data that is not accounted for here. Ideally, the same ranking of constraints should account for any language input, so all simplif ication strategies should be possible outputs of the theory. However, because of variability in speaker strategies, it is necessary to talk about general patterns in production a nd in learning. That is the purpose of the Gradual Learning Algorithm. Modeling Learning in the Gradual Learning Algorithm Boersmas Gradual Learning Algorithm (G LA) in stochastic OT (Boersma, 1999; Boersma & Hayes, 2001) is a program in whic h learning can be represented in Optimality Theory. The difference between traditional OT and stochastic OT is the possibility of overlapping constraints. While OT implies stri ctly ranked constraints that produce the same output on every try, stochastic OT ranks constraints along a conti nuum (see figure 3-5), therefore allowing for various outputs on diffe rent trials (see figure 3-6). The GLA within Stochastic OT is a valuable model within which to discuss learning, for given an initial state grammar and target-language data, the program can predict possible stages that learners may go through during the language acquisition process. The program does not assume that learners move from one stage to th e next abruptly; rather, as they move between stages, variation occurs. When programming an OT grammar, the user gives every constraint a ranking value on a scale of 0 100, a nd noise, or possible variation, is added to each trial. As the algorithm repres ents error-driven learning, small changes in the ranking va lue occur as the given grammar compares its constr aint ranking with input-o utput pairs from the target grammar. The Gradual Learning Algorithm was first used within Stochastic OT to model firstlanguage learning; however, others have applied the program to second-language learning (Xu, 2003 in Broselow et. al., 2004; Wiltshire, 2006) and it is also the purpose of this section to apply

PAGE 39

the algorithm to Chinese learners of English. Since all learners in this experiment began studying English after the age of nine, it is assumed that th e first language had been acquired before English learning began. Therefore, the ini tial state grammar in this analysis is Mandarin Chinese. As previously discussed, the Ma ndarin language prefers unmarked structures; therefore, the Mandarin OT gram mar is set to favor markedness constraints over faithfulness constraints. All constraints are considered to be on a ranking scale, a nd following the example of Xu (2003) (in Broselow et. al., 2004), the marke dness constraints were given a ranking value of 100 while the faithfulness constraints received a ranking value of 88. The twelve-point difference in the ranking values insures that the constraints to do not overlap in the grammar. The constraints and rankings are given in Table 6. The standard deviat ion was set at the standard 2.0 (Boersma & Hayes 1999). Given the initial state grammar, Praats output distribution predicts that the program will give open syllables 100% of the time, until constraints shift in OT ranking. All other outputs shoul d never occur. According to Broselows description of the GLA, the rate at which a markedness constraint is demoted is a functi on of the frequency with which the constraint is violated by input structuresthe more general constraint will be demoted more quickly than the more specific (and therefore less frequently violated) constraint (137). Theref ore, I considered the frequency of complex codas and planned th e input-output pairs a ccordingly. The desired output of the GLA for the input /bulb/ is the fa ithful form [bulb]. Therefore, I gave the desired output of [lb] the highest score in the input-out put pairs. Other outputs are ra nked according to frequency, for according to the BYU Corpus of American Englis h, English words ending in liquids [r] and[l] are more common than those ending in voiceless obstr uents [p] [t] and [k]. Therefore, since we might expect to see users simplifying [lb] clusters to [l] before [p], [l ] is ranked as the second

PAGE 40

most common in the GLA, and [p] as the third most common. [b] was consid ered to be the least common output given the marked nature of the segm ent in word-final position (see figure 3-7). When the GLA is set to learn at a rate of 300 trials, the grammar accurately describes the learners production data collected for this experiment. In order to move from the totally unmarked Mandarin grammar to the totally ma rked English data, the algorithm goes through three stages, thus capturing three important obs ervations about the learning process. The grammar tells us that sonorant codas are acquired be fore obstruent codas, that voiceless obstruent codas are acquired before voiced ones, and that complex codas with voiced obstruents are acquired last. I will consider each of these in turn. The first observation is that Sonorant Codas ar e acquired before obstr uent codas. In the first change, the markedness constraint NoObsCoda and the faithfulness constraint Max-IO are ranked above markedness constraints NoComplexCoda, NoCoda, and NoVoicedObsCoda, and the corresponding output is [bul]. This predic tion corresponds with research documenting that learners tend to acquire non-co mplex, sonorous syllable codas be fore complex, obstruent codas (Yoo, 2004; Wiltshire, 2006); however, it does not di rectly correspond to the data collected, as learners were much more likely to delete the so norant consonant in the consonant cluster (Figure 3-8). I suspect the reason the algorithm produced this output was due to the frequency ratings liquids and obstruents. Howeve r, the algorithm doesnt take s onority sequencing of consonant clusters into consideration; therefore, the predicted output was not found in the data. The second observation is that voiceless obs truent codas are ac quired before voiced obstruent codas. After another 300 tries, NoVc dObsCoda and Max-IO were moved in front of Ident-IO[ObsVce] therefore producing optimal outputs with voicel ess obstruent clusters (see

PAGE 41

Table 8). Again, this prediction by the GLA co rresponds to numerous cr oss-linguistic studies citing that voiceless obstruents are less marked than voiced ones, and more specifically, this prediction corresponds to studies showing that Chinese learners of English tend to acquire voiceless obstruent codas first (Tarone, 1980; Eckman, 1981; Anderson, 1983; Wang, 1995; Broselow et. al., 2004). This prediction by the GLA also corresponds with the production data collected in this experiment, for two speakers were recorded producing [bulb] as [bulp] and [bump]. The third observation is that complex codas with voiced obstruents were acquired last. After a final 300 tries, Max-IO became a highly-ranked constraint while IdentObsVce moved lower, therefore allowing [bulb] to surface as th e correct output (Figures 3-9 and 3-10). This makes sense given the constraint rankings and the target data; how ever, what is not certain is whether language learners actually ever reach na tive-like competency or native-like constraint rankings. The results do correspond with the output distri butions given by Praat for the initial state grammar that has undergone learning. For the in put /bulb/, the output [bul] was predicted to occur 26% of the time; [bulb] was predicted to occur 25% of the time; [bub] was expected 22% of the time; [bup] was predicted 11% of the time; [bulp] was pred icted 9% of the time; and [bu] was predicted to occur 0% of the time. To test the GLAs prediction for comple x CCC clusters, the algorithm was once again programmed for the input /markt/. Once again, th e initial state grammar is that of Mandarin, with all markedness constraints being ranked above all faithfulness constraints. The OT tableau is given in Table 10; the optimal output is th e unmarked syllable form, [ma] (Figure 3-11).

PAGE 42

The desired output of the GLA for the input /markt/ is the faithful form [markt]. Therefore, the pair distribution ranked the desired output [markt] the highest. Simplified codas were ranked according to the BYU Corpus of S poken American English, with [rk] being the most common coda, [r] and [kt] the second most common, and [k] the least common coda in the spoken English database. When the GLA is set to learn at a rate of 300 trials, the grammar accurately describes the learners production data collected for this experiment. In moving from the initial state of Mandarin to the target state of English, the algorithm goes through only two stages, one simplified form and the actual targ et. After 300 trials, the GLA still predicts [ma] as the output; however, after 600 and 900 trials, the GLA predicts [makt] as the outpu t. After 1200 trials, *CCC is demoted, and the GLA predicts the correct output of [markt] (F igures 3-12 and 3-13). The prediction of the GLA is accurate for la nguage data collected, for /markt/ was only produced as such 27% of the time, and100% of a ll simplified forms were produced as [makt]. Results also correspond to output distributions gi ven by the Praat program: that [park] will surface 47% of the time after learning; that [pak ] will surface 26% of the time; that [par] will surface 24% of the time; and that [pa] will surface 0% of the time. Conclusions The production data expands the observation that has been made about word-final obstruent production by Chinese learners of English by showing that similar strategies also apply to production of word-final cluste rs. These strategies are subject to first-language constraints and constraints on universal markedness, and th ey are predictable based on the first language (Mandarin) and the seco nd language (English). The Gradual Learning Algorithm accurately pred icts the results of the production task, namely that consonant clusters containing voiced obstruents are the most difficult for Chinese

PAGE 43

learners of English to acquire and that simplification strategies such as consonant deletion and devoicing frequently occur. Additionally, the ge neral observations about language learning are important, as they reflect the role of universal markedness in language acquisition. What is unrealistic about the GLA, however, it its somewhat static representation of the learning. In actual language acquisition, people dont learn at the same rate nor do they always progress through the same stages. This could be because of different frequencies of input. For example, language learners immersed in a second-language environment have high frequencies of input and may learn more quickly than those who ha ve almost no input. Additionally, most people never achieve the native-like produc tion; rather, if they are able to communicate effectively, many language learners remain in an intermediate stage. The perception data shows us that speakers from different first-language backgrounds interpret the same input in different ways. In this study, Chinese listeners performed significantly better at un derstanding Mandarin-ac cented speech, and Englis h listeners performed better at understanding the Mandarin-accented speech than did Spanish listeners. Since the input was exactly the same, it seems that these three groups of listeners were paying attention to different cues in the data. Based on the predic tions of Wang (1995) related to aspiration and the findings of Flege (1992) about native-speaker attunement to vow el length, the results of the experiment seem to suggest that Mandarin listene rs are more aware of aspiration as a cue to voicing while native speakers of English are more attuned to vowel length as the most important cue; however, it is still unclear if Spanish li steners were aware of any specific cues that influenced their performance on the perception task. This is an important finding, as the use differing cues will certainly lead to misunderstandings between language groups. Additionally, the finding speaks to the

PAGE 44

relationship between perception a nd production. Previous literatu re has proven that perception and production are closely related (Flege & E ffting, 1987; Flege, 1993; Bradlow et. al., 1996), while it is still debated whether perception precedes production or vice versa. In this case, it is not clear whether Mandarin learne rs of English may be perceivi ng aspiration as a cue to voicing because that is what they are trained to produce. Or, whether they may be using aspiration in their production because they perceive this distinc tion in the speech of other language learners. Studies have come to various conclusions about the relationship between production and perception. One study by Flege & Effting (1987) suggests that production precedes perception. This study examined Dutch learners of English production of /t/-/d/ contrast in English, finding that although Dutch learners were able to produc e the sounds with relative accuracy, they were not able to perceive the difference between the sounds in a perception study. However, multiple other studies suggest that per ception of non-native contrasts bo th precedes and influences production. For example, a study by Bradlow et. al. (1996) proved that Japanese learners of English who were trained to perceive the /r/-/l/ contrast also improved in their production of the sounds, and another study by Flege (1993) showed th at perception of English words ending in /t/ or /d/ by Chinese learners of English was correla ted with the amount of foreign accent they were judged to have by native speakers of English. Therefore, Flege c oncludes that pe rception of the /t/-/d/ contrast is related to native-like production of these sounds. The relationship between perception and production is important for our discussion of Chinese learners of English, for if these learners are indeed using non-native cues to voicing in their production and perception, mi sunderstandings could occur wh en speaking to other language groups. A better understanding of how cues to vo icing are used in perception and production could impact the way we think about second-language learning and teaching. For example,

PAGE 45

perceptual training methods could be devised that train listeners to use more native-like cues, and if these training methods are able to improve the perception of word -final obstruents and obstruent clusters, it may also positively affect the production of these second-language forms. The limitations of this study include a small sample size and a limited number of cluster types. Additionally, studying production within such a limited speech environment such as a frame sentence does not result in completely natural production patterns; therefore, findings in this study cannot be generalized to conversational situations. Fi nally, all transcriptions were completed by one researcher, with no outside ju dgments, and the profic iency level of speakers and listeners could be more rigi dly controlled in order to exam ine the role of proficiency on perception and production. Future research could focus on training methods and perceptual acquisition. In particular, I would like to focus future research on psycho linguistic training methods, for previous research has shown that perceptual training can cause improvements in behavioral accuracy and production as well as preattentive reaction times For example, Tremblay et. al. (1997) examined the effects of training on segment percep tion. Native English participants were trained to hear and identify a prevoiced la bial stop, and a prevoiced alveolar sound was used as a transfer stimuli. After only one training session, the pr eattentive response increased significantly in duration and area. Bradlow et. al.s (1997) study also shows that perceptual training can also increase production accuracy. After training Japa nese learners of English to perceive the difference between /r/ and /l/, it was found that th eir ability to produce the segments accurately also increased. This study only begins to touch upon important issues for Chinese learners of English, namely how the first language and universal prin ciples influence the ability to learn a second

PAGE 46

language. While these theoretical findings are important, I believe they have implications for further research on training methodol ogy and perceptual cues, which I leave to future research. Table 2-1. Markedness a nd Faithfulness Constraints Markedness NoCoda Syllables are open (Kager, 1999) *ComplexCoda Codas are simple (Kager, 1999) NoObsCoda No obstruents in the coda (Broselow et al., 2004) NoVoicedObsCoda No voiced obstruents in the coda (Broselow et al., 2004) *CCC No CCC clusters (Kager, 1999) Faithfulness MaxI-O Input segments must ha ve output correspondents (Kager, 1999) DepI-O Output segments must ha ve input correspondents (Kager, 1999) IdentI-O No featural changes (Kager, 1999) IdentI-O[ObsVce] Correspondent obstruents are identical in their specification for voice (Kager, 1999) MaxC [ -son ] Inputs segments with the feature [-son] must have output correspondents /rd/ NoVoicedObsCoda MaxIO IdentIO[ObsVce] *ComplexCoda NoCoda rt d *! rd *! r *! Figure 3-1. Obstruent devoicing as the preferred strategy /rk/ *ComplexCoda MaxC[-son] DepI-O MaxI-O rk *! k r *! rik *! Figure 3-2. Consonant deletion as the preferred strategy

PAGE 47

/rkt/ *CCC MaxC[son] DepI-O *ComplexCoda MaxI-O rk *! kt r *! ** kit *! rkt *! Figure 3-3. Three-consonant coda simplification /lb/ NoVcdObs Cda MaxC[son] DepIO *Comple xCoda MaxI-O IdentIO[ObsVce ] p b *! lb *! lp *! lepe *! l *! Figure 3-4. Deletion and devoicing Figure 3-5. Example of ranked constraints in stochastic OT (Boersma, 2003)

PAGE 48

Figure 3-6. Possible output of ranking in stochastic OT Figure 3-7. Ranking values and tableau fo r the initial state Mandarin Chinese

PAGE 49

Figure 3-8. Ranking values and tableau at 300 trials Figure 3-9. Ranking value a nd tableau at 600 trials

PAGE 50

Figure 3-10. Ranking values and tableau for output [bulb]. Figure 3-11. Initial state grammar: input /markt/ Figure 3-12. GLA after 600 trials

PAGE 51

Figure 3-13. GLA predica tion after 1200 trials

PAGE 52

LIST OF REFERENCES Anderson, J. (1983). The difficulties of English syllable structure for Chinese ESL learners. Language Learning and Communication, 2, 53-61. Bedore, L. (1999). The acquisition of Spanis h. In O. Taylor & L. Leonard (Ed.), Language acquisition across North America: Cross-cu ltural and cross-linguistic perspectives (pp. 157-207). San Diego, CA: Singular. Best, C.T. (1995). A direct realist perspective on cross-language speech perception. In: W. Strange (Ed.), Speech perception and linguistic experience: issues in cross-language speech research (pp. 171-204). Timonium, MD: York Press. Birdsong, D. (1999). Whys and Why Nots of the Critical Period Hypothesis for Second Language Acquisition. In D. Birdsong (Ed.), Second Language Acquisition and the Critical Period Hypothesi (pp. 1-22). Mahway, NJ: Lawr ence Erlbaum Associates. Boersma, P. (1999). Optimality-Theoret ic Learning in the PRAAT program. Proceedings of the Institute of Phonetic Sciences, 23, 17-35. Boersma, P. & Hayes, B. (2001). Empirical tests of the Graduate Learning Algorithm. Linguistic Inquiry, 32, 45-86. Bradlow, A., Pisoni, D., Akahama-Yamada, R., & Tohkura, Y. (1997). Training Japanese listeners to identify /r/ and /l/: IV. Some effects on perceptual learning and speech production. Acoustical Society of America 101, 2299-2310. Broselow, E. (1984). An investigation of transfer in second language acquisition. International Review of Applied Linguistics, 22 253-269. Broselow, E., Chen, S., & Wang, C. (1998). The em ergence of the unmarked in second language phonology. Studies in Second Language Acquisition, 20, 261-280. Broselow, E., & Xu, Z. (2004). Differential difficulty in the acquisition of second language phonology. International Journal of English Stud ies: Advances in Optimality Theory, 4(2), 135-163. Cheng, L. (1991). Assessing Asian language performance: Guidelines for evaluating limitedEnglish proficient students, second edition Oceanside, CA: Academic Communication Associates. Eckman, F.R. (1977). Markedness and the contrastive analysis hypothesis. Language Learning, 27, 315-330. Eckman, F.R. (1981). On predicting phonological difficulty in second language acquisition. Studies in Second Language Acquisition 4, 18-30. Gordon, R. (2005). Ethnologue: Languages of the World, fifteenth edition Dallas, TX: SIL International. Flege, J. (1989). Chinese subjects' percepti on of the word-final English /t/-/d/contrast: Performance before and after training. Journal of the Acoustical Society of America 86 1684-1697. 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. 233-277). Baltimore, MD: York Press. Flege, J. & Davidian, R. (1984). Transfer and developmental pro cesses in adult foreign language speech production. Applied Psycholinguistics 5, 323-347. Flege, J., Munro, M., & Skelton, L. (1992). Producti on of the word-final English /t/-/d/ contrast by native speakers of Englis h, Mandarin and Spanish. Journal of the Acoustical Society of America 92, 128-143.

PAGE 53

Hancin-Bhatt, B. (2000). Optimality in s econd language phonology: codas in Thai ESL. Second Language Research, 16, 201-232. Hurford, J.R. (1991). The evolution of th e critical period for language acquisition. Cognition 40, 159-201. Jongstra, W. (2003). Variable and Stable Cluste rs: Variation in the Re alisation of Consonant Clusters. Canadian Journal of Linguistics, 48, 265. Kager, R. (1999). Optimality Theory Cambridge, MA: Cambridge University Press. Kuhl, P.K. (1991). Human adults and human infants show a per ceptual magnet effect for the prototypes of speech categories, monkeys do not. Perception & Psychophysics, 50 93107. Kuhl, P.K., Williams, K.A., Lacerda, F., Stevens, K.N., & Lindblom, B. (1992). Linguistic experience alters phonetic perception in infants by 6 months of age. Science, 255, 606608. Lenneberg, E.H. (1969). Biological Foundations of Language New York, NY: Wiley. Mayberry, R. (1993). First-language acquisition after childhood differs from second-language acquisition: the case of American Sign Language. Journal of Speech and Hearing Research 36 1258-1270. McLeod, S., Van Doorn, J., & Reed, V. (2001). Cons onant cluster development in two-year-olds: general trends and individual difference. Journal of Speech, Language, and Hearing Reseach 44 1144-1171. Munro, M. J., Derwing, T. M., & Morton, S. L. (2006). The mutual intelligibility of L2 speech. Studies in Second Language Acquisition 28, 111-131. Ohala, Diana (1999). The influence of sonority on childrens cluster reductions. Journal of Communication Disorders, 32, 397. Pater, J. and Barlow, J.A. (2003). Constraint conflict in cluster reduction. Journal of Child Language. 30, 487-526. Pilus, Z. (2003). Second language speech: Produc tion and perception of voicing contrasts in word-final obstruents by Malay speakers of English. Dissertation Abstracts International: The Humanities and Social Sciences 63, 11. Pinker, S. (1994). The language instinct: How the mind creates language New York, NY: Morrow. Prince A. & Smolensky, P. (1993). Optimality Th eory: Constraint Interaction in Generative Grammar. Rutgers University Center for C ognitive Science Tec hnical Report 2. Roach, P. (2000). English Phonetics and Phonology: A Practical Course Cambridge, MA: Cambridge University Press. Tarone, E. (1980). Some influences on the syllable structure of interlanguage phonology. IRAL 18, 139-152. Tremblay, K., Kraus, N., Carrell, T., & McGee, T. (l997). Central auditory system plasticity: Generalization to novel stimuli following listening training. Journal of the Acoustical Society of America 6, 3762-3773. Vanderweide, T. (2005). A Perceptual An alysis of Consonant Cluster Reduction. Proceedings of the 2005 annual conference of the Canadian Linguistic Association. Wang, C. (1995). The acquisition of English wo rd-final obstruents by Chinese speakers. Dissertation Abstracts International the Humanities and Social Sciences 56, 10.

PAGE 54

Wardrip-Fruin, C. (1982). On the status of tem poral cues to phonetic categories: preceding vowel duration as a cue to voici ng in final stop categories. Journal of the Acoustic Society of America 71, 187-195. White, L., & Genesee, F. (1996). How native is nea r-native? The issue of ultimate attainment in adult second language acquisition. Second Language Research 12, 238-265. Wiltshire, C. (2006). Word-final Consonant and Cluster Acquisition in Indian English(es). A Supplement to the Proceedings of the 30th Boston University Conference on Language Development. Yoo, H. (2004). A Longitudinal stu dy of consonant cluster acquisition. Studies in Phonetics, Phonology and Morphology, 10.3. 481-503.

PAGE 55

BIOGRAPHICAL SKETCH Milla Chappell is a graduate student in Lingui stics at the Un iversity of Florida who is interested in second language acquisition, particularly that of phonology and phonetics. Milla graduated in 2006 with an M.A. in English from Auburn University and began her studies in linguistics at the University of Florida in August 2006.