<%BANNER%>

Training Native Speakers of American English to Perceive Thai Tones using High Stimulus Variability

Permanent Link: http://ufdc.ufl.edu/UFE0041182/00001

Material Information

Title: Training Native Speakers of American English to Perceive Thai Tones using High Stimulus Variability
Physical Description: 1 online resource (110 p.)
Language: english
Creator: Laphasradakul, D
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: american, english, high, perceive, stimulus, thai, tone, training, variability
Linguistics -- Dissertations, Academic -- UF
Genre: Linguistics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: 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 TRAINING NATIVE SPEAKERS OF AMERICAN ENGLISH TO PERCEIVE THAI TONES USING HIGH STIMULUS VARIABILITY By Donruethai Laphasradakul December 2010 Chair: Ratree Wayland Major: Linguistics The goal of this research was to investigate whether training nai umlautve American English listeners to perceive Thai tones employing high-variability stimuli produced by multiple talkers in conversational speech would be more effective than training using low-variability, prototypical stimuli produced in citation forms. Due to time constraints, it was not realistic to conduct real time teaching sessions using the high stimulus variability in classrooms; therefore, the experiment was carried out in a laboratory setting. Three groups (two experimental and one control) of native speakers of American English (NE) and one group of native Thai (NT) speakers participated in the experiment, which consisted of four phases: Pre-test, Training, Post-test and Retention test. There were 20 participants in each experimental group. In the Pre-test, all four groups of participants were asked to identify the low and mid tones in Thai. Immediately after the Pre-test, the two NE experimental groups underwent training. During training, participants in one experimental group, the Prototype group, were trained with prototypical stimuli produced in citation forms, while the other group, the High Variability group, was trained with high-variability stimuli produced by multiple speakers in a faster, conversational speech style. Participants in the control group did not receive any training and only participated in the Pre-test and the Post-test, and NT speakers only participated in the Pre-test. The effectiveness of the two training methods was assessed immediately after training and again two weeks later. Besides accuracy rate and reaction time, the ability to make perceptual generalizations when presented with both old and new stimuli by new speakers was also evaluated. As expected, NT speakers outperformed all three groups of NE speakers on the Pre-test, and there was no significant difference among the three non-native groups at that time. In addition, it was found that the training significantly improved tone identification accuracy among participants in both experimental groups, but there was no statistically significant difference in the two groups? amount of improvement. The effectiveness of the training was also retained two weeks after training. In addition, the training effects seemed to generalize to novel stimuli, both immediately after training and two weeks after training. However, a closer examination of the data suggested that the high-variability training paradigm may result in a more robust, longer-term representation of the two Thai tones than the low-variability training method. However, the high-variability effectiveness may not be immediate and takes longer to realize. Pedagogical values and implications of these findings are discussed. Shortcomings of the current study are noted and suggestions for further research are included.
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 D Laphasradakul.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Wayland, Ratree.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041182:00001

Permanent Link: http://ufdc.ufl.edu/UFE0041182/00001

Material Information

Title: Training Native Speakers of American English to Perceive Thai Tones using High Stimulus Variability
Physical Description: 1 online resource (110 p.)
Language: english
Creator: Laphasradakul, D
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: american, english, high, perceive, stimulus, thai, tone, training, variability
Linguistics -- Dissertations, Academic -- UF
Genre: Linguistics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: 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 TRAINING NATIVE SPEAKERS OF AMERICAN ENGLISH TO PERCEIVE THAI TONES USING HIGH STIMULUS VARIABILITY By Donruethai Laphasradakul December 2010 Chair: Ratree Wayland Major: Linguistics The goal of this research was to investigate whether training nai umlautve American English listeners to perceive Thai tones employing high-variability stimuli produced by multiple talkers in conversational speech would be more effective than training using low-variability, prototypical stimuli produced in citation forms. Due to time constraints, it was not realistic to conduct real time teaching sessions using the high stimulus variability in classrooms; therefore, the experiment was carried out in a laboratory setting. Three groups (two experimental and one control) of native speakers of American English (NE) and one group of native Thai (NT) speakers participated in the experiment, which consisted of four phases: Pre-test, Training, Post-test and Retention test. There were 20 participants in each experimental group. In the Pre-test, all four groups of participants were asked to identify the low and mid tones in Thai. Immediately after the Pre-test, the two NE experimental groups underwent training. During training, participants in one experimental group, the Prototype group, were trained with prototypical stimuli produced in citation forms, while the other group, the High Variability group, was trained with high-variability stimuli produced by multiple speakers in a faster, conversational speech style. Participants in the control group did not receive any training and only participated in the Pre-test and the Post-test, and NT speakers only participated in the Pre-test. The effectiveness of the two training methods was assessed immediately after training and again two weeks later. Besides accuracy rate and reaction time, the ability to make perceptual generalizations when presented with both old and new stimuli by new speakers was also evaluated. As expected, NT speakers outperformed all three groups of NE speakers on the Pre-test, and there was no significant difference among the three non-native groups at that time. In addition, it was found that the training significantly improved tone identification accuracy among participants in both experimental groups, but there was no statistically significant difference in the two groups? amount of improvement. The effectiveness of the training was also retained two weeks after training. In addition, the training effects seemed to generalize to novel stimuli, both immediately after training and two weeks after training. However, a closer examination of the data suggested that the high-variability training paradigm may result in a more robust, longer-term representation of the two Thai tones than the low-variability training method. However, the high-variability effectiveness may not be immediate and takes longer to realize. Pedagogical values and implications of these findings are discussed. Shortcomings of the current study are noted and suggestions for further research are included.
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 D Laphasradakul.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Wayland, Ratree.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041182:00001


This item has the following downloads:


Full Text

PAGE 1

1 TRAINING NATIVE SPEAKERS OF AMERICAN ENGLISH TO PERCEIVE THAI TONES USING HIGH STIMULUS VARIABILITY By DONRUETHAI LAPHASRADAKUL A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULF ILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010

PAGE 2

2 2010 Donruethai Laphasradakul

PAGE 3

3 To my family, Khaodeedech and Laphasradakul

PAGE 4

4 ACKNOWLEDGMENTS I am deeply indebted to my advisor, Dr. Ratree Wayla nd, who has always been constantly supportive and encouraging, without her this study would not have been finished. My gratefulness also goes to the committee members, Dr. Caroline Wiltshire who always kindly helped me, ever since the days when I applied f or the program, in 2004. Dr. Andrea Pham, Dr. Edit Kaan, and Dr. Jane Townsend have given me invaluable suggestions. I would like to express my gratitude to the University of the Thai Chamber of Commerce under whose scholarship I studied at the Department of Linguistics, Universi ty of Florida, for five years. I would also like to express my appreciation to all of my friends. I always received help and support from them through the five year pursuit of my study. My deepest gratitude goes to my family, Khaode edech and Laphasradakul.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 2 BACKGROUND ................................ ................................ ................................ ...... 19 Thai Tone System ................................ ................................ ................................ ... 19 Differences between Thai and English Prosodic Systems ................................ ...... 23 Previous Studies on Thai Tone Acquisition ................................ ............................. 27 Effects of Experience in Speech Perception and Production Learning ................... 31 The Speech Learning Model (SLM) ................................ ................................ .. 31 The Perceptual Assimilation Model (PAM) ................................ ....................... 32 The Native Language Magnet Effects (NLM) ................................ ................... 33 Exemplar Based Models and Prototype Based Models ................................ .......... 34 Training Studies (High Variability Training Paradigm) ................................ ............ 36 Relationship between Speech Perception and Speech Production ........................ 38 3 METHODOLOGY ................................ ................................ ................................ ... 45 Particip ants ................................ ................................ ................................ ............. 45 Stimuli ................................ ................................ ................................ ..................... 46 Stimuli for the Pre Test ................................ ................................ ..................... 47 Training Stimuli for the Prototype Training Group ................................ ............ 49 Training Stimuli for the High Variability Training Group ................................ .... 49 Stimuli for the Post Test ................................ ................................ ................... 49 Stimuli for the Generalization Test ................................ ................................ ... 50 Stimuli for Retention Tests ................................ ................................ ............... 50 Task ................................ ................................ ................................ ........................ 51 Procedure ................................ ................................ ................................ ............... 51 Pre Test ................................ ................................ ................................ ............ 51 Training ................................ ................................ ................................ ............ 52 Post Test ................................ ................................ ................................ .......... 52 Retention Test ................................ ................................ ................................ .. 53

PAGE 6

6 4 PERCEPTION ACCURACY: PERCENT CORRECTION IDENTI FICATION .......... 56 Pre Test ................................ ................................ ................................ .................. 57 Effects of Training ................................ ................................ ................................ ... 58 Pre Test to Post Test Comparison ................................ ................................ ......... 58 Pre Test, Post Test and Retention Test Comparison ................................ ............. 59 Effects of Training in the Generalization Test ................................ ......................... 61 Generalization Test 1 and Generalization Test 2 Comparisons .............................. 61 General Summary of Findings ................................ ................................ ................ 66 5 REACTION TIME ................................ ................................ ................................ .... 74 Pre Test ................................ ................................ ................................ .................. 74 Effects of Training ................................ ................................ ................................ ... 75 Pre Test to Post Test Comparison ................................ ................................ ......... 76 Pre Test, Post Test and Retention Test Comparisons ................................ ........... 76 Effects of Training in t he Generalization Test ................................ ......................... 78 Generalization Test 1 and Generalization Test 2 Comparisons .............................. 78 Summary of Findings ................................ ................................ .............................. 80 Summary of Accuracy and Reaction Time Data ................................ ..................... 80 6 DISCUSSION AND CONCLUSION ................................ ................................ ........ 87 Pre Test ................................ ................................ ................................ .................. 87 Pre Test and Post Test ................................ ................................ ........................... 88 Pre Test, Post Test and Retention Test ................................ ................................ .. 89 Generalization Test 1 and Generalization Test 2 ................................ .................... 91 Applications to the Classroom ................................ ................................ ................ 95 Limitations of the Present Stud y ................................ ................................ ............. 95 Future Directions ................................ ................................ ................................ .... 96 APPENDIX: STIMULI ................................ ................................ ................................ .... 98 LIST OF REFERENCE S ................................ ................................ ............................. 104 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 110

PAGE 7

7 LIST OF TABLES Table page 2 1 Thai consonants ................................ ................................ ................................ 41 2 2 Thai vowels ................................ ................................ ................................ ......... 41 2 3 Characteristics of Thai tones ................................ ................................ .............. 41 2 4 Th ai tones ................................ ................................ ................................ ........... 41 2 5 C omparison of phonetics of Thai tones (from 1911 1990) ................................ .. 42 3 1 Minimal pairs of the pre test and the post t est stimuli ................................ ......... 54 3 2 Minimal pairs of the training stimuli ................................ ................................ ..... 54 3 3 Minimal pairs of the generalization (post test) stimuli ................................ ......... 54 3 4 Summary of procedure ................................ ................................ ....................... 55 4 1 Mean percent correct in mid and low tones identification in the Pre test comparing between tw o stimulus types for each group ................................ ...... 68 4 2 Comparison of mean percent correct and SE between Pre test and Post test for low and high variability stimulus type conditions ................................ .......... 68 4 3 Mean percent correct tone identification under low variability stimulus type condition ................................ ................................ ................................ ............. 68 4 4 Mean percent correct tone identification under high variability stimulus type condition ................................ ................................ ................................ ............. 68 4 5 Mean percent correct tone identification for Prototype group in the Generalization test 1 and 2 under low variability stimulus type condition ........... 68 4 6 Mean percent correct tone identification for Prototype group in the Generalization test 1 and 2 under high variability stimulus type condition .......... 69 4 7 Mean percent correct tone identification for High Variability group in the Generalization test 1 and 2 under low variability stimulus type condition ........... 69 4 8 Mean percent correct tone identification for High Variability group in the Generalization test 1 and 2 under high variability stimulus type condition .......... 69 4 9 Means percent identification accu racy for factors that the main effect is significant collapsed over other variables ................................ ........................... 69

PAGE 8

8 4 10 S ummary of significant results of the generalization tests ................................ .. 70 5 1 Reaction time in mid and low tones identification in the pre test comparing between two stimulus type conditions for each group ................................ ........ 82 5 2 Comparison of react ion time and SE between Pre test and Post test for low and high variability stimulus type conditions ................................ ....................... 82 5 3 Reaction time in tone identification comparing between the Pre Post Retention tes t under low variability stimulus type condition ................................ 82 5 4 Reaction time in tone identification comparing between the Pre Post Retention test under high variability stimulus type condition .............................. 82 5 5 Reaction time in tone identification for Prototype group in the Generalization test 1 and 2 under low variability stimulus type condition ................................ ... 8 2 5 6 Reaction time in tone identification for Prototype group in the Generalization test 1 and 2 under high variability stimulus type condition ................................ .. 83 5 7 Reaction time in tone identification for High Variability group in the Generalization test 1 and 2 under low variability stimulus type condition ........... 83 5 8 Reaction time in tone identification for High Variability g roup in the Generalization test 1 and 2 under high variability stimulus type condition .......... 83 5 9 Reaction times means collapsed over other variables ................................ ........ 83

PAGE 9

9 LIST OF FIGURES Figure page 2 1 F0 contours of the five Thai tones ................................ ................................ ...... 43 2 2 Thai mid tone syllable [ma:] produ ced in conversational style by 3 speakers ..... 43 2 3 Thai low tone syllable [ma:] produced in conversational style by 3 speakers ..... 44 2 4 The fundamental frequency traces of the standard Thai low and mid tones. [reprinted with permission from Wayland, R. & Guion, S.G. (2004). Training English and Chinese listeners to perceive Thai tones: A Preliminary Report. Language L earning, 54 (4), (page 694).] ................................ ............................ 44 4 1 Comparison of mean percent correct and SE in the Pre test between two stimulus types for each group ................................ ................................ ............. 71 4 2 Comparison of mean percent correct and SE between the Pre test and the Post test under two stimulus type conditions ................................ ...................... 71 4 3 Mean percent correct tone identification and SE between the Pre Post Retention test under low variability stimulus type condition ................................ 72 4 4 Mean percent correct tone identification and SE between the Pre Post Ret ention test under high variability stimulus type condition .............................. 72 4 5 Mean percent correct tone identification and SE in the Generalization test 1 and 2 under low variability stimulus typ e condition ................................ ............. 73 4 6 Mean percentage correct in tone identification and SE in the Generalization test 1 and 2 under high variability stimulus type condition ................................ .. 73 5 1 Reaction time in mid and low tones identification in the Pre test comparing between the two stimulus type conditions and SE for each group ...................... 84 5 2 Comparison of reaction time and SE between Pre test and Post test under two stimulus type conditions ................................ ................................ ............... 84 5 3 Reaction time in tone identification and SE comparing betwee n the Pre Post Retention test under low variability stimulus type condition ................................ 85 5 4 Reaction time in tone identification and SE comparing between the Pre Post Retention test under high variability stimulus type condition .............................. 85 5 5 Reaction time in tone identification and SE in the Generalization test 1 and 2 under low variability stimulus type condition ................................ ....................... 86

PAGE 10

10 5 6 Reaction time in tone identification and SE in the Generalization test 1 and 2 under high variability stimulus type condition ................................ ..................... 86

PAGE 11

11 Abstract of Dissertation Pr esented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy TRAINING NATIVE SPEA KERS OF AMERICAN ENG LISH TO PERCEIVE THAI TONES USING HIG H STIMULUS VARIABIL ITY By Donruethai Laphasradakul December 2010 Chair: Ratree Wayland Major: Linguistics The goal of this research was to investigate whether training nave American English listeners to perceive Thai tones employing high variability stimuli produced by multiple talkers in conversational speech would be more effective than training using low variability, prototypical stimuli produced in citation forms Due to time constraints, it was not realistic to conduct real time teaching sessions using the high stim ulus variability in classrooms; therefore, the experiment was carried out in a laboratory setting. Three groups (two experimental and one control) of native speakers of American English (NE) and one group of native Thai (NT) speakers participated in the ex periment, which consisted of four phases: Pre test, Training, Post test and Retention test. There were 20 participants in each experimental group. In the Pre test, all four groups of participants were asked to identify the low and mid tones in Thai. Immed iately after the Pre test, the two NE experimental groups underwent training. During training, participants in one experimental group, the Prototype group, were trained with prototypical stimuli produced in citation forms, while the other group, the High V ariability group, was trained with high variability stimuli produced by multiple speakers in a faster,

PAGE 12

12 conversational speech style. Participants in the control group did not receive any training and only participated in the Pre test and the Post test, and NT speakers only participated in the Pre test. The effectiveness of the two training methods was assessed immediately after training and again two weeks later. Besides accuracy rate and reaction time, the ability to make perceptual generalizations when pre sented with both old and new stimuli by new speakers was also evaluated. As expected, NT speakers outperformed all three groups of NE speakers on the Pre test, and there was no significant difference among the three non native groups at that time. In addit ion, it was found that the training significantly improved tone identification accuracy among participants in both experimental groups, but there was effectiveness of the training was also retained two weeks after training. In addition, the training effects seemed to generalize to novel stimuli, both immediately after training and two weeks after training. However, a closer examination of the data suggested that the high v ariability training paradigm may result in a more robust, longer term representation of the two Thai tones than the low variability training method. However, the high variability effectiveness may not be immediate and take s longer to realize. Pedagogical v alues and implications of these findings are discussed. Shortcomings of the current study are noted and suggestions for further research are included.

PAGE 13

13 CHAPTER 1 INTRODUCTION In recent years, there has been a growing interest in teaching and learning Thai as either a second or foreign language because the country is considered a paradise for tourists, an increasing number of foreigners are coming to the country. These outsiders want to learn Thai to be able to com municate with Thai people and to become integrated into Thai society (Poomsan, 1995; Ponmanee, 2002a, 2002b) Nowadays, Thai is taught as a second language in Thailand and as a foreign language outside of the country. For instance, China, Korea, Japan, Vie tnam, Malaysia, Indonesia, India, Australia, New Zealand, Canada, the United States, the United Kingdom, Sweden, and Germany, among others, all have programs for Thai as a foreign language (Department of Information, Ministry of Foreign Affairs, Bangkok, 2 000). These programs play a major role in facilitating nonnative that Thai teachers in these programs implement optimal pedagogical methods that are effective in teac hing the most difficult aspect of Thai which is Thai tones (Ponmanee, 2002a, 2002b; Wittayasakpan et al., 2002; Sittikesorn, 2005; Sathiansukon, 2005; Potibal, 2005; Udompan, 2005; Aroonmanakun, 2006). This research aimed to empirically examine if the comm on teaching method that has students perceive and produce Th ai tone in citation forms such as tonal prototypes is effective in teaching Thai tones. P revious studies ( Abramson, 1979; Luksaneeyanawin, 1998; Arunreung, 1990) have shown that pitch contours of Thai tones vary greatly in conversational speech. Despite this fact, many Thai teachers still employ the traditional method in teaching Thai

PAGE 14

14 to non native speakers. They usually present Thai tones in their prototypical, idealized citation forms and have L2 learners mimic his or her production As a result, the learners become used to this idealized representation of Thai tones and use them in their production and perception Producing and perceiving tones in fast, conversational speech where the idealized p itch contours are not retained, however, may prove difficult for them. T his may be one of the reason s why perception and production of Thai tones remains challenging for non native speakers even after years of trying to learn Thai a s reported in a number o f studies reviewed in the next chapter The traditional method of teaching Thai tones could be said to have been influenced by the Prototype theory and the abstractionist view of lexical representation. The Prototype theory maintains that humans categorize objects by matching them to the contain the most representative features of the category (Kuhl, 1992). According to the abstractionist model, cognitive or mental c ategories such as words are represented in the lexicon in the form of normalized, abstract phonological representations ( McClelland & Elman, 1986; Liberman & Mattingly, 1985 ; Norris, 1994) S peech input is abstracted from exact speech and compared to these prototypes or idealized representations by learners during processing. Consistent with this view, providing learners with only ideal exemplars of the five contrastive Thai tones in citation forms should facilitate their learning of these tones. However, i n recent years, new views on how humans categorize objects including mental objects known as the exemplar based models have emerged in various domains of linguistic research including syntax, phonetics, phonology and language

PAGE 15

15 acquisition (Gahl & Yu, 2006 ). Central to the exemplar based model is the idea that mental representations are comprised of memory traces of specific tokens. In other words, under the exemplar based view, a category is comprised of a collection of ch exemplar may, however, belong to many categories simultaneously. In an exemplar based speech processing system, inputs are recognized and outputs are generated based on analogical evaluation across distinct memory traces of remembered tokens. Different exemplar based models differ on how new tokens are assigned to relevant categories or integrated with pre existing exemplars. All exemplar based models assume, however, that the entire system of the category will shift slightly with each addition of a new experience. That is, as perceptual memories associated with a category accumulate and are incrementally updated, their distribution may shift. Empirical evidence in support of the exemplar based models came from different domains of linguistic research inc luding studies in speech perception. Studies by Goldinger (1996) & Johnson (1997), for example, have shown that listeners relied on fine grained phonetic details in word recognition suggesting that these phonetic details are coded in long term representati on and are available during processing. Consistent with the exemplar based models, speech categories comprised mainly of idealized exemplars should be different from the ones that consist of diverse and varied samples of the category, and that processing m echanisms and/or outputs from these two different category systems may vary. In this study, perceptual memories associated with two different category formation systems of Thai tones were compared and contrasted: input to one system comprised only of ideal ized exemplars of the category, while those

PAGE 16

16 of the other was phonetically diverse and varied. It was hypothesized that mental representations of Thai tones developed by the second type of input would be richer and thus more robust than those formed by the first system. To test this hypothesis, a study designed to compare and contrast the effectiveness of two different training methods for lexical tone perception was conducted. In the first training method, the prototype method, participants were trained to perceive a contrast between the Thai mid and low tones using ideal exemplars of these two tones produced slowly and carefully in isolation by a female and a male native Thai speaker. On the other hand, participants administered the second training method w ere trained to discriminate between these two Thai tones u variability stimuli produced in different phonetic contexts, and in a relatively faster, conversational style by multiple speakers, two female and two male native Thai sp eakers. The effectiveness of the two training methods was measured twice, once immediately after training and then again two weeks after training had ended. Effectiveness was measured by (1) perceptual accuracy rate (percentage of correct responses), (2) r eaction times, and (3) ability to make generalizations when (a) old stimuli are produced by new speakers, (b) new stimuli produced by old speakers, and (c) new stimuli produced by new speakers. Participants were tested on two ty pes of stimuli: prototypical such as low variability stimuli produced in citation forms and high variability stimuli produced in fast, conversational style by multi talkers. Specific research questions are the following: 1 Which training method (prototype vs. high variability) will res ult in a higher percentage of perceptual accuracy immediately after training and for which type of stimuli ?

PAGE 17

17 2 Which training method will produce faster reaction times in perception immediately after training and for which type of stimuli ? 3 Which training meth od will lead to a greater ability to make generalizations immediately after training? 4 Which training method will be more effective (as measured by perceptual accuracy, reaction time, and ab ility to make generalizations) two weeks after training? Based on t he above research questions, it was hypothesized that: 1 Perception of the two Thai tones would improve after training among participants in both groups. However, since learning to categorize the two tones from highly variable stimuli may take longer than le arning from low variable stimuli, a greater amount of improvement would be observed among participants in the Prototype training group than among those in the High Variability training group, particularly on low variable, prototypical stimuli immediately a fter training. On the other hand, if high variability training resulted in a more robust representation of the two tones, then participants in the High Variability training group should outperform those in the Prototype group on both types of stimuli two w eeks after training. 2 Along the same line of reasoning provided in hypothesis 1 above, it was expected that reaction time would be shorter, particularly for low variable sti muli among participants in the P rototype training group when tested immediately afte r training. However, two weeks after training, those in the High V ariability training group were expected to exhibit shorter reaction time s for all types (low and high variable; new and old) of stimuli at two weeks after training. 3 A more robust representat ion of the two tones resulted from training with highly variable stimuli which should also lead to better generalization abil ity. Thus, participants in the High V ariability training group should be more accurate in their perception of the tw o Thai tones th an those in the P rototype training group when tested with new stim uli two weeks after training. 4 Similarly, their reaction time to new stimuli should also become shorter when compared t o those of participants in the Prototype training group. The remainder o f the dissertation is organized as follows. In Chapter 2, I give general background on the Thai language then present differences between Thai and English prosodic systems, issues of second language learning and cross language speech learning, abstractioni st vs. prototype theories of learning, and effects of training on SLA in general and on tone in particular. Chapter 3 presen ts the research

PAGE 18

18 methodology. Chapter 4 presents the results and statistical analysis on percent correct identification and Chapter 5 present s the results and statistical analysis on reaction time. In the last chapter, Chapter 6 I focus on the implications of the results, the limitation of the present study and s uggestions for future research.

PAGE 19

19 CHAPTER 2 BACKGROUND In this chapter, b ackground information on the Thai tone system is first presented. This is followed by a brief overview of differences and similarities between Thai and English prosodic systems. Previous studies on Thai language acquisition by foreigners are also reviewed and discussed. Prototype and exemplar based theories of memory and learning are briefly reviewed and three current speech perception models, namely Magnet effect (NLM) are discussed. Since the focus of the current study was on the effectiveness of laboratory training, results of previous training studies are also reviewed. Thai Tone System Thai is the national and official language of Thailand, a cou ntry in Southeast Asia It is the mother tongue of the Thai people, the dominant ethnic group in Thailand The Thai language is spoken in Thailand, Northern Malaysia, Cambodia, Southern Myanmar, an d Laos. There are about 63 million total speakers (Pakasri, 2006) Thai is a member of the Tai group of the Tai Kadai language family, the language family that is believed to have originated in what is now southern China F our main dial ects of Thai spoken in Thailand include Northern Thai, Central Thai, Northeaste rn Thai, and Southern Thai (Tingsapat, 1982) The Central Thai spoken in the capital c ity of Bangkok is considered the standard and official dialect of Thailand; it is used i n the government school system. The Thai phonological system consists of 21 conson ants (see Table 2 1) and 21 vowel nuclei: nine monophthongs and three dip h thongs (see Table 2 2). Phonemic

PAGE 20

20 contrastive length is found in Thai as each of the nine monop h thongs may occur as long or short with different meaning s for example, short [ a ] and l ong [ a: ] as in [ ak ] [ a:k ] Phonetically, long vowels are approximately twice as long as short vowels ( Hudak, 1990) Thai is a tonal language, which means that the tone or pitch is used to dete rmine the meaning of a word ( Smith, 1995, 2005) In other words, besides consonants and vowels, difference s in lexical tones or voice pitch are employed to distinguish the meaning of Thai words Therefore, people learning Thai must be able to perceive and reproduce tones correctly in order to effectively communicate G ood pronunciation of consonants and vowels alone are not sufficient ( Poomsan, 1995, Vongvipanond, 2000; Wittayasakpan et al., 2002; Wittayasakpan 2005). In his book, David Smith not onl y informs the learners that Thai is a tonal language and that tones are very important in determining a T meaning, but he also not ed that complex than in more familiar European languages, the learner will probabl y find Thai grammar considerably easier to absorb, for there are none of the complex verb tenses experience of language learning (Smith, 1995: 1). In Thai there are five phonemic tones and four tone marks. The characteristics of Thai tones are presented in Table 2 3. Static tones are classified as having a level pitch, while dynamic, or moving tones, are identified as having a contour pitch (Satravaha, 2002). The fundamental frequency (F0) of the five tones in isolated monosyllabic words is shown in Figure 2 1 (adapted fro m Luksaneeyanawin, 1998: 377). It is these prototypes that are usually taught to Thai learners. Although the five tone system in

PAGE 21

21 Thai seems simple at first glance, their d istributions are in fact rather complex. Not all syllables can bear all five ones which means that syllable structure plays a significant role in Thai tone assignment. According to Zsiga (2006), only CVV, CVS, and CVVS can bear all five tones. S stands for sonorant in Thai: /m, n, sound that is produced without turbulent airflow in the vocal tract such as nasals and approximants). However, on CVO syllables, only high and low tones can occur, and on CVVO syllables, only falling and low tones are allowed. O stands for obstruent in Thai including /p, t, k/ (obstruent means a sound that is produced by obstructing airflow, causing increased air pressure in the vocal tract such as stops and fricatives). Zsiga also claimed that the tone bearing unit (TBU) in Thai is the mora, not the syllable, and static tones are aligned to the rightmost m ora, while contour tones such as falling and rising have one tone per mora. Concerning variation of Thai tones, it has been observed that the phonetic realization of Thai tones v aries due to several factors. The tonal context (tonal co articulation between neighboring syllables) (Abramson, 1979) and sentence intonation (Luksaneeyanawin, 1998) are important factors influencing the pitch/fundamental frequency patterns of Thai tones. Additionally, Arunreung (1990) found that in Bangkok Thai, there was variation between generations of speakers which suggests that the language is undergoing change and that age is also a factor. Pitch levels and/or pitch contour of Thai tones have change d somewhat over time (Bradley 1911, Henderson 1949, 1982, Abramson 1962, and Arunrueng 1990) as shown in Table 2 5. In this table, t he number s 1 through 5 represen t the pitch level of the tones starting from 1 (low) to 5 (high). As shown in the table, Thai tones have change d gradually over

PAGE 22

22 time, particularly the contour tones. F or example, the falling tone has changed from 31 ( mid to low ) in 1911 and 1928 to 343 (mid hig h mid) in 1990. Similarly, the pitch contour of the rising tone has changed from a l inear rising tone (35 or 14) before the tones (mid, low and high) seem to have been alternately changed from a falling to a level pitch contour over time. Howeve r, it is uncertain to what extent the changes thus their notations, particularly among earlier studies where supporting acoustic measurements were not available. It is im portant to note that fluent Thai speech is usually fast and not every word is clearly produced. A s shown below, Thai tones are not actually pronounced in complete forms but rather are generally spoken in reduced forms While native Thai listeners are able to perceive the tones by relying on the context L2 learners of Thai have difficulty perceiving the se reduced tones According to my p reliminary study, tones p roduced in connected speech are generally shortened and do not show the complete forms of pitch c ontours. For example, the rising t one does not rise in natural sp eech as the prototype does and is therefore similar to the mid and the low tones Because of this variation, rising tone may be incorrectly perceived by L2 learners as either mid or low tone Figure 2 2 and F igure 2 3 show the pitch track s of the mid tone and the low tone produced in conversational style by three speakers As seen in the figure s the pitch levels of these tones v ary from speaker to speaker conside rable amount of overlap between the pitch heights of the two tones such that a low tone produced by a high pitched speaker is similar to a mid tone produced by a low

PAGE 23

23 pitched speaker. Moreover, preceding or following tones affect the target tone. For exampl e, as seen in Figure 2 3, the F0 of the low tone has a higher ending point than the low tone in citation form because the following tone is the high tone. According to previous studies, the mid and low tones are the most difficult pair to discriminate, eve n for native speakers of Thai, especially when they are perceived in isolation (Wayland & Guion 2004; Sathiansukon, 2005; Sittikesorn, 2005) Differences between Thai and English P rosodic S ystem s Prosody can be considered as the supras egmental features whi ch involve patterns in tone, tempo, and loudness that constitute the melody or rhythm or tune of a spoken language (Warren, 2000) The main difference between Thai and English lies in how voice pitch or an auditory impression of rate of vocal fold vibratio ns functions linguistically in these two languages. Thai is a tone language, or a language in which a difference in voice pitch is used to differentiate word meaning. It is important therefore, to pronounce the tone associated with each word accurately in order to distinguish meanings (Kruatrachue, 1960). It has been estimated that as many as 60 70% of the (Goldsmith, 1994; Yip, 2002). Some examples of tone languages are Mandarin Chinese (885,000,000 speakers), Thai (60,000,000 speakers), Yoruba (20,000,000 speakers), and Swedish (9,000,000 speakers). In non tonal languages such as English, on the other hand, variation in pitch level or pitch contour is used to express syntactical information (intonation) ( Bolinger 1949, 1962, 1 978; Henderson, 1965; Pike 1975). accent, stress, intonation and rhythm have been examined by linguists such as Rudaravanija (1965), Hiranburana (1971), Luangthongkum (1978), Luksaneeyanawin

PAGE 24

24 (1983), Potisuk et al. (1996). The effects of stress and intonation on Thai tones have also been investigated by Satravaha (2002). According to Luksaneeyanawin (1983), features such as a higher degree of respiratory effort, length, pitch, loudness etc. as compared with the unstressed syllable. However, the terms used interchangeably in studies on Thai accentual system. S ome Thai linguists (Luksaneeyanawin, 1983; Potisuk et al., 199 0; Thubthong et al. 2000) have reported that Thai is a fixed accent language in which the accent or the stres s is always assigned to the last sy llable of the word. F or polysyllabic words, the syllable final position is also the position of primary stress Concerning rhythmic categories, Thai has been considered a stress timed language (Luangthongkum, 1977) In th e original formulation, a stress timed language refers to a language that each syllable ha s approximately equal duration whereas a syllable timed language refers to a language that each foot ha s equal duration (Dauer, 1983) However, according to Grabe & L features, Thai should pattern with the stress timed languages because Thai has high Pairwise Variability Index (PVI) values. According to Low, Grabe, & Nolan (2001), the PVI i s a phonetic measure of variability between vocalic and intervocalic speech durations and they predicted that stress timed languages should have high PVI values. In an analysis of intonation in Thai, Luksaneeyanawin (1983: 182) proposed that ntour of any syntactic unit is the realization of the integrated system of

PAGE 25

25 phonetic characteristics such as pitch leveling and lengthening of duration. The common phone tic characteristics in stylized forms of speech such as chanting, recitation, calling, etc. are the following: 1 The pitch configuration of the tones is levelized: the static tones become level while the dynamic tones become a stepping narrow fall or rise. 2 T he pitch range is narrower than normal intonation 3 In some extreme cases, such as in chanting and professional calling where the meaning of the words is unimportant, the dynamic tones are realized with level pitch 4 The extension of the syllable duration is v ery prominent 5 The degree of loudness is higher than normal speech, especially in callings Luksaneeyanawin (1998) who investigated T hai lexical tone and sentential intonation, proposed that, to distinguish the meaning of the word, the system of tone in Tha i is more important than the system of intonatio n since the latter just adds attitudinal and emotional meanings to the word Unlike in Thai, pitch in English is used to distinguish 'lexical stress' In other words, non tone languages such as English do not depend on variation in pitch in order to distinguish the meaning of one word from another In English, variation in pitch height at the word level is one of the cues used to make one syllable in a multi syllabic word more pro minent than others. A t the syn tactic level, differences in pitch are employed to signify a statement or a question Eng lish has two types of stress: (1) lexical stress which refers to an accentuation of syllables within words, and (2) sentential stress which refers to an acc entuation of words within sentences (Cutler, 2005) Stres s in English facilitates word recognition. L exical stress is

PAGE 26

26 used to distinguish the word class of stress m inimal pair words such as pro duce (verb) vs. produce (noun) (Wiltshire & Moon, 2003) It is worth no ting that there are varieties between English used in different countries such as British English, American English, Australian English, and Indian English. For instance, in an investigation of the phonetic properties of stress in Indian English (IE), it w as found that the primary cue to lexical stress is increased duration. Moreover, unlike American English (AE), IE stress rules are conditioned by quantity or the number and size of the syllables in a word; for example, the main IE stress rule states that s tress is placed on the rightmost heavy syllable, if there is one; if not, stress is placed on the first syllable of the word (a heavy syllable ends with a tense vowel or with a lax vowel followed by at least two consonants). The IE stress rules are not rel ated to morphological category such as nouns or verbs like the rules in AE (Vijayakrishnan 1978 cited in Wiltshire & Moon, 2003). Also, in IE amplitude and durational differences between stressed and unstressed syllable are much smaller than those in AE. F or example, on stressed syllable, IE speakers produce only a .15 dB average increase in volume, while the AE speakers produce a 3.27 dB increase on average. These differences may cause miscommunications between the two groups or slower processing for the A E speakers (Wiltshire & Moon, 2003). Due in part to a difference in the linguistic function of pitch in tone and non tone languages, n ative speakers of non tonal languages often have difficulty acquiring languages with tonal systems such as Mandarin Chines e or Thai (Bluhme & Burr, 1971; Kiriloff, 1969; Wang et al., 1999; Potibul, 2005). It has been suggested that perceiving tonal distinctions is o ne of the most difficult challenges American English speakers face

PAGE 27

27 when they learn a tone language. ( Wang 1999, 2001; Wa yland & Guion 2003, 2004). Because AE speakers lack prior experiences using pitch differences t o distinguish the meaning of word s, they find it difficult to notice the pitch differences when they learn Thai tones. Previous S tudies on Thai T one A cqu isition Even after non native speakers of Thai have studied the language for a long period of time and are able to communicate in Thai, they may still have problems perceiving and pronouncing the tones (Ponmanee, 2002a, 2002b; Wittayasakpan, 2002; Sittikes orn, 2005; Sathiansukon, 2005; Potibal, 2005; Udompan, 2005; Aroonmanakun, native language (L1). Native knowledge of one prosodic system such as lexical tone or lexi cal stress may affect the acquisition of other systems. Wayland & Guion (2006) argue that adult learners can learn some aspects of a new prosodic system differing from their native system although they may acquire it imperfectly. According to Best (1995) a nd Flege (1995), similarities and differences between the L1 and L2 phonological systems can lead to perception and production difficulties among adult L2 learners Wayland et al. (1997, 2003, 2004, 2007) investigated the effects of native language on lexi cal tone perception They mention that although nave listeners demonstrate the ability to discriminate lexical tones, their performance is conditioned by their native language background Wayland (1997) examined the contribution of both segmental and supr a segmental parameters to the perception of foreign accent s in order to investigate whether the production of Thai consonants, vowels, and tone s by native Thai speakers and native English speakers differ in terms of acoustic

PAGE 28

28 measurements and auditory eval uation by native Thai speaking listeners The results showed that adult native English listeners were almost successful in learning temporal aspects, such as voice onset time and vowel duration, of certain Thai consonants and vowels However, vowel quality and tones in Thai that is, formant frequencies and fundamental frequency, still proved difficult for the native English speakers to master. Moreover, the production of Thai tones by adult native English learners was easily distinguished from native produ ction by native Thai listeners, while some non native ve demonstrated that each Thai tone poses different degree s of difficulty for non native sp eakers. L evel ton es seem to be more difficul t than contour tones for native English learners to master; the rating scores for level tones were lower M oreover, it was demonstrated that first language background indeed affects th e perception of Thai tones: native Chinese listeners had a greater abili ty of detecting and discriminating Thai tones (phonetic variations of the mid and low tones) than native Engli sh speakers both before and after training (Way land & Guion, 2004). Based on the scores of the P re test and P ost test, the native C hinese listener s showed grea ter improvement than the native English listeners in both the shorter and longer inter stimulus interval ( ISI ) conditions (500ms vs. 1500ms.) Thus, these re sults suggest that speakers of tonal language s such as Chinese have an advantage in L2 tonal perception because they can transfer the ir prior experience from their L1 tone system to another tone language like Thai whereas speakers of non tonal language s such as English are unable to transfer their experience with pitch successfu lly. In acc ordance with my preliminary study, many Thai teachers who teach Thai to speakers of both tonal

PAGE 29

29 languages like Chinese and non tonal languages like American E nglish have anecdotally found the above observation to be true. A numb er of studies also support th e findings that non tone speakers have problems in learning another tone language. In previous studies of cross language differences in supra segmental features, the results suggest ed that the native speakers of a non tone language such as English have pro blems percei ving or acquiring tones in tone languages such as Thai and Chinese (Burnham et al., 1992; Burnham & Francis, 1997; Wang, 1999, 2001) For instance, in a comparison between native Thai speakers and native Australian English speakers with no know ledge of Thai, the Thai speakers do better in the discrimination of tone contrasts (Burnham et al., 1992). Further, it has been reported that even after two or three years of studying Thai at Chiang Mai University, Thailand, Korean students self reported t hat the most difficult aspect in learning Thai was tone pronunciation (Ponmanee, 2002a). Potibal (2005) examined other Korean students who have been studying Thai for four years at Hankuk University of Foreign Studies, Seoul, Rebublic of Korean and similar ly found that they had great difficulty pronouncing Thai tones. He noted that their errors stemmed from their inability to correctly produce pitch height for level tones and pitch contour for contour tones. Their production of contour tones was too short f or the pitch contour to be fully and accurately realized. Korean students showed the influence of their mother tongue as they applied the Korean stress and intonation system to Thai. For instance, they pronounced words at the beginning of sentences incorre ctly by changing a mid tone to a high tone because they usually placed stress there and this was consistent with Korean stress pattern (Potibal, 2005). In addition, conforming to Korean intonation

PAGE 30

30 (Potibal, 2005), Koreans tend to produce high intonation at the middle of sentences in their L1 and they often do s o when they speak Thai as well. Udompan (2005) pointed out that Indonesian speakers also frequently have problems with tonal pronunciation due to the influence of their mother tongue The significant differences between Thai and Bahasa Indonesian reside in the length of vowels and in tones. N either of these features is phonemic in Bahasa Indonesia, but they have phonemic status in Thai Due to the influence from their native language, Indonesian stud ents do not pay enough attention to tones Interestingly, it has been suggested that different tones pose different levels of difficulty for Indonesian students ; they were able to accurately perceive and identify the contour tones better than the le vel ton than level tones for non native listeners to perceive and perhaps produce. This is because a relatively more drastic change in fundamental frequency from high to low or f rom low to high found in contour tones may be more perceptually salient than a gradual linear change of F0 exhibited by level tones (Wayland, 1997 ). Udompan (2005) confirmed that changing the to nes. Danish students also have difficulty acquiring Thai tones. This was made evident by a study designed to investigate self reported perception (using open ended The results confirm that, although these Danish students finished an Intensive 60 hour Thai Language Course, they were still highly concerned about their ability to perceive Thai tones correctly (Sathiansukon, 2005)

PAGE 31

31 Effects of E xperience in S peech P erception and P r oduction L earning Differences and similarities between the L1 and L2 sound systems, and experiences with the target language are important factors affecting L2 speech perception and production learning Both phonemic and phonetic differences and similariti es are considered the main sources in cross language speech misperception and misproduction While L1 and L2 phonemic system comparison known as a comparison at the phonetic level is believed to be tter predict or account for cross language speech learning outcome. Current speech perception models such as the Speech Learning Model (SLM) (Flege, 1995), the Perceptual Assimilation Model (PAM) (Best, 1995), and the Native La nguage Magnet (NLM) (Iverson & Kuhl, 1995) take differences and similarities at both the phonemic and particularly at the phonetic level as main factors in the models The Speech Learning Model (SLM) The purpose of the SLM is to account for success or fai lure in producing and perceiv ing phonetic segments in second language acquisition The SLM focuses on difficulty of learning certain L2 speech sounds, age related limits in L2 sound learning, and speech perception in terms of phonetic dissimilarity between L1 and L2 The core assumpti on of the SLM is that the ability to learn new speech sounds r emains malleable across on e life span The SLM proposes that the greater the dissimilarity between an L2 speech sound and a closest L1 sound is, the more likely the learners will notice the di fferences between the L1 and L2 sound Moreover, the SLM claims that learning L2 sounds at an early age is more effective than learning as adult learners This claim has been supported by a number of studies ( Yamada, 1993, Flege & Munro et. al., 1995)

PAGE 32

32 The Perceptual Assimilation Model (PAM) In the domain of cross language speech perception, the Perce ptual Assimilation Model (PAM) ha s been one of the most influential models Best and her colleagues proposed and developed the model to account for the process of speech perception of non n ative sounds (Best, McRobert & Sithole, 1988; Best, 1995) The PAM viewed that different degrees of discriminability of pairs of non native (L2) sounds is related to the perceived distance between articulatory gestures of a pair of L2 sounds and the L2 sound from L1 sound. Specifically, learners perceive L2 sound contrasts based on the degree of similarity as well as the discrepancy between the native and non native sounds. According to PAM, there are three main perceptual a ssimilation types First ly a tw o category (TC) assimilation type where members of an L2 contrast assimilate to two different L1 sound categories and discrimination of this contrast is predicted to be relatively easy to discriminate. Secondly, a Category goodness (CG) assimilation type: a pair of L2 sounds assimilate to the same L1 category, but is judged to differ in degrees of goodness of fit to the L1 sound. Discrimination is predicted to be relatively more difficult than the TC assimilation pattern. T hirdly, a single category (SC) type: two L2 sounds assimilate to the same L1 category and both are judged to be equally deviant or similar to the L1 sound. Discrimination is predicted to be the most difficult. Therefore, according to their difficulty in di scrimination, these three perceptual assimilation types can be ranked from easy to hard as follows: (TG) > (CG) > (SC). Moreover, the PAM also formulated three other types of perceptual assimilation The first one is the Uncategorized Categorized (UC) type: on e L2 sound is classified as one of the L1 sounds but the other is not It is predicted that discrimination for the (UC) is predicted to be high or good. The second one is the Uncategorized Uncategorized

PAGE 33

33 (UU) type: both L2 sounds cannot be classified into any L1 category and discrimination is predicted to range from fair to good Finally, the Non Assimilable (NA) type: both L2 sounds cannot be classified into any L1 category and are perceived as non speech sounds Discrimination is predicted to range from good to excellent depending on phonetic difference between the two foreign sounds. Assimilation Model (PAM) (Best, 1995), it is possible that American English speakers may perceive two lexica l tones in Thai as the (UU) or the (NA) type. Discriminability of a pair of tone will thus depend on the phonetic distance between them. The Native Language Magnet Effects (NLM) This perception and the model was also propose perception development so that it is particular to their linguistic experience or their native language (Kuhl, 1991). The Native Language Magnet Effects (NLM) proposes that the native category goodness of speech sou nd is related to speech perception. According to NLM, an exceptional exemplar of phonetic categories, the prototype (P), was more difficult to be differentiated from its variants than a poor exemplar or the non prototype (NP) because the prototype in a cat egory assimilates the non prototype members of the same category and shrinks the perceptual boundaries around it. This causes discrimination between prototype and non prototype to be more difficult. This effect was referred to as the perceptual magnet effe ct (Kuhl, 1991, 1993; Kuhl et al., 1992, Iverson & Kuhl, 1996). However, the NLM does not offer specific prediction for learners who acquire a speech sound that does not fall into any category in their native language such as American English speaker s who learn tones.

PAGE 34

34 Exemplar B ased Models and Prototype B ased Models When teaching Thai tones, most Thai teachers have long employed a traditional method consistent with learning model Acco model mental representation of a category is the prototype Although different researchers defined a (Barsalou, 1990; Nosofsky, 1987, 1992b; Posner & Keele, 1968; Reed 1972), the is the exemplar with average values on all of the s (Reed 1972; Rosch, Simpson, & Miller, 1976) as a member of the category if it i s sufficiently similar to the prototype. The prototypes of the five contrastive Thai tones in c itation forms (as shown in F igure 2 1) are usually taught by Thai teachers to L2 learners Unf ortunately, as discussed earlier after learning the Thai tones usi ng this method, a substantial number of L2 learners of Thai still have difficulty in producing the tones correctly. exemplars they experience as whole, independent memory tra ces. Each category thus has no independent representation and instead is represented as collections of exemplars (Estes, 1986; Medin & Florian, 1992; Nosofsky, 1986; Reed, 1972) A new stimulus is s and accepted as a memb er if it is similar enough to the exemplars (Estes, 1986; Nosofsky, 1986) As a result, all exemplar s do not always share the same properties Based on the different degree of an the others Therefore, instead of using unrealistic prototypes to teach T hai tones, using high variability stimuli to teach tones would incorporate learning the tones in a variety of

PAGE 35

35 contexts. If nonnative learners of Thai learn to perceive high variabi lity stimuli, their mental representations of Thai tones would be richer and more robust enabling them to perceive Thai tones correctly as they are reali zed in natural, native speech. It has been demonstrated that high variability stimuli positively affec t speech perception learning (Mullennix & Pisoni, 1990; Jusczyk, Pisoni & Mullennix, 1992). For example, after learners were trained with high variability stimuli produced by different speakers and at different speaking rates, their ability to perceive a n ew set of h igh variability speech stimuli such as produced by different speakers and in different phonetic contexts significantly increased Although their performance declined at the beginning of the training, it significantly improved later (Mullennix, 1989; Mullennix & Pisoni, 1990; Sommers et. al., 1994). Moreover, word identification accuracy has also been shown to initially decrease under the high variability training paradigm, but was found to later call of words (Goldinger, Pisoni, & Logan, 1991). Similarly, Sommers & Barcroft (2007) found that, although variability in speaking rate initially impedes word identification, it ultimately enhances L2 Variability of stim uli significantly improves Participants trained in this manner are also able to apply this ability to new speech sounds as well as to sounds produced by unfamiliar speakers (Lively et al., 1993, 19 94) The ability to make generalizations to new talkers and new contexts is very important in learning Thai tones since there is an enormous amount of tonal variability in a real wor l d environment. It should be noted that the effectiveness of this training paradigm has been tested mainly on the learning of consonants and vowels The goal of this study, on the other hand, is

PAGE 36

36 to explore if the high variability training paradigm is more effective in improving L2 the traditional method Training S tudies ( H igh V ariability T raining P aradigm) Numerous studies have shown that non system in both segmental and suprasegmental distinctions can be improved by perceptual/auditory trainin g in a laboratory ( Pisoni, Aslin, Perey, & Hennessy, 1982; Jamieson & Morosan, 1986; Logan, Lively, & Pisoni, 1991) For example, after American English listeners were trained in identifying non native voicing contrasts, their performance significantly imp roved The American English listeners were able to perceive voiceless aspirated, voiceless unaspirated, and voiced stops differing in voice onset time (VOT) after training (Pisoni, Aslin, Perey, & Hennessy (1982) Similarly, after auditory training, native speakers of French were able to discriminate between the two American English fricatives: / / (Jamieson & Morosan, 1986). Besides perception of segments, perception of supra segments such as tones has also been shown to improve with training. For exampl e, it has been found that both perception and production of Mandarin tones by non native speakers of Mandarin were significantly improved after the non native speakers were trained with short perceptual tone training (Wang et al., 1999, 2003a). Moreover, L eather (1990) found that, in a group of Dutch speakers, production training on Mandarin tones had an effect on the tones. Based on these results, Wayland & Guion, (2004 ) conducted a study focusing on the transfer of prior experience with a tone system in a native language to experience with another tonal language. The findings suggested that the native speakers of

PAGE 37

37 Chinese (NC) transferred their native ability in discrimi nating Chinese tones to their perception of Thai tones. Unlike the NC group, the native speakers of English (NE) lacked phonological representation of tones in their native long term memory to help facilitate them in this process. Thus, even after the trai ning, the NE speakers did not show significant improvement in perceiving Thai tones. Further, a comparison between nave American English speakers of Thai and experienced American English learners of Thai confirms that experience with the target language ( Thai) does facilitate these nave American English listeners in acquiring tones (Wayland & Guion, 2003). The above findings that perception of non native segments and supra segments improved with experience either in a laboratory or in natural setting chal lenged the strong version of the Critical Period Hypothesis (CPH) (Lenneberg, 1967; Scovel, 1969; Patkowski, 1989). According to CPH, plasticity of the human brain declines with age and language learning may be difficult or impossible beyond a critical tim e window perception of L2 sounds can be improved suggested, instead, that the human span (Wayland & Guion, 2004). native speakers of Thai might need more extensive training in order to reach the ability to perceive tones at the same level, or near the same l evel as native speakers. To using high variability stimuli produced by different speakers, both male and female, and with a fast speaking rate, conversational style w as employed. Furthermore, besides

PAGE 38

38 accuracy rate, the effectiveness of this training paradigm was also measured by novel stimuli including new words produced by both new and old talkers. Relationship between Speech Perception and Speech Production This current research focused entirely on perception and excludes production. Nonetheless a brief theoretical overview of the relationship between speech perception and spee ch production is presented here. According to the Motor Theory of speech perception, there is a link between speech perception and speech production (Liberman, 1996; Liberman & Mattingly, 1985). Moreover, training in the perception domain can be transferre d to better performance in the production domain. This notion is supported by a case study of Japanese learners who were trained to identify English /r/ /l / minimal pairs. These learners showed a close link between speech perception and production with tr aining in one domain such as perception resulted in significant improvement in the other domain such as production (Akahane yamada, Bradlow & Pisoni, 1996). The direct realist theory of speech perception (Fowler, 1986) supports the Motor Theory in terms of an inextricable link between speech perception and speech production However, according to the direct realist theory of speech perception, integrated event perception and action systems play a role in the perception production link, not a specialized pho netic module as proposed by the Motor Theory. Moreover, according to acoustic properties is the object of speech perception. On the other hand, besides taking acoustic and au ditory patterns of speech as the object of speech perception, proponents of the acoustic auditory theories of speech perception prescribe only an

PAGE 39

39 indirect link between speech production and perception (Steven & Blumstein, 1981; Diehl & Kluender, 1989). In this study, two Thai tones, mid and low, were selected due shown that these two tones are the most difficult to discriminate, even for native speakers of Thai, especiall y when they are perceived in isolation (Wayland & Guion 2004; Sathiansukon, 2005; Sittikesorn, 2005). The similarity between the low and mid tones is shown in Figure 2 4. For this experiment, native speakers of American English were chosen as participants because their native language is non tonal which makes it difficult for them to perceive Thai tones (Wayland & Guion 2003, 2004). According to Best (1995) and Flege (1995), similarities and differences between the phonological systems of Thai and American English may constitute significant factors causing tonal perception difficulties for American learners. Importantly, this research focuses on finding a better, more efficient method in teaching and learning Thai tones. Despite the fact that a number of stu dies have been conducted on teaching Thai to foreigners, none have exclusively focused on tones (Ponmanee, 2002a; Potibal, 2005; Sathiansukon, 2005; Sittikesorn, 2005). Even among those that emphasized phonology including consonants, vowels, and tones, ton e is usually considered to be only a secondary aspect to investigate. Most importantly, the most effective method in teaching Thai tones has yet to be decided. Therefore, an experiment grounded in current linguistic theories should be conducted to find a m ore efficient solution. Equipped with a better teaching method, teachers of Thai would be able to help non native learners to achieve their goal of learning Thai tones. A faster method would enable non native learners learn Thai more quickly.

PAGE 40

40 In sum, the a bove literature review suggested that due in part to a difference in the linguistic function of pitch in English and Thai and to the phonetic similarity between the Thai mid and low tones, native speakers of English would have difficulty perceiving the dif ference between these two tones. However, previous studies also suggested that the amount of perceptual difficulty they experience could be reduced with training. In addition, consistent with the exemplar based theory of categorization and learning, a trai ning paradigm using high variability stimuli has been proven effective in improving non native speech contrasts. The current study focused on the effectiveness of a high variability perceptual training paradigm on the p tone perception, it was expected that improved perception, if observed, would result in improved production without further production training given the relationship bet ween speech perception and production discussed above. Results of this currently study are of both the oretical and practical values. They would deepen our understanding of the mechanisms underlying the acquisition of a non native prosodic system in general and of lexical tones in particular, and inform teachers, particularly of tone languages, of a perhaps more theoretically sound and effective teaching method.

PAGE 41

41 Table 2 1. Thai c onsonants Bilabial Labio dental Alveolar Post alveolar Palatal Velar Glottal Plosive p p h b t t h d k k h Nasal m n Fricative f s h Affricate t t h Trill r Approximant j w Lateral Approximant l [Reprinted with permission from Abramson, A S. & Tingsabadh, K. (1999) Thai final stops: Cross language perception. Phonetica, 56, ( p age 118 ) ] Table 2 2. Thai vowels Place of articulation front central Back High /i i:/ / :/ /u, u:/ Mid /e, e:/ / :/ /o, o:/ Low / :/ /a, a:/ / :/ diphthongs /ia, i:a/ / :a / /ua, u:a/ [ adapt ed from Jitapunkul, S., Maneenoi, E., Ahkuputra, V., & Luksaneeyanawin, S. (2003). Performance evaluation of phonotactic and contextual onset rhyme models for speech recognition of Thai language. In Eurospeech 2003 (page 184 1 ) ] Table 2 3 Characteristics of Thai t ones Tone Tone m ark Pitch contour Pitch height Mid Unmarked Level Medium Low ` Level Low Falling Contour High to low High Level High Rising Contour Low to high [adapted from Hudak, T. (1990). Thai. In B. Comrie, (ed.) New York: Oxford University Press ( p age 760 ) ] Table 2 4. Thai t ones Tone Word Meaning Mid [na:] Low [n:] Falling [n:] High [n:] Rising

PAGE 42

42 Table 2 5 C omparison of phonetics of Thai tones (from 1911 1990 ) Tone Bradley 1911 Jones 1928 Henderson 1949 Abramson 1962 Henderson 1982 Arunrue ng 1990 Mid 332 332 33 332 33 32 Low 221 22 22 21 21 21 High 341 341 343 343 34 34 Falling 31 31 51 451 52 343 Rising 35 14 14 14 14 213 [adapted from Arunrueng, A. (1990). Variation of the fal ling tone by age of speakers of Bangkok Thai. MA thesis. Chulalongkorn University (page 16) ]

PAGE 43

43 Figure 2 1. F0 contours of the five Thai tones Figure 2 2. Thai mid tone syllable [ma:] produced in conversational style by 3 speakers Variat ion in Pitch Contour of Thai Mid Tone [ma:] 300 250 150 100 50 0 0 % 20 % 40 % 60 % 80 % 100 % Duration % 200

PAGE 44

44 Figure 2 3. Thai l ow tone syllable [ma:] produced in conversational style by 3 speakers Figure 2 4. The fundamental frequency traces of the standard Thai low and mid tones [ reprinted with permission from Wayland, R. & Guion, S.G. (2004). Training English and Chinese l isteners to perceive Thai tones: A Preliminary Report. Language Learning, 54 (4), ( p age 694 ) ] Variation in Pitch Contour of Thai Low Tone [ma:] 300 250 150 100 50 0 0 % 20 % 40 % 60 % 80 % 100 % Duration % 200

PAGE 45

45 CHAPTER 3 METHODOLOGY The overall design of this study involved testing three groups of native speakers of American English on their ability to perce ive the di stinction between two Thai tones: mid and low None of the participants had prior experience with Thai or any other tonal language such as Chinese or Vietnamese as determined by a language questionnaire These part icipants were divided into three groups: two experimental groups and one c ontrol group In addition, a control group of Thai speakers was tested. The study consisted of four phases: P re test, Training, P ost tes t, and R etention During training, participants in one experimental group (the Prototy pe group) were trained with prototypic al stimuli produced in citation forms while the other group (the High Variability group) was trained with high variability stimuli produced by multiple speakers in a faster, conversational speech style. Participant s in the control group did not receive any trainin g and only participated in the P re test and the P ost test (American English control group); or only in the Pre test (Thai control group). The effectiveness of the two training methods was assessed immediately a fter training and again two weeks after training. Besides accuracy rate and reaction time, the ability to make perceptual generalizations when presented with both old and new stimuli by new speakers was also evaluated. Participants Sixty nave native speak ers of American English: 27 men and 33 women and 10 Thai native speakers (5 men, 5 women) were recruited from the student population of University of Florida A ll were right handed with ages ra nging from 18 to 35 years old (mean = 24 years). All subjects were compensated for their participation at a rate of $ 8/hour, or if students participated for course credit, they received 1 credit/hour or up to a

PAGE 46

46 maximum of 2 credits for their participants The English participants were randomly divided into three gro ups: two experimental groups and one control group All three groups of participants were administered a Pre test and a Post test However, only two experimental groups underwent training. The Thai Control group only did the Pre test. 1. The Prototype tra ining group (20 people: 8 m en and 12 women) was trained to perceive the Thai mid and low tones with prototypical, low variability stimuli produced in citation form by two Thai speakers (one male and one female) at a single slow, clear speaking rate. 2. Th e High Variability training group (20 people: 9 men and 11 women) was trained with high variability stimuli. Stimuli varied due to the use of multiple speakers : 4 speakers (2 men and 2 women ), and speech occurred in a fast conversational style 3. The NE Control group (20 people: 10 men and 10 women) was administered the P re test and, four days later, the P ost test, but did not undergo training. 4. The Thai Control Group (10 people: 5 men a nd 5 women) w as administered the Pre test, but did not undergo training. Stimuli Thirty minimal pairs (60 words) that contrasted the mid and low tones in standard Thai wer e used as stimuli in this study. The 60 words we re real Thai words that we re frequently used in the standard such as Bangkok variety of Thai Addit ionally, in order to prevent an effect of syllable structures on tones, all stimuli we re monosyllabic words containing the same or similar syllable structures They we that is, syllables that end with a long vowel or a sonoran t final consonant: CVV, CVVS, or CVS (S are sonorants such as /m/ /n/ / / /w/ and /j/) This study focused on live syllables/words, not dead or checked syllables/words such as those ending with a short vowel or a stop final cons onant such as /p/, /t/, and /k/ Previous research has shown that subjects performed better on open syllables than on closed syllables (Wayland & Guion, 2003), perhaps due to the fact that the acoustic properties of low and mid tones

PAGE 47

47 in Thai are more salient in open syllables than in closed syllables This is because F0 onset value is different in open syllables while both F0 onset and offset values are different in closed syl lables (Wayland & Guion 2003). In addition, the mid and low tones were chosen for use in training due to their difficulty already discussed in Chapter 2. The inter stimulus interva l was set at 1 500 ms processing at the phonological level. According to Werker & Tees (1984) a phonological mode of processing occurred at the longer ISI (1500 ms) while a phonetic mode of processing occurred at the shorter ISI (500 ms) In addition, it is believed that prior linguistic experience affects processing at the phonological level, but not at the phonetic level of processing. The phonetic mode of processing is, therefore, viewed as a language general mode of perception (Burnham et al. 1992; Werker & tees 1983) Evidence for these two levels of processing has been reported on lexical tone native Thai speakers discriminated Thai tones better under th e longer ISI of 1500 ms while native Australian English speakers did better under the short, 500 ms ISI. The researchers asserted that this might have been due to the fact that native Thai speakers processed the tones at the phonological level and that the phonetic mode of processing was employed by native Australian English speakers. For further discussion on different modes of proce ssing, see Wayland & Li (2008). Stimuli for the Pre Test Two se ts of stimuli were used in the P re test. The first set consist ed of 10 minimal pairs of syllables produced with a mid or low tone in Thai. These 20 target words (as shown in T able 3 1 ) were produced in isolation by a Thai male ( M1 ) and a Thai female speaker ( F1 ) using a slow, clear, reading style. Each of the 20 targ et words was

PAGE 48

48 repeated 2 times generating a total of 80 tokens (20 words x 2 speakers x 2 repetitions). These 80 tokens were presented to participants twice yielding 160 trials in total. In addition, the trials were presented in two blocks: 80 trials from the male speaker and 80 trials from the female speaker The order of speakers (female male and male female) was random ized such that half of the participants were presented with stimuli produced by the male speakers first while the other half heard stimul i produced by the female first. The sec ond set of stimuli was the high variability set To create high stimulus variability, the same 10 target minimal pairs used to generate the first set were produced in various phonetic contexts Specifically, they were produced at the end of the carrier sentences (see A ppendix ) and may be preceded by a low tone, a mid tone, or a high tone As mentioned earlier in C hapter 2, preceding or following tones affect the target tone. For example, the F0 of the mid tone has a hig her beginning point if the preceding tone is the high tone; the F0 has the lower beginning point if the preceding tone is the low tone. Thus, variations in the tones that precede the target tone produce desired variation in the pitch contour of that target tone Therefore, each of the 20 target words (10 pairs ) was produced in one sentential context preceded by various tones yielding 20 tokens in total. These 20 target tokens were produced by the same male and the same female speakers as the first set, but produced in a fast, conversational speed, thus generating 40 tokens (20 tokens x 2 speakers). The 40 tokens were presented four times for a total of 160 tokens These 160 tokens were presented in two blocks: 80 tokens from the male speaker and 80 tokens fr om the female speaker with a random order of speakers (See A ppendix)

PAGE 49

49 Training Stimuli for the Prototype Training Group Training stimuli for the Prototype training group consisted of 10 minimal pairs (shown in T able 3 2 ). The se 20 target words (10 pairs) w ere produced by a Thai male ( M2 ) and a Thai female speaker ( F2 ), using a single speaking rate such as slow, careful reading pronunciation, generating a total of 40 tokens Note that these two speakers were not the same speakers who produced the pre test st imuli. These 40 tokens were presented to participants four times yielding 160 tri als in total (see A ppendix ). Training Stimuli for the High Variability Training Group The same 20 target words (10 pairs) used in the Prototype training group were used in the High Variability training group. However, these 20 target words (10 pairs) were produced in a sentence final context similar to the high variability stimuli set used in the P re test To increase stimulus variability, the target words were pronounced by tw o male (M2, M3) and two female (F2, F3) speakers using fast conversational speaking rate and style None of these speakers were the same as those who produced tones in the Pre test materials. A total of 8 0 tokens (20 words x 4 speakers) were produced and presented two times These 160 tokens were presented in two blocks: 80 tokens from the male speaker and 80 tokens from the female speaker ( See Appendix ). Stimuli for the Post Test Stimuli for the post test were identical to those used in the pre test. Ther e were two sets of stimuli. The first set consisted of 10 minimal pairs of syllables produced with a mid or low tone in Thai. The sec ond set of stimuli was the high variability set T he same 10 target minimal pairs used to generate the first set were produ ced in various phonetic contexts

PAGE 50

50 Stimuli for the Generalization Test Four sets of stimuli were used in the Generalization test which included five old minimal pairs (previously used in the P re tes t) and 5 new (not used in the P re test ) (see Table 3 3). Th ese target words were p roduced by either old speakers such as M2, F2 speakers who produced training stimuli or new speakers such as M4, F4 Similar t o P re test and Post test stimuli, these stimuli were produced with different speaking styles: in isolation, slow, citation style; and in sentence final context at fa st, conversational style. The Generalization test consisted of 80 tokens and 320 trials in total. Specific characteristi cs of the stimuli in each of the four sets were as follows. Set 1: 20 tokens ( 10 words x 2 talkers) of 80 tokens, old and new words produced by old talkers in sentence final context at fast, conversational speaking rate. Set 2: 20 tokens (10 words x 2 talkers) of 80 tokens, old and new words produced by new talkers in sentence fina l context at fast, conversational speaking rate. Set 3: 20 tokens (10 words x 2 talkers) of 80 tokens, old and new words produced by old tal kers in isolation and in slow, citation form. Set 4: 20 tokens (10 words x 2 talkers) of 80 tokens, old and new wo rds produced by new tal kers in isolation and in slow, citation form. to high variability stimuli while set 3 and Set 4 tested their ability to generalize to low vari ability stimuli. Stimuli for Retention Tests Two weeks after training, participants in the Prototype and the High Variability training groups were administered the same Post test and the Generalization test again. To avoid confusion they will be referred t o as Post test vs. Retention test and

PAGE 51

51 Generalization 1(or Gen 1) and Generalization 2 (or Gen 2) respectively. The stimuli and the format of these tests were otherwise identical. Task Tone identification task was used in this study. Two Thai tone, mid and low, were presented to participants. The p articipants were asked to indicate for each trial whether the tone they heard was a low or mid tone Their responses were recorded and calculated as mean reaction time. Procedure Pre Test Before the experiment beg an, all participants were informed that they were participating in a speech perception experiment. Then they were asked to complete a language and education questionnaire. Participants were tested individually in a quiet, laboratory room. The P re test cons isted of four blocks: two (one male speaker and one female speaker) with low variability stimuli and two (one male speaker and one female speaker) with high variability stimuli) and lasted approximately 30 minutes. Before taking the P re test, participants were presented with a short practice session with 10 trials in order to familiarize them with the stimuli and presentation rate (one stimulus per trial). They were told to focus on the difference between the mid and the low tone produced by Thai speakers During the test, they were asked to identify the low and mid tone by Half of the participants were tested with the low variability stimuli fir st and the other half were tested with the high variability stimuli first. The order of speakers (male speaker first/ female speaker first) was also counterbalanced across participants.

PAGE 52

52 Training After taking the P re test, participants in the High V ariabili ty training group and the P rototype training group were trained to identify the two Thai tones. The participants in the c ontrol group did not participate in any training. The participants heard one stimulus per trial and were asked to identify the tone cat egory by selecting the corresponding button: A for the mid tone and B for the low tone They could replay each trial as many times as they desired before making a decision However, after the participants made a response, they could not replay the sounds Feedback was given if an incorrect button was selected, the correct button would blink for 5 seconds The two tones, the mid tone and the low tone, were randomly presented Each participant completed four training sessions which lasted approximately 30 min utes each. The first session began on Day 1 after the Pre test. The second and the third sessions took place on Day 2 and the last session took place on Day 3. Post Test On Day 3 (after the fourth training session) participants in both training groups took the Post test To ascertain that the time lag between the Pre and the Post test was comparable for both the exp erimental (Prototype and High Variability Training) groups and the NE c ontrol group, participants in the control group were administered the Po st test on the fourth day after the y took the Pre test. Due to time limitations, we were not able to test them on the same day such as D ay 3, as those in the experimental groups. As mentioned earlier, the Post test was identical to the Pre test. This test thus tested the effectiveness of the training on stimuli heard on the Pre test. It is important to note that participants were not trained on Pre test stimuli. On da y 4 ( one day after the traini ng and the immediate Pos test) participants in the two train ing

PAGE 53

53 groups were administered the Gen 1 test designed to t est which training method wa s more effective in terms of how well it generalize s to new stimuli. Retention Test Finally, two weeks after training, participants in the two training groups were asked t o come back and took the Retention test ( a delayed Post test) and Gen 2 test. These test stimuli and on Gen 1 stimuli two weeks later. After training, p ercent correct i dentification of the two Thai tones for all subjects was calculated for statistical analyses. B ased on previous research, it was expected that the higher degree of variability in the stimulus set used to train native speakers of American English in the Hig h Variability training group would result in a more robust representation of the Thai low and mid tone s in their long term memory It was thus predicted that the American listeners in the High Variability training group would outpe rform those in the Protot ype training Group on both the Post test and the Generalization test. They were expected to gain the ability to distinguish the two Thai tones produced in new words and in both old and new words produced by new talkers.

PAGE 54

54 Table 3 1. Minimal pairs of the p r e test and the p ost test stimuli Mid tone Low tone 1 2 3 4 [ 5 [ [k 6 [t [ 7 8 [ph [ph 9 [ k [ k 10 [t Table 3 2 Minimal pairs of the training stimuli Mid tone Low tone 1 [pha j ] [ph j 2 [kh 3 [ph 4 [ph 5 [kh :j] 6 [th 7 [k 8 [t 9 10 [k Ta ble 3 3. Minimal pairs of the g eneralizatio n (p ost test) stimuli Mid tone Low tone 1 [ph 2 [ph n] 3 [t 4 [ kho m 5 [tha 6 [ th a [ th a 7 [pu : [pu : 8 [tha 9 [ th [ th 10 [ k : [k :

PAGE 55

55 Table 3 4. Summary of procedure Group Day 1 Day 2 Day 3 Day 4 Two weeks later High V ariability Pretest /t raining 1 Training 2, 3 Training 4 /p ost test Gen 1 R et ention G en 2 Prototype Pre test /t raining 1 Training 2, 3 Training 4 /p ost test Gen 1 Retention G en 2 Control (NE) Pre test Post test Control (Thai) Pre test

PAGE 56

56 CHAPTER 4 PERCEPTION ACCURACY: PERCENT CORRECTION I DENTIFICATION Percent co rrect identification of the two Thai tones obtained from both the Pre and the Post test was statistically analyzed using the Statistical Package for the Social Sciences (SPSS17) to test for the effects of Group (Prototype, High Variability, NE control, a nd Thai), Test Time (Pre test and Post test), and Stimulus Type (low variability and high variability). In this study, i ndividuals whose scores fell more than two standard deviations (SD) below the group mean were considered outliers and were eliminated fr om the statistical analyses. Therefore, two outliers were removed and it was three percentages. No training data was obtained from the NE control group since they did not participate in the training phase. The native Thai speakers did not participate in e ither the training or the Post test phase; therefore, only their Pre test scores were analyzed and compared with those of the non native groups. The results will be reported starting with Pre test and Post test. The Pre test data are compared among four gr oups: Prototype, High Variability, NE control and Thai groups; the Post test data among three groups: Prototype, High Variability and NE control. The effects of two stimulus types are also examined: low variability stimuli (produced in citation forms by tw o speakers, a male and a female speaker) and stimuli with high degree of variability (produced in sentence final position by two male and two by comparing the Pre tes t with the Post test among the three groups: Prototype, High Variability and NE control. Moreover, the results of the Generalization test comparing the two groups: Prototype and High Variability will be reported. In the last section, the results of the Ret ention test (two weeks after the training) will be reported, by comparing

PAGE 57

57 the Pre test to the Retention test, and the Generalization test 1 to the Generalization test 2 between the two groups: Prototype and High Variability respectively. Pre Test Mean perc ent correct identification of the two Thai tones, mid and low, for the four groups: Prototype (n=20), High Variability (n=20), NE control (n=20), and Thai groups (n=10), under two stimulus types: low variability and high variability and standard errors (SE ), obtained during the Pre test are reported in Table 4 1 and shown in Figure 4 1. Native Thai speakers earned the highest scores for both stimulus types (99.6% and variability stimulus than the high variability stimulus condition (Prototype group mean = 77.9% vs. 73.5%, High Variability group mean = 80.0% vs. 77.7%, NE control group mean = 82.8% vs. 77.4%, Thai group mean = 99.6% vs. 99.6% and overall mean = 83.0% vs. 79.6%). A repeate d measure s ANOVA with Group (Prototype, High Variability, NE c ontrol, and Thai) as the between subject s factor, and Stimulus Type ( low variability and h igh variability) as the within subjects factor was performed on this data As expected, the analysis yie lded a significant main effect of Group [ F ( 3, 66) = 6.90, p = .000]. Post hoc pair wise comparison using the Bonferroni method (p=.05) revealed that the Thai group outperformed all thre e non native groups [means = 99.6%, 75.7%, 78.8%, 80.1%, for Thai, Pro totype, High Variability, and NE control groups respectively, p <.01]. No significant difference among the non native groups was found [ F ( 2, 57) = .44 p = .646] T he main effect of S timulus T ype was also significant [ F ( 1, 66) = 12.06 p = .001] confirm ing the above observation that participants performed significantly better on low variability stimuli than on high variability stimuli (mean=85.1% vs. 82.1%). No significant

PAGE 58

58 interaction b etween Group and Stimulus Type [ F ( 1, 66) = 1.64, p = .187) was found In sum, these results suggested that, prior to training, native Thai speakers outperformed all non variability than on high variability stimuli, and performances of all 3 grou ps of non native speakers were comparable. Effects of Training The e ffect the four tests: Pre test, Post test, Retention test (or delayed Post test) and Generalization test. The Pre test was compared with the Post test (which were identical to those of the Pre test) and Retention test to evaluate the effect of training on the same words produced by the same speakers. Gen 1 and Gen 2 tests were also compared to assess if the effect of trai ning is generalized to (1) new stimuli produced by new speakers, (2) new stimuli produced by old speakers, and (3) old stimuli produced by new speakers as well as to investigate retention of the effects of the training. Pre Test to Post Test Comparison Me an percent correct identification for th e three groups: Prototype (n=20), High Variability (n=20), and c ontrol groups (n=20) under two s timulus type conditions : l ow and high variability, and two test times: pre test and p ost test, are reported in Table 4 2 and shown in Figure 4 2. Both experimental groups (Prototype and High Variability) performed better after the training under both stimulus type conditions : l ow variability and high variability. Concerning stimulus type, both experimental groups showed be tter performance for the low variability stimuli than for the high variability stimuli. Interestingly, the NE control group scored higher on the Post test than on the Pre test without training.

PAGE 59

59 This data was submitted to a repeated measure s ANOVA with Grou p (Prototype, High Variability, and NE c ontrol ) as the between subjects factor and Test Time (P re and P ost test) and Stimulus Type (low variability and high varia bility) as the within subjects factors The analysis yielded a significant main effect of Tes t Time [ F ( 1, 57) = 4 6.02, p = .001] and Stimulus Type [ F ( 1, 57) = 33.93, p = .001] ; however, a significant main effect of Group was not found [ F ( 2 57) = 00, p = .999 ] These results indicate that, when pooled across both types of stimuli, the participa significantly improved after training (mean = 78.1% vs. 85.3%). Similarly, across both Test Times, the participants obtained significantly higher scores for low variability stimuli than for high variability stimuli (mean = 83.6% vs.79.9%). N o significant interaction between Test Time and Stimulus Type was found [ F ( 1, 57) = .51, p = .475 ] Taken together, these results suggest ed that immediately after training (Post test), significantly and equally im proved. Moreover, their performance was better for the low variability stimulus type than for the high variability stimulus type. Pre T est, Post T est and Retention T est Comparison Two weeks after training, participants were administered a Retention test th at was identical to the Post test. Recall that the post test was also identical to the pre test. The administration of the retention test allowed for an investigation of the effectiveness of retention of the tone discrimination skills they had gained during the training, Pre test, Post test and Retention test scores for the two experimental groups (the Prototype and the High Variability groups) were compared. Table 4 3, Table 4 4, Figure 4 3 a nd Figure 4 4 show mean percent correct identification of the two Thai tones, mid and low for the two

PAGE 60

60 experimental groups: Prototype and High V ariability, under two stimulus type conditions: low and high variability, at three test times: P re test Post te st and Retention test. As shown in the Table 4 3, Table 4 4, Figure 4 3, and Figur e 4 4 above, both g roup s showed an improve ment after training. I t was also evident that the Prototype g roup did better than the High Variability group in the P ost test Howev er, a s show n in the Retention test scores, two weeks after training, the High V performance was better than that of the P rototype group : 97.2% vs. 93.3% for the low variability stimuli and 94.0% vs. 88.0% for the high variability stimuli. The Retention test also showed that a identification was more a ccurate under the l ow variability stimulus type condition than the h igh variability stimulus type condition (P ro totype mean = 93.3% vs. 88.0%, High Variability mean = 97.2% vs. 94.0% and overall mean = 95.2% vs. 91.0%). A repeated measures ANOVA with Group (P rototyp e and High Variability ) as the between subjects factor, and Test Time (P re test, P ost test and Retention test ) and Stimulus Type Condition (low variability and h igh vari ability) as the within subjects factors revealed marginally significant main effect of Group [ F ( 1, 18) = 4.20, p = .055]. This result suggested that, when pooled across three test times and two stimulus types, the High Variability training group outperfor med the Prototype training group almost to a significant level ( mean= 78.8% vs. 75.7%). Moreover, the analysis yielded a significant main effect of Test Time [ means= 77.3%, 86.3%, and 93.1% for Pre Post and Retention tests respectively [ F ( 2, 18) = 21.76 p = .000] P ost hoc pair wise comparisons using the Bonferroni method suggested that the significant main effect of Test Time was large ly due to the improvement from P re test to Post test [p = .015], and

PAGE 61

61 from P re test to R etention test [p = .002], but no t from Post test to R etention test [p = 1.000] Furthermore, a main effect of Stimulus Type was found [ F ( 1, 18) = 10.93 p = .004] with the l ow variability stimuli responded to more accurately compared to the h igh variability stimuli (87.2 % vs. 83.9%). N o significant interaction was found between Stimulus Type and Group [ F ( 1, 18) = .53 p = .474] nor between Test Time and Group [ F ( 2, 17) = 2.04 p = .159] nor between Stimulus Type and Test Time [ F ( 2, 17) = 1.73, p = .206]. These results indicated that the training significantly improved the performance of participants in both experimental groups. The improvement was evident immediately after training and was retained two weeks after training. Interestingly, no additional amount of improvement was obser ved during the two weeks period between the Post test and the Retention test. More importantly, overall, the High Variability training group performed almost significantly better than the Prototype group. Effects of Training in the Generalization Test Besi des the Post test and the Retention test, the effectiveness of the training was also examined in the Generalization test administered immediately and two weeks after training. In this test, the effectiveness of the training on (1) new stimuli, and (2) new speakers was explored. It was hypothesized that the training affects to both new stimuli and new speakers. Generalization Test 1 and Generalization Test 2 Comparisons Generalization tests 1 and 2 were administered to examine if the effects of training gen eralized to new stimuli and new talkers. The two generalization tests were identical, except that Generalization test 1 (Gen1) was administered immediately after training and Generalization test 2 (Gen2) was administered two weeks after training. The

PAGE 62

62 Gener alization tests consisted of four sets of stimuli: (1) old words produced by old tal kers (2) new words produced by old tal kers (3) old words produced by new tal kers and (4) new words produced by new tal kers Mean percent correct identification scores fr om Gen 1 and Gen2 by the Prototype and the High Variability training groups are reported in Table 4 5 through Table 4 8, and Figure 4 5 This data was submitted to a repe ated measures ANOVA with Group (Prototype, and High Variability ) as the between subje c ts factor, and Test Time (Gen1, and Gen 2), Stimulus Type (l ow va riability and high variability), Talker (old and new), and Word (old and new) as the within subjects factors The analysis yielde d a significant main effect of T alker [ F ( 1, 18) = 7.27 p = .0 15] with performance being significantly better for stimuli produced by new talkers than by old talkers (mean= 85.0% vs. 82.9%); a main effect of Stimulus Type [ F ( 1, 18) = 85.30% p = .000] confirming that significantly higher scores were obtained for low variability stimuli than for high variability stimuli (mean= 92.3% vs. 75.6% ) ; a main effect of Word [ F ( 1, 18) = 27.7 6 p = .000] with mean score for new words significantly higher than old words (mean = 85.7% vs. 82.3%); and a marginally significant mai n effect of Test Time [ F ( 1, 18) = 3. 99 p = .061] with mean score of Gen1 higher than that of Gen 2 (mean score = 83.3% vs. 79.2%). However, no significant main effect of Group was found [ F ( 1, 18) = 2.13 p = .161] even though the High Variability train ing group obtained higher score than the Prototype group (mean = 87.5 % vs. 80.5%) overall. These results suggested that the effectiveness of the training generalized to new talkers. However, the generalizability of the training effects appeared to decline over time (higher mean score for Gen1 than Gen2). High variability stimuli were more difficult

PAGE 63

63 than low variability stimuli. More interestingly, albeit statistically non significant, the high variability training paradigm appeared to result in a higher deg ree of generalizability than the prototype training method. The analysis also yielded several 2 and 3 way significant interaction s among factors. Significant interactions between Stimulus Type and Test Time [ F ( 1, 1 8) = 11.3 0 p = .003 ] Stimulus Type a nd Talker [ F ( 1, 1 8) = 8.59, p = .009 ] Test Time and Word [ F ( 1, 1 8) = 5.28, p = .034 ] Talker and Word [ F ( 1, 1 8) = 7.68, p = .013 ] Stimulus Type and Word [ F ( 1, 1 8) = 10.09, p = .005 ] and Stimulus Type, Talker, and Word [ F ( 1, 18) = 34.62 p = .000], were found However, no significant interactions between S timulus Type, Test Time, and Group [ F ( 1, 18) = .00, p = 1.000], S timulus Type, Talker, and Group [ F ( 1, 18) = .34, p = .564], Test Time, Word, and Group [ F ( 1, 1 8) = .21, p = .647 ] Talker, Word, and Group [ F ( 1, 1 8) = .75, p = .398 ] S timulus Type, Word, and Group [ F ( 1, 1 8) = 2.72, p = .116 ] Further analyses using one way ANOVAs and T tests were conducted to explore the significant interactions. Multiple comparisons were corrected using the Bonf erroni method with the significant alpha level set at .05. The significant three way interaction was explored first followed by 2 way interactions. Multiple t tests conducted to explore the significant three way interaction between Stimulus Type, Talker, a nd Word. First, the difference between the two stimulus types was compared. The analysis revealed variability stimuli was always significantly better than on the high var iability stimuli; old words produced by old talkers [Mean= 89.3 vs. 74.5, t (19) = 6.07, p = .000]; new words produced by old talkers [Mean=91.7 vs. 76.2, t (19) = 8.22, p = .000]; old words produced by new talkers [Mean = 93.9 vs. 71.4,

PAGE 64

64 t (19) = 10.54, p = .000; and new words produced by new talker [Mean = 94.3 vs. 80.5, t (19) = 10.82, p = .000]. When the difference between old and new talkers for each stimulus type (low and high variability) and for each word type (old and new) were compared, it was found that the difference was significant for l ow variability, old words [mean = 89.3 % vs. 93.9%, t (19) = 4.74 p = .000]; low variability, new words [mean = 91.7% vs.94.3%, t (19 ) = 2.74, p = .013]; high variability, new w ord s [mean= 76.2% vs.80.5%, t (1 9) = 3.78, p = .001], but not for high vari ability, old words [mean = 74.5% vs. 71.4%, t ( 19) = 2.29, p = .033]. These results, thus, suggested that overall participants performed significantly better on stimuli produced by new talkers than by old talkers. The only exception was found for high vari ability, old words; their performance was better when these stimuli were produced by old talker, but the difference was not statistically significant. T test analyses comparing the difference between the two word types (old and new) for each stimulus type (low and high variability) and for each talker (old and new), showed that participants performed significantly better on new words produced by new talker under the high variability stimulus condition (mean = 80.5% vs. 71.4% ) [ t (1 9) = 5.89, p = .000 ], but not by old talker under the low variability condition (mean = 91.7% vs. 89.3%) [ t (19) = 2.40, p = .027], or old talker under the high variability condition (mean = 76.2% vs. 74.5%) [ t (19) = 1.44, p = .164 ] or new talker under the low variability co ndition (mean= 94.3 vs. 93.9) [ t (19) = .79, p = .437 ] In other words, words, but only when the new words were produced by new talkers under the high variability stimulus cond ition such as conversational style

PAGE 65

65 A significant interaction between Stimulus Type and Test Time was further examined in one stimulus types for both Gen1 and Gen2. The analysis revealed that par ticipants performed significantly better on low variability stimuli than on high variability stimuli immediately after training (Gen 1) [mean =92.5% vs. 74.1%, F ( 1, 39) = 88.68, p = .000] and also two weeks after training (Gen2) [mean= 92.1% vs. 77.1%, F ( 1, 39) = 15.16, p = .000]. The difference that was large in Gen1 than Gen2 was the source of this interaction. Turning to the significant interaction between Test Time and Word, one way omparable for both old [mean = 84.3% vs. 87.0%), F ( 1, 39) = .00, p = .994] and new words [mean = 84.3% vs. 87.0%), F ( 1, 39) = .53, p = .470]. However, performance on new words was significantly better than on old words for Gen2 [ mean = 82.3% vs. 87.0%], t (19) = 8.02, p = .000], but not for Gen1 [mean=82.3% vs. 84.3%), t (19) = 1.90, p = .072]. Thus, the Test Time x Word significant interaction resulted mainly from the fact that, two ficantly better than their performance on old words. In sum, analyses performed on the Gen1 and Gen2 tests which were designed to test the generalizability of the training effect revealed several interesting results. First, results obtained from the main r epeated measures ANOVA analysis revealed that the effectiveness of the training generalized to new talkers. In addition, the generalizability of the training effects was, overall, more apparent immediately after training and seemed to have declined two wee ks after training (higher score on Gen1 than onGen2).

PAGE 66

66 Also, participants in the High Variability training group showed higher, but not statistically significant, degree of generalizability than those in the Prototype training group. Furthermore, participan ts found low variability stimuli easier overall. Second, analyses conducted to examine interaction effects further revealed that a) participants preferred low variability stimuli over high variability stimuli regardless of whether they were old or new stim new words was superior to their performance on old words, when these new words were produced by new talkers under the high variability stimulus condition, c) participants performed signif icantly better on low variability stimuli than on high variability stimuli two weeks after training (Gen2), but not immediately after training significantly better than on old words. General Summary of Findings accuracy data obtained from the Pre Post Retention and Generalization tests were reported. Results of the Pre test showed that, as e xpected, native Thai speakers were superior to all three groups of non native speakers, and that all groups performed significantly better on low variability stimuli produced in citation forms than on fast, conversational, high variability stimuli. After t raining, both training groups (Prototype and High Variability), and, surprisingly, the NE control group showed a significant and comparable amount of improvement (there was no significantly main effect of Group). The effectiveness of the training remained significant two weeks after training for both training groups. More interestingly, when all three tests (Pre Post and Retention) were considered, the High Variability training group outperformed the Prototype group

PAGE 67

67 at a near significant level. The effec tiveness of the training was also found to extend to new words, particularly those produced by new talkers in conversational, high variability style. Interestingly, the High Variability training group exhibited a higher degree of generalizability, overall (higher mean scores for Gen1 and Gen2) than the Prototype group, but the difference did not reach significance.

PAGE 68

68 Table 4 1. Mean percent correct in mid and low tones identification in the P re test comparing between two stimulus types for each group Group Prototype High V ariability NE C ontrol Thai Stimulus t ype Low vari High vari Low vari High v ari Low vari High vari Low vari High vari Mean 77.9 73.5 80 .0 77.7 82.8 77.4 99.6 99.6 SD 14.5 15.5 15.0 14.8 16.5 17.2 0.4 0.5 SE 3.2 3.4 3.3 3.3 3.7 3.8 0.1 0 .1 Table 4 2. Comparison of mean percent correct and SE between Pre test and P ost test for low and high variability stimulus type conditions Prototype g roup High V ariability g roup NE C ontrol g roup Low vari High vari Low vari High vari Low vari High vari Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post Mean 77.9 89.0 73.5 86.3 80 .0 85.8 77.7 84.0 82.8 86.1 77.4 81 .0 SD 14.5 14.0 15.5 12.7 15.0 15.0 14.8 15.1 16.5 18.2 17.2 16.8 SE 3.2 3.1 3.4 2.8 3.3 3.3 3.3 3.4 3.7 4.0 3.8 3.7 Table 4 3. Mean percent correct tone identification under low variability stimulus type condition Prototype High V ariability Pre test Post test Retention Pre test Post test Retention Mean 77.9 89.0 93.3 80.0 85.8 97.2 SD 14.5 14.0 9.6 15.0 15.0 2.4 SE 3.2 3.1 3.0 3.36 3.3 0.7 Table 4 4. Mean percent correct tone identificatio n under high variability stimulus type condition Prototype High V ariability Pre test Post test Retention Pre test Post test Retention Mean 73.5 86.3 88.0 77.7 84.0 94.0 SD 15.5 12.7 12.5 14.8 15.1 4.1 SE 3.4 2.8 3.9 3.3 3.4 1.3 Table 4 5. Mean percent corr ect tone identification for P rototype group in the G ene ralization test 1 and 2 under low variability stimulus type condition Prototype g roup l ow vari condition Gen1 Gen2 Old t alker New Talker Old t alker New Talker OW NW OW NW OW NW OW NW Mean 87 .0 87.7 90.7 92.2 85.5 90.2 89.7 89.7 SD 16.0 18.7 15.3 14.0 19.5 14.8 18.1 17.1 SE 5.0 5.9 4.8 4.4 6.1 4.7 5.7 5.4

PAGE 69

69 Table 4 6. Mean percent corr ect tone identification for P rototype group in the Generalization test 1 and 2 under high variability stimulus type condition Prototype Group h igh vari condition Gen1 Gen2 Old t alker New Talker Old t alker New Talker OW NW OW NW OW NW OW NW Mean 69.8 69 .0 70.1 73 .1 72.5 75 .0 67.8 77.9 SD 17.5 11.2 11.3 15.7 13.6 14.4 14.1 17.4 SE 5.5 3.5 3.5 4.9 4.3 4.5 4.4 5.5 Table 4 7. Mean percent corr ect tone identification for High V ariability group in the Generalization test 1 and 2 under low variability stimulus type condition H igh Variability Group l ow vari condition Gen1 Gen2 Old t alker New Talker Old t alker New Talker OW NW OW NW OW NW OW NW Mean 93.7 93.7 97.5 97.5 91.2 95.2 97.7 97.7 SD 5.0 9.8 2.6 3.1 5.1 5.9 2.7 4.6 SE 1.5 3.1 0.8 0.9 1.6 1.8 0.8 1.4 Table 4 8 Mean percent corr ect tone identification for High V ariability group in the Generalization test 1 and 2 under high variability stimulus type condition High Variability Group high vari condition Gen1 Gen2 Old t alker New Talker Old t alker New Talker OW NW OW NW OW NW OW NW Mean 75.8 77.5 73.8 84.0 79.8 83.3 74.0 87.0 SD 12.5 8.7 10.4 7.2 7 .0 8.0 10.1 7.7 SE 3.9 2.7 3.3 2.2 2.2 2.5 3.2 2.4 Table 4 9 Means percent identification accuracy for factors that the main effect is significant collapse d over other variables Talker Stims Word Test Time O ld New Low High Old New Gen test1 Gen test2 Mean 82.9 85.0 92.3 75.6 82.3 85.7 83.3 79.2

PAGE 70

70 Table 4 10. S ummary of significant results of the generalization test s Factors tested F value P v alue Talker (1, 18) = 7.27 P = .015 Stimulus T ype (1, 18) = 85.30 P = .000 Word (1, 18) = 27.75 P = .000 Test T ime (1, 18) = 3.99 P = .061 (marginally sig.) Stimulus Type x Test T ime (1, 18) = 11.30 P= .003 Stimulus T ype x T alker (1, 18) = 8.59 P= .009 T est T ime x Word (1, 18) = 5.28 P= .034 Talker x Word (1, 18) = 7.68 P= .013 Stimulus T ype x Word (1, 18) = 10.09 P= .005 Stimulus T ype x Talker x Word (1, 18) = 34.62 P= .000

PAGE 71

71 Figure 4 1. Comparison of mean percent correct and SE in the Pre test b etween two stimulus types for each group Figure 4 2. Comparison of mean percent correct and SE between the Pre test and the P ost test under two stimulus type conditions

PAGE 72

72 Figure 4 3 Mean percent correct tone identification and SE between the Pre Post Retention test under low variability stimulus type condition Figure 4 4. Mean percent correct tone identification and SE between the Pre Post Retention test under high variability stimulus type condition

PAGE 73

73 Figure 4 5. Mean percent corr ect tone ident ification and SE in the Generalization test 1 and 2 under low variability stimulus type condition Figure 4 6 Mean percentage corr ect in tone identification and SE in the Generalization test 1 and 2 under high variability stimulus type condition

PAGE 74

74 CHAPT ER 5 REACTION TIME reaction time to correct responses were also collected and analyzed and the results will be reported in the same format as the mean percent correct identifi cation. First of all, the results of reaction time to correct responses only were compared among four groups: Prototype, High Variability, NE control and Thai groups for Pre test and among the three groups for Post test: Prototype, High Variability, and NE C ontrol. The effects of two stimulus types were also examined: low variability stimuli (produced in citation forms by two speakers, a male and a female speaker) and stimuli with high degree of variability (produced in context by two male and two female sp eakers). Next, the training Pre test with the Post test among the three groups: Prot otype, High Variability and NE C ontrol. Moreover, reaction time data from the General ization tests will be reported comparing between the two groups: Prototype and High Variability. In the last section, reaction time data from the Pre Post and Retention tests and from the Generalization Test 1 and 2 will be reported. Pre T est Reaction time s (in milliseconds) in identification of the two Thai tones, mid and low, for the four groups: Prototype(n =20), High Variability (n=20), C ontrol (n=20), and Thai groups(n=10), under two stimulus type conditions: low variability and high variability obtaine d during the pre test, are reported in Table 5 1 and Figure 5 1. R eaction time for the native Thai speakers and the NE control groups were faster than those of the Prototype and the High Variability training groups. However, for all groups,

PAGE 75

75 reaction time f or high variability stimuli was longer th an for low variability stimuli (mean = 675 msec. and 642 msec.). A repeated measures ANOVA with Group (Prototype, High Variability, NE control, and Thai) as the between subjects factor, and Stimulus Type (low and hi gh variability) as the within subjects factor was performed on this data. The analysis yielded a non significant main effect of Group (group mean averaged across all other variables: Prototype = 690 msec High Variability = 710 msec., NE control= 615 mse c., Thai = 620 msec.), F (3, 66) = 1.43, p = .240], but a significant main effect of Stimulus Type (mean for each stimulus averaged across all other variables: low variability stimulus type = 642.5 msec., and high variability = 675 msec.) F (3, 66) = 4.33, p = .041]. However, no interaction between Group and Stimulus Type [ F (3, 66) = .05, p = .984] was found. These results suggested that, on the average, participants took longer to decide whether a mid or a low tone was heard when listening to high variabi lity stimuli. Effects of Training Effects o f training on reaction times were also examined by comparing test, Post test, Generalization test, and Retention test Pre test was compared with Post test and Retenti on test to evaluate the effect of training on the same words produced by the same speakers Gen 1 and Gen 2 tests were compared to assess if the effect of training is generalized to (1) new stimuli produced by new speakers, (2) new stimuli produced by old speakers, and (3) old stimuli produced by new speakers as well as to investigate retention of the effects of the training.

PAGE 76

76 Pre T est to Post T est Comparison Reaction Times for the three Groups: Prototype (n=20), High Variability (n=20), and NE control grou ps (n=20), under two stimulus type conditions: low variability and high variability are reported in Table 5 2 and shown in Figure 5 2. Reaction time became faster after training for the Prototype and the NE control groups. The opposite, however, was true f or the High Variability training group. Concerning stimulus type, all variability stimuli than for high variability stimuli. T his data was submitted to a repeated measure s ANOVA with Group (Prototype, High Variab ility, and NE c ontrol ) as the between subjects factor, and Stimulus Type (low variability and high varia bility) and Test Time (Pre and P ost test ) as the within subjects factor. The analysis yielded a marginally significant main effect of Group [ F ( 2, 57) = 3.01, p =.057], a significant effect of Stimulus Type [ F ( 1, 57) = 22.32, p = .000], but a non significant main effect of Test Time [ F ( 1, 57) = .80, p =.374]. These results indicated that, on the average, reaction times for all three groups were compara ble (mean for all three groups collapsed over both test times and both types of stimuli: Prototype = 670 msec., High Variability = 732 msec., control = 587 msec.), t hat they responded more quickly (640 msec. vs. 686 msec.) to low variability stimuli than t o high variability stimuli, and that their response time did not improve significantly after training (mean for Pre test = 673 msec. and Post test = 653 msec.). Pre T est, Post T est and Retention T est Comparison s Recall that the Post test was identical to t he Pre test. The administration of the Retention test allowed for an investigation of the effectiveness of the training on reaction time two weeks after training. To test for this effect, reaction times for Pre test,

PAGE 77

77 Post test and Retention test for the tw o experimental groups (the Prototype and the High Variability groups) were compared. Table 5 3, Table 5 4, Figure 5 3 and Figure 5 4 show reaction time for the two experimental groups from the three tests: Pre test, Post test and Retention test, under the low variability and the high variability stimulus conditions. variability stimulus type condition than the high variability stimulus type condition (mean= 590 msec. vs. 680 msec.) for the Protot ype group, and 540 msec. vs. 630 msec. for the High immediately after training (Post test) while that of the High Variability group became longer. This was true for both ty pes of stimuli (Prototype group mean = 620 msec. and 680 msec. and High Variability group mean = 720 msec. and 780 msec. for low and high variability stimuli type, respectively). However, two weeks after training (Retention test), the opposite was true: r eaction time for the High Variability group was faster than that of the Prototype group for both types of stimuli (mean = 540 msec. vs. 590 msec. and 630 msec. vs. 680 msec. for low and high variability stimuli respectively). However, a repeated measures ANOVA with Group (Prototyp e and High Variability ) as the between subject s factor, and Test Time (Pre test, Post test and Retention test), and Stimulus Type (low variability and h igh variability ) as the within subjects factor revealed a significant main ef fect of only Stimulus Type [ F (1, 18) = 16.26, p = .001], but not of Group [ F (1, 18) = .00, p = .969], or Test time [ F (2, 36) = .50, p = .608]. Furthermore, n o significant interaction was found between Stimulus Type and Group [ F ( 1, 18) = .25, p = .826 ] nor betwe en Test Time and Group [ F ( 2, 36) =

PAGE 78

78 1.09, p = .345] nor between Stimulus Type and Test Time [ F ( 2, 36) = 1.26, p = .295]. These results suggested response time, neither immediately or t wo weeks after training, and that the high reaction time than the prototype method. Effects of Training in the Generalization Test Besides the Post test and the R etention test, the effectiveness of the training on reaction time was also examined in the Generalization tests administered immediately and two weeks after training. In these tests, the effectiveness of the training on reaction time on (1) new stimuli, an d (2) new speakers was explored. It was anticipated that the training also affects on reaction time through both new stimuli and speakers. Generalization Test 1 and Generalization Test 2 Comparison s Effect of training in generalization to new stimuli was examined using two Generalization tests: Gen1 and Gen2. The Gen2 was identical to the Gen1 but it was administered two weeks after training. The Generalization tests consisted of four sets of stimuli: (1) old words produced by old talkers, (2) new words p roduced by old talkers, (3) old words produced by new talkers, and (4) new words produced by new talkers. Reaction time of the Gen1 and Gen2 by the Prototype and the High Variability groups are reported in Table 5 5 to Table 5 8, Figure 5 5, and Figure 5 6 This data was submitted to a repeated measure s ANOVA w ith Group (Prototype, and High V ariability) as the between subject s factor, an d Test Time (Gen1, and Gen2), Stimulus Type (l ow variability and h igh variability), Talker (Old and New), and Word (Old an d New ) as the within subjects factor. The analysis indicated that participants responded significantly more quickly to low variability stimuli than to high variability

PAGE 79

79 stimuli (596 msec. vs. 829 msec.), [ main effect of Stimulus Type, F ( 1, 18) = 77.72 p = .000] More interestingly, their reaction time became significantly faster two weeks after training (mean = 781 msec. v s. 644 msec.), [main effect of Test Time, F ( 1, 18) = 12.00 p = .003] However, no significant mai n effect of Group (711 msec. vs. 714 m sec. ) [ F ( 1, 18) = .25, p = .618], Talker (709 vs. 716 ) [ F ( 1, 18) = .21 p = .646], and Word ( 710 msec vs. 716 msec ) [ F ( 1, 18) = 2.82 p = .110] was found suggesting that ount of time to react to old and new talkers and to old and new words. Interestingly there were significant interactions between Stimulus Type and Word [ F ( 1, 18) = 5.80 p = .027], and Talker and Word [ F ( 1, 18) = 7.39 p = .014] The significant interac tion between Stimulus Type and Word was further examined by t test s comparing the difference between the two W ord types ( o ld a nd new) for each Stimulus Type c ondition ( l ow and h igh var iability). T he analysis revealed that the difference between the o ld and n ew w ord s was not significant unde r the low variability stimulus type c ondition (584 msec. vs. 579 msec. ) [ t (19) = 57 p = .573] but it was significant under the h igh variability o ne (809 vs. 836 ) [ t (19) = 2,76 p = .012] In other words, participants responded to old words significantly faster than to new words, but only when these words were produced in the high variability stimulus condition. The significant interaction between Talke r and Word was further examined by comparing the difference between the two Word (old and New) for each Talker (o ld and New), the analysis revealed that the difference between the Old and New Word was not significant for the Old Talker (702 msec. vs. 694 msec. ) [ t (19) = .77 p = .450] was significantly different for the New Talker (691 msec. vs. 720 msec. ) [ t (19) =

PAGE 80

80 3.34 p = .003] new words when they were produced by new talkers. Summary of Findings Statistical analyses performed on reaction time data revealed that, before training, all four groups of participants (Prototype, High Variability, NE control and Thai) took comparable amount of time to identify the low level and mid falling tones in Thai. However, they were significantly faster at iden tifying low variability stimuli than high variability stimuli. Interestingly, neither training method was found to significantly shorten the response time of any group or for any type (low and high variability) o f ( old) stimuli, either immediately or two weeks after training. Furthermore, even though there seemed to be an overall improvement in reaction time from immediately to two weeks after training on novel stimuli (shorter mean RT for Gen 1 than Gen 2), the training effects did not appear to generaliz e to new words or to new talker as it was found that reaction time for old words was significantly faster than for new words produced under high variability condition, and that reaction time for new words produced by new talker was significantly longer tha n by old talker. Summary of Accuracy and Reaction Time Data When accuracy and reaction time data were considered together, the combined results were as follows. Firstly, results of the Pre test showed that all groups of participants were significantly more accurate and faster at identifying low variability than high variability stimuli. The Pre test results also showed that native Thai speakers were significantly more accurate, but not significantly faster than the non native groups in identifying the two T hai tones. Secondly, immediately and two weeks after training, both training methods significantly improved tone identification accuracy of both training

PAGE 81

81 groups to a comparable degree. Their reaction times were also significantly shortened. When the Pre Post and Retention tests were considered together, it was found that the High Variability training group almost significantly outperformed the Prototype training group. A careful examination of Figure 4 3 and Figure 4 4 suggested that this result was like ly due to the fact that the High Variability training group performed much better at the Retention test than the Prototype group. Interestingly, their response time remained statistically comparable to the Prototype group suggesting that they have not sacr ificed processing time for accuracy or vice versa. Thirdly, the effectiveness of both methods of training seemed to generalize to novel stimuli, particularly high variability new words produced by new talkers at two weeks after training. That is, the analy ses suggested that participants were more accurate at identifying the two Thai tones when they were new words produced in fast, conversational style by new talkers. Reaction time data suggested, however, that this effect may have come at the expense of a l onger reaction time as it was found that reaction time for new words produced by new talker were significantly longer than for old words. These above results will be discussed in Chapter 6 along with tone language pedagogical implications, limitations of t he study and future research.

PAGE 82

82 Table 5 1. Reaction time in m id and low tones identification in the pre test comparing between two stimulus type conditions for each group Group Prototype High V ariability NE C ontrol Thai Stimulus t ype Low vari High vari Lo w vari High v ari Low vari High vari Low vari High vari Reaction t ime 670 710 700 720 600 630 600 640 SD 190 210 210 210 160 150 100 120 SE 40 40 40 40 40 40 60 60 Table 5 2. Comparison of reaction time and SE between Pre test and P ost test for low and high variability stimulus type conditions Prototype g roup High V ariability g roup NE C ontrol g roup Low vari High vari Low vari High vari Low vari High vari Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post RT 670 620 710 680 710 720 720 780 60 0 520 630 600 SD 190 210 210 210 210 340 210 230 150 520 150 600 SE 40 60 40 50 40 60 40 50 40 60 40 50 Table 5 3. Reaction time in tone identification comparing between the Pre Post Retention test under low variability stimulus type condition Protot ype High V ariability Pre test Post test Retention Pre test Post test Retention RT 670 620 590 710 720 540 SD 190 210 160 210 340 170 SE 40 60 50 40 60 50 Table 5 4. Reaction time in tone identification comparing between the Pre Post Retention test under high variability stimulus type condition Prototype High V ariability Pre test Post test Retention Pre test Post test Retention RT 710 680 680 720 780 630 SD 210 210 150 210 230 170 SE 40 50 50 40 50 50 Table 5 5. Reaction time in tone identif ication for Prototype group in the Generalization test 1 and 2 under low variability stimulus type c ondition Prototype group Low vari condition Gen1 Gen2 Old Talker New t alker Old Talker New t alker OW NW OW NW OW NW OW NW Mean 620 620 620 600 560 560 600 580 SD 300 290 20 200 160 140 170 150 SE 70 70 50 40 50 40 50 50

PAGE 83

83 Table 5 6. Reaction time in tone identification for Prototype group in the Generalization test 1 and 2 under high variability stimulus type c ondition Prototype group High var i condition Gen1 Gen2 Old Talker New t alker Old Talker New t alker OW NW OW NW OW NW OW NW Mean 880 890 850 890 800 780 740 800 SD 270 290 290 300 210 220 230 240 SE 60 70 60 70 60 70 70 80 Table 5 7. Reaction time in tone identification for Hig h Variability group in the Generalization te st 1 and 2 under low variability stimulus type condition High V ariability group Low vari condition Gen1 Gen2 Old Talker New t alker Old Talker New t alker OW NW OW NW OW NW OW NW Mean 730 690 720 720 500 470 480 480 SD 270 260 320 270 100 150 130 130 SE 60 50 70 60 30 40 40 40 Table 5 8 Reaction time in tone identification for High V ariability group in the Generalization test 1 and 2 under high variability stimulus type c ondition High V ariability gr oup High vari condition Gen1 Gen2 Old t alker New t alker Old t alker New t alker OW NW OW NW OW NW OW NW Mean 880 910 910 980 730 730 740 760 SD 230 240 290 280 200 220 230 220 SE 50 50 70 60 60 70 70 70 Table 5 9 Reaction times means collapsed over other variables Talker Stims Word Test Time Group O ld Ne w Low High Old New Gen test1 Gen test2 Prototype High Variability Mean 709 716 596 829 710 716 781 644 711 714

PAGE 84

84 Figure 5 1. Reaction time in mid and low tones identification in the Pre test comparing between the two stimulus type conditions and SE for each group Figure 5 2. Comparison of reaction time and SE bet ween Pre test and Post test under two stimulus type conditions

PAGE 85

85 Figure 5 3. Reaction time in tone identification and SE c omparing between the Pre Post Retention test under low variability stimulus type condition Figure 5 4. Reaction time in tone identification and SE comparing between the Pre Post Retention test under high variability stimulus type condition

PAGE 86

86 Figure 5 5 Reaction time in tone identification and SE i n the Generalization test 1 and 2 under low variability stimulus type c ondition Figure 5 6. Reaction time in tone identification and SE in the Generalization test 1 and 2 under high variability s timulus type c ondition

PAGE 87

87 CHAPTER 6 DISCUSSION AND CONCL USION Relative efficacy of two methods used to train native speakers of English to identify the mid and the low tones in Thai was investigated. The two training methods differed mainly in the types of stimuli involved; the Prototype training paradigm used low variability stimuli produced in isolation in a clear, citation style while the High Variability training method employed stimuli produced in a sentence final context with a relatively faster, conversationa l style. The effectiveness of the training was measured as well as their reaction time (RT). Training effects on no vel stimuli were also examined. Pre Test The Pre tes t results were compared among four groups: Prototype, High Variability, NE control, and Thai. F irst as predicted, native Thai speakers were significantly more accurate at identifying the two Thai tones than all three groups of American English speakers un der both stimulus type conditions: low variability and high variability Secon d, all three groups of non native speakers performed significantly better under the low variability stimulus set. T hese results showed that the American groups did worse than the Thai speakers did, most likely because they had not had any experience with tones. As discussed in Chapter 1, non native speakers of a tone language have difficulty when they learn tones primarily because of their lack of prior experience with the target language ( Piske, 2001 ). RT data was consistent with the identification accuracy data and showed that RT was longer for high variability stimuli than for low variability s timuli for all groups of participants These results suggested that additional variab ility in the stimuli resulted in

PAGE 88

88 an increase in processing load among both native and non native speakers. The fact that native Thai speakers were not significantly faster than non native speakers was somewhat surprising. No firm explanation can be provide d at present. However, a relatively high identification accuracy (above 70%, chance = 50%) among non native speakers suggested that the stimuli were easy for them to discriminate, which could explain why they responded as quickly as the native Thai speaker s. Pre Test and Post Test Turning to the effect of training, compari ng the Pre test and Post test performance among the three groups of American English listeners and the two stimulus types, the result s show ed that the training significantly improved the Thai tones, but the difference in improvement across the three groups was not significant. Interestin g ly, the increase in ability to discriminate the tones did not lead to an accompanying decrease in response time. Participants still thought about their responses for the same amount of time; they just achieved more accurate responses in the end. However, there was marginal evidence that the Prototype group reacted faster than the High Variability group, indicating that the perfor mance of participants exposed to high variability stimuli improved with training, but this may come at the cost of a higher processing load in comparison to low variability stimuli. Overall, the pre post training results suggested that training with simple laboratory techniques can improve or modify the adult human perceptual system that categorizes speech sounds in a short period of time not only for consonant and vowel perceptions, as previous studies explored (Pisoni, Aslin, Perey, & Hennessy, 1982) bu t also for lexical tones. This evidence indicates that the human perceptual system is malleable and can be changed

PAGE 89

89 It should be pointed out that the NE c ontro l group did not undergo training, yet they performe d significantly better on the Post test than on the Pre test. However, unlike participants in the training groups, half of the participants in the NE c ontrol group were l inguistics major students. Therefore, it is possible that their speech sound discrimin ation ability was superior to non linguistics major students due either to their previous linguistic training or their linguistic inclination or propensity. Nonetheless, this result might lead one to think that simply exposing participants to target stimul i, with or without feedback, could lead to perception improvement. However, the fact that perceptual gain realized by participants in the NE control group was numerically lower tha n that of the two training groups suggested the additive benefit of the trai ning or the provision of feedback. The statistically non significant group difference observed was likely due to the duration of the training. In comparison to other studies, training duration employed in this study was relatively short. For example, the t raining lasted two weeks identification was observed. In this study, participants only underwent three days of training and perceptual gains of 11.99 % and 6.05 % for the Protot ype and the High Variability training groups, respectively, were recorded. A longer training period, such as two weeks or a month, may have yielded stronger and more obvious evidence of the training efficacy within and across groups. Pre T est, Post T est an d Retention T est To investigate the longer term effects of the training, a Retention test that was identical to the Pre test and to the Post test was administered two weeks after training. When all 3 tests are considered, w hat i s interesting is that the di fference between the two training groups reversed. Recall that the Prototype training group numerically

PAGE 90

90 outperformed the High Variability training group on the Post test, which was administered immediately after training, but the High Variability training group numerically outperformed the Prototype training group on the Retention test two weeks later. In other words, perceptual gains realized among participants in the Prototype group were greater than those in the High Variability training groups immediat ely after training, but the opposite was true two weeks after training, leading to a nearly significant (p=.055) difference between the two groups. These results strongly suggested that long term representations of Thai tones formed under the high variabil ity training paradigm were more robust than those formed under the low variability (prototype) training method. The fact that this result was found for both low and high variability stimuli lends further support to this claim. Interestingly, it was also f ound that s ignificant improvement accuracy scores was found from P re to Post test and from P re test to the Retention test, but not from Post test to the Retention test This finding is interesting as it suggests that, similar to previous studies (Bradlow et al. 1999) the effectiveness of the training was retained at least two weeks after training. In addition, the fact that the High Variability group improved more between the Post test and the Retention test than they did from the Pre tes t to the Post test suggests that the non significant Post test to Retention test difference was driven mainly by the data from the Prototype group. That is, while perceptual gains seemed to dissipate two weeks after training for the Prototype group, percep tual sensitivity seemed to continue to develop for the High Variability group.

PAGE 91

91 Turning to reaction time, although the difference between the two training groups was not significant, a trend in the differential effects of training could be observed. Interes tingly, the Prototype training group reacted numerically faste r than the High Variability training group did for the P ost test whereas the High Variability training group reacted faster than the Prototype training group for the Retention test or two weeks after training in both stimulus t ype conditions. This result is consistent with mean percent correct identification mean percent correct identification was higher than those in the High Variability training group for the Post test, whereas the High Variability training group performed better than the Prototype training group on the Retention test two weeks after training. In short, results of the combined Pre Post and Retention tests suggested the superior long term ef fects of the High variability training paradigm. That is, contrary to the immediate, but quickly dissipating effects of the Prototype training paradigm, the High variability training group endured a slow start, but had longer lasting improvement. This find ing is consistent with findings of previous studies that performance after exposure to high variability stimuli declined at the beginning of training and then significantly improved later (Mullennix, 1989; Mullennix & Pisoni, 1990; Sommers et. al., 1994). Generalization Test 1 and Generalization Test 2 T he results obtained from the Generalization test s illustrated the effects of training to new words and new talkers. On both Generalization tests the partic ipants showed better performance on Generalization Test 1 than Generalization Test 2 under Low variability Stimulus Type condition, New Talker and New word T he better performa nce on New Word stimuli might suggest that the

PAGE 92

92 participants were better at ident ifying new words because they match ed the new words to the tonal representation such as pitch contour, built/formed on old word during training. This finding provides evidence that the participants relied on the tonal representations they built/formed, no t on each old word. This means that they were able to focus more on identifying the new stimuli without the conflict related to the old information of each old word I t may be easier for them to ident ify a new word than an old word, since for the new words they did not have the old information in their me mory to be perfectly matched which may have confused the participants. Concerning the result that the participants did better on New Talker than Old Talker stimuli, it is possible that the new talkers in th e study produced the stimuli more clearly or more intelligible than the old talkers did. Other studies have shown that female talkers are more intelligible than male talkers (Bradlow et. al., 1996). However, the stimuli in this study were produced by both genders to avoid this effect. In interviews immediately after training, participants stated that the new talkers were more intelligible than the old talkers, and this may account for the result. Turning to reaction time the participants reacted faster in Generalization test 2 than on Generalization test 1. This finding supports the hypothesis that they needed time to form tonal representations in their memory, as they were able to react more quickly two weeks after training. Although a difference between t he two experimental significant the High Variability training group showed numerically better performance than the Prototype training group did. This suggests that the High variabilit y training method may lead to a greater ability to make generalizations immediately afte r training. The ability to make

PAGE 93

93 generalizations is important to learning any phonetic aspect of any foreign language, particularly in learning Thai tones, given the eno rmous amount of tonal variability in a real wor l d environment The High Variability training group seems to obtain more benefits in generalization from the higher degree of variability in the stimulus set they were exposed to during the training This is c onsistent with previous studies stating the new speech sounds as well as to understand new speakers (Lively et al., 1993, 1994). T he initial resear ch questions will now be addressed once more to illustrate the results. 1 Which training method (prototype vs. high variability) will result in a higher percentage of perceptual accuracy immediately after training? 2 Which training method wi ll produce faster reaction time in perception immediately after training? 3 Which training method will lead to a greater ability to make generalizations immediately after training? 4 Which training method will be more effective (as measured by perceptual accuracy, reaction time, and ability to make generalizations) 2 weeks after training? For the first question, the results from all tests suggest that the Prototype training method did result in a higher percentage of perceptual accuracy immediately after training. Moreover, for the second question, the Prototype train ing group also reacted faster than the High Variability group at the Post test administered immediately after training. However, for the third question, the High variability training method led to a greater ability to make generalizations immediately after training. Most importantly, for the last question, the High variability training method helped participants retain the Thai tones in the longer term, as shown by the Retention test administered two weeks after

PAGE 94

94 training. However, these are numerical data o nly; the differences by training group were not statistically significant. In conclusion, this study investigated which method of teaching second language perception is more effective, the traditional Prototype approach or the more radical High Variability approach. Unlike previous studies that focused on consonants and vowels, this study explored the effect of training on tone perception. Moreover, in this reaction tim e and their ability to generalize to novel words and speakers. Tests were administered three times, before training, immediately after training and two weeks after training, the latter to investigate which training method is more effective for long term ga in. Importantly, the non High Variability training, particularly two weeks after training, in both mean percent correct identification and reaction time. The results of this study suggest that the High Variability training method has a longer and more enduring effect than the Prototype training method on tone perception for non native speakers of tone languages. The results support the Prototype theory in that providing learners with only ideal exem plar of the two Thai tones in citation forms does facilitate their learning of these tones immediately after training However, the study also illustrated that, for long term gain, the results are consistent with the exemplar based view. The entire system of the category shifted slightly with a new experience as seen from the result that the High Variability training method has a longer and more enduring effect than the Prototype training method.

PAGE 95

95 Applications to the Classroom The results of this study provi ded useful information on effective training procedures on tone perception. Therefore, teachers who teach a tonal language suc h as Thai, Chinese, Vietnamese, and Yoruba, to L2 learners can employ the High Variability training paradigm as a pedagogical tool in their teaching. Since long term gain is obviously more important than immediate but temporary improvement, teachers should employ the High Variability training paradigm. It would be more beneficial to get students to build a better long term representa tion of tones. Alternatively, the teachers could combine the low and high variability training procedures, which may prove a more effective teaching method in their classrooms. For example, they may start their teaching with the Prototype training method providing the Prototype tones by themselves. After that, they may assign students to do activities to gain exposure to a variety of tones in real life, for instance, going out to Thai restaurants or travelling to tourist attractions to perceive tones from different talkers, different speaking styles, and speaking rates. Having been trained by a combination of these teaching procedures, the students should have more robust mental representations of tones in their memory. Limitation s of the P resent S tudy In t his study, although the results showed that the participants performed better after training, some of their performances are numerically better but not statistically better. Therefore, further studies should use many more participants, more tokens of the s ame tones, more talkers, longer training, and a longer delay, perhaps of one to three months, before the Retention test. The most challenging task is designing an experiment to find out how much variability in each factor word, talker, stimulus type cond ition, for instance will provide optimal benefits in tone perception. Degree of

PAGE 96

96 variability should be considered very carefully to find a better solution. Moreover, to avoid a problem about the control group that might perform better than other participa nts due to their background; the participants should not be linguists or linguistics students although this may not always be possible. Future Directions Concerning the effectiveness of the training on generalization ability, the factors like Talker (New a nd Old) and Word (New and Old) should be expanded upon in further studies. For talker, it would be better if there were variations in speaking styles. For word, it would be better to employ high er variability stimuli produced by multiple talkers in more se ntential contexts, for example, the beginning, the middle, and the end of the sentences. However, a researcher needs to consider whether the stimuli can still be representative of a tone since a high degree of variability may cause excessive variation in a tone, and that tone may change and be realized as another tone. In addition, since this study did not test the generalization ability of the Control group, it is suggested that their generalization ability be further compar ed with the experimental groups to confirm the effectiveness of training. To investigate the effect of training on working memory and long term memory, the participants may be tested at multiple times, for instance, immediately after training and then two weeks, one month, two months, an d three months after training. Then the results obtained from each test time will be compared to find significant improvement and the optimal period of time that is sufficient for learners to store high variability information in their memory. In addition it is suggested that other Thai tones or all five Thai tones be investigated and the results be compared. Moreover, concerning participants, comparing the ability of non native speakers of tone languages like

PAGE 97

97 American English and native speakers of tone languages like Chinese to perceive Thai tones using high variability stimuli will generate better understanding of the effectiveness of training on tone perception for various listeners.

PAGE 98

98 APPENDIX STIMULI 1 20: Words in sentence final contexts of the Pre test and the Post test stimuli 1 [p b k ha j ju t pa ] Father told (Nid) to stop throwing. 2 [pu pen k on ra k pa ] My grandfather loves the forest. My grandfather is a forest lover. 3 [t r t a n la j pi ] She has been waiting for me for years. 4 [k a w p p ] He likes pipe. 5 [ja ha j r a k a k a ] 6 [m kamla plu k k a ] Mother is planting a galangal (plant). 7 [rawa a d n to k k u ] 8 [k a w ma j klua m wa a t u k k u ] He is not afraid even though he has been threatened. 9 [p u yi ma j k j p u t ku ] Koo 10 [sa t pa tako n ku ] Wild animals howl loudly.

PAGE 99

99 11 [k a w t u k t si p t ] He is/was hit/bang/beaten 10 times. 12 [k un k uan k i t ha j t u an t ] You should think carefully 13 [p m k uan ra k lu k ha j t u k t a ] Parents should show/demonstrate their affection to their children in the right way. 14 [k a w d n k a t a ] He walks with legs wide apart 15 [ a n p na kin k a w t p ] I like to have dinner on a raft. 16 [ja j pen kon af ap a p ] My grandmother is a generous woman. 17 [t la ba n mi k a si p k ] There are ten clum ps of galangal in the backyard. 18 [panha ni : k raj k ] Who cause this trouble? 19 [pu pu aj pen ro k taj ] Grandfather is sick with a kidney disease. 20 [kra r k tua ni ma j p ta j ] This s 21 40: Words in sentence final contexts of the training stimuli 21

PAGE 100

100 [k ha j pra tsa a k ro k p aj] May you stay clear from illness. 22 [k r a t a k sa n ni t am t a k p a j ] This handcraft is made of bamboo. 23 [t ma j k j mi ra k i k a w ] She has been virtuous. 24 [t a n t a wa pen ka n ku k a w ] I believe this is a rumor. 25 [lu ksa w k on ni p m wa pen t i p p a ] This daughter, the parents hope to rely on 26 [ni t klua fa p a ] Nid fears thunder. 27 [t t a t d kma j sa j p a n ] She arranged the flowers into the tray. 28 [t a n s p p a n] I passed the test. 29 [t a t a p baj j t a ru s k k a j] If you touch a sugar cane leaf, your hand will feel irritated/rashed. 30 [pla ti t k a j] Fish was trapped in the fishing net. 31 [t a n ra k pra t e t t aj ] I love Thailand.

PAGE 101

101 32 [de k t u k t a p paj ri ak k a t a j] A child was kidnapped (for ransom). 33 [ja t a j le p kaw] 34 [k a w ma j t p k a p ro t ka w ] 35 [m t am k wa m sa a t taw ] My mother cleaned the stove. 36 [de k ba k on ma j t p ta w] 37 [k wa m k i t t t i ap k om ] Her mind is so sharp. 38 [ma j mi k raj t p t u k k o m ] No one likes to be pressed. 39 [na kji p n kamla la n kaj ] A sharpshooter is pulling the trigger. 40 [p k rua kamla sa p ka j ] The cook is chopping chicken. 41 6 0: Words in sentence final contexts of the G eneralization stimuli (20tokens: 10 tokens of new words and 10 tokens from old words) 41 [lu ksa w k on n i p m wa pen t i p p a ] This daughter, the parents hope to rely on

PAGE 102

102 42 [ni t klua fa p a ] Nid fears thunder. 43 [t t a t d kma j sa j p a n ] She arranged the flowers into the tray. 44 [t a n s p p a n] I passed the test. 45 [m t am k wa m sa a t taw ] My mother cleaned the stove. 46 [de k ba k on ma j t p ta w] 47 [k wa m k i t t t i ap k om ] Her mind is so sharp. 48 [ma j mi k raj t p t u k k o m ] No one likes to be pressed. 49 [t a n ra k pra t e t t aj ] I love Thailand. 50 [de k t u k t a p paj ri ak k a t a j] A child was kidnapped ( for ransom). 51 [t a ma j ma n t aj k ja t a j 52 [de k t aro k kamla t a j ] Babies are defecating.

PAGE 103

103 53 [ri t l at ka p pu ] To get blood from a crab. (Englis 54 [p anba n k ra k pu ] The neighbors also love my grandfather 55 [t am k s p se t l w ja l m tru at t a n ] t to go over all your answers. 56 [fajt a j ru n ni ma j t t a j t a n ] This kind of flashlight does not require batteries. 57 [ma j t t i m t ] Real silk that my mother weaved 58 [k a w t a ma j k a m t ] He carries/holds a pole 59 [man ma j ki aw ka p k ] It is not your business 60 [la j k on b k wa ni t k ] Many people told Nid that she is old

PAGE 104

104 LIST OF REFERENCES Abramson, A.S. (1962). The Vowels and Tones of Standard Thai: Acoustical Measurements and Experiments. Columbia University. Abramson, A. S. (1979). The coarticulation of tones: An acoustic study of Thai. In T. L. Thongkum, P. Kullavanijaya, V. Panupong, & K. Tingsa badh (Eds.), Studies in Tai And Mon Khmer phonetics and phonology : Indigenous Languages of Thailand Research Project, 1 9. Aroonmanakun, V. (2006). Using Language Data Base for learning and Teaching Thai Paper presented at the Conference on the Teaching Thai as a Foreign Language, Bangkok, Thailand Arunrueng, A. (1990). Variation of the falling tone by age of speakers of Bangkok Thai MA thesis. Chulalongkorn University Bluhme, H.,&Burr, R. (1971). An audio visual display of pitch for teaching Chinese tones. Studies in Linguistics, 22, 51 57. Bradley, C. B. (1911) Graphic analysis of the tone accents of the Siamese Language. JAOS 282 289 Bradlow, A. R., Lively, S. E. Magnuson, J. S., Yamada,R. A. & Tohkura, Y. (1995). Testing the importance of tal ker variability in non native speech contrast training Paper presented at the 129th Meeting of the Acoustical Society of America, Washington, D.C. Bradlow, A. R., Pisoni, D. B., Yamada, R. A., & Tokhura, Y. (1997) Training Japanese listeners to identify English /r/ and /l/: IV. Some effects of perceptual learning on speech production. Journal of the Acoustical Society of America 101, 2299 2310. Best, C., & Strange, W. (1992). Effects of phonological and phonetic factors on cross language perception of approximants. Journal of Phonetics, 20, 305 350. Best, C. T. (1995). A direct realist view of cross language speech perception. In W. Strange (ed.), Speech perception and linguistic experience: Issues in cross language Research (pp. 171 204). Baltimore: Y ork. Chun, D. (1982). A contrastive study of the suprasegmental pitch in modern German, American English and Mandarin Chinese Ph.D. dissertation, University of California. Clark, J., & Yallop, C. (1995). An Introduction to Phonetics and Phonology Cambr idge: Blackwell.

PAGE 105

105 Clopper, C. G. & Pisoni, D. B. (2004). Effects of Talker Variability on Perceptual Learning of Dialects. Language and Speech. 47(3). 207 239. Catford, J. C. (1989). A Practical Introduction to Phonetic Oxford: Clarendon Press. 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. 229 273). Baltimore: York. Gahl, S. & Yu, C. A. (2006) Introduction t o the special issues on exemplar based models in Linguistics The Linguistics Review 23, 213 216. Goldinger, S. D., Pisoni, D. B. & Logan, J. S. (1991). On the nature of talker variability effects on recall of spoken word lists. Journal of Experimental P sychology: Learning Memory & Cognition, 17, 151 162. Goldinger, S. D (1996) Words and voices: Episodic traces in spoken word identification and recognition memory Journal of Experimental Psychology: Learning, Memory, and Cognition 22, 1166 1183. H enderson, E., J.A. (1982) Tonogenesis: Some Recent Speculations on the Development of tones Transaction of the Phisological Society. Hudak, T. (1990) Thai. In B. Comrie, (ed.), 756 775 New York: Oxford University Press. Ja mieson, D.G., & Morosan, D.E. (1989). Training New, Nonnative Speech Contrasts: A Comparison of the Prototype and Perceptual Fading Techniques. Canadian Journal of Psychology 43(1), 88 96. Jitapunkul, S., Maneenoi, E., Ahkuputra, V., & Luksaneeyanawin, S (2003). Performance evaluation of phonotactic and contextual onset rhyme models for speech recognition of Thai language. In Eurospeech 2003 1841 1844. Johnson, K. (1997) Speech perception without speaker normalization: An exemplar model In Talker var iability in Speech Processing Johnson, K. & Mullenix, J. (eds.), 145 166. San Diego: Academic Press. Kirk, K.I., Pisoni, D.B. & Miyamoto, R.C. (1997). Effects of stimulus variability on speech perception in hearing impaired listeners. Journal of Speech, Language, and Hearing Research, 40, 1395 140. Kirioff, C. (1969). On the auditory discrimination of tones in Mandarin. Phonetica, 20, 63 69.

PAGE 106

106 Kuhl, P. K. (1992). Psychoacoustics and speech perception: Internal Standards, Perceptual Anchors, and prototype s. In L. Werner & A. james (eds.): Developmental Psychoacoustics. American Psychological Association (pp. 293 332). Washington, DC: American Psychological Association. Lenneberg, E. (1967). Biological foundations of language New York: Wiley. Leather, J. (1990). Perceptual and productive learning of Chinese lexical tone by Dutchand English speakers. In J. Leather and A. James (eds.): New Sounds 90: Proceedings of the Amsterdam Symposium on the Acquisition of Second Language Speech (pp. 305 341). Amsterdam : University of Amsterdam. Lively, S. E., Logan, J. S., & Pisoni, D. B. (1993). Training Japanese listeners to identify English /r/ and /l/: II. The role of phonetic environment and talker variability in learning new perceptual categories. Journal of the Acoustical Society of America 94, 1242 1255. Lively, S. E., Pisoni, D. B., Yamada, R. A., Tokhura, Y., & Yamada, T. (1994). Training Japanese listeners to identify English /r/ and /l/: III. Long term retention of new phonetic categories. Journal of the Acoustical Society of America, 96 2076 2087. Logan, J.S., Lively, S.E., & Pisoni, D.B. (1991). Training Japanese listeners to identifyEnglish /r/ and /l/: A first report. Journal of the Acoustical Society of America, 89, 874 886. Luksaneeyanawin, S. (19 93). Speech computing and speech technology in Thailand. In Proceedings of the Symposium on Natural Language Processing, 276 321 Bangkok, Thailand Luksaneeyanawin, S. (1998) .Intonation in Thai. In Hirst, D. and Di Christo, A. (Ed.), Intonation Systems A Survey of Twenty Languages Cambridge University Press, Cambridge. Morn, B., Zsiga, E. (2006). The lexical and postlexical phonology of Thai tones. Natural Language & Linguistic Theory, 24 113 178. Nosofsky, M., R. (1988) On Exemplar Based Exempla r Representations: Reply to Ennis (1988) Journal of Experimental of Psychology, 117, 412 414. Patkowski, M. (1989). Age and accent in second language: A reply to James Flege. Applied Linguistics, 11 73 89. Piske, T., MacKay, I. R. A., & Flege, J. E. (2 001) Factors affecting degree of foreign accent in an L2: a review. Journal of Phonetics, 29(2), 191 215.

PAGE 107

107 Pisoni, D.B., Aslin, R.N., Perey, A.J. & Hennessy, B.L (1982). Some effects of laboratory training on identification and discrimination of voicing c ontrasts in stop consonants. Journal of Experimental Psychology: Human Perception and Performance, 8, 297 314. Pakasri, Y., Wallapa, T. & Supak, M. (2007). Thai in Daily Life Bangkok: Aempan. Pisoni, D.B. (1997). Some Thoughts on "Normalization" in Spee ch Perception In K. Johnson and J.W. Mullennix (Eds.). Talker Variability in Speech Processing 9 32. San Diego: Academic Press. Ponmanee, S. (2002a). Foundations of teaching Thai as a foreign language Bangkok: Chulalongkorn University Press. Ponmanee, S. (2002b). Language and Teaching Bangkok: Klet Thai. Potibal, P. (2005). The Problems of Thai Pronunciation for Korean Students. Paper presented at the International Conference on Teaching Asian Languages to Non Native Speakers, Bangkok, Thailand, Aug ust 4 5 Poomsan, B. B. (1995). Thai for Beginners. Bangkok: Paiboon Publishing. Scovel, T. (1969). Foreign accents, language acquisition, and cerebral dominance. Language Learning 19, 245 253. Smith, D. (1995) Thai: A Complete Course for Beginners L ondon: Hodder Headline Publishing. Smith, D. (2005). Teaching Thai to non native speakers: some lessons from the past Paper presented at the International Conference on Teaching Asian Languages to Non Native Speakers, Bangkok, Thailand, August 4 5 Sat hiansukon, S. (2005). Situations and Problems Found in Intensive 60 hour ThaiLanguage Course: A Case Study of Danish Students Taking 10 Day and 15 Day Courses Paper presented at the International Conference on Teaching Asian Languages to Non Native Speake rs, Bangkok, Thailand, August 4 5 Satravaha, N., Klinkhachorn, P. & Lass,N (2002). Tone classification of syllable segmented Thai speech based on multilayer perception. The Journal of the Acoustical Society of America 111, 2366 2366. Sittikesorn, S. (2 005). The relationship between the error of reading and the perception of Thai monosyllables: A case study of Japanese students who learn Thai as a foreign language. Wannawitat, 5, 168 185.

PAGE 108

108 Sommers, M.S. & Barcroft, J. (2007). An integrated account of th e effects of acoustic variability in first language and second language: Evidence from amplitude, fundamental frequency, and speaking rate variability Applied Psycholinguistics, 28, 231 249. Abramson, A. S. & Tingsabadh, K., (1999). Thai final stops: Cross language perception. Phonetica, 56, 111 122. Udompan, C. (2005). Problematic cases in pronouncing Thai sounds for learners. Paper presented at the International Conference on Teaching Asian Languages to Non Native Speakers, Bangkok, Thailand, August 4 5. Vongvipanond, P. (2000). Survival Thai. Bangkok: Chulalongkorn University Press. Wang, Y., Spence, M.Jongman, A., & Sereno, J. A. (1999). Training American listeners to perceive mandarin to ne. Journal of the Acoustical Society of America, 106, 3649 3658. Wang, Y., Sereno, J., Jongman, A., & Hirsch, J. (2001). Cortical reorganization associated with the acquisition of Mandarin tones by American learners: An fMri study. In Proceedings of the Sixth International Conference on Spoken Language Processing II (pp. 511 514), Beijing. Wang, Y., Jongman, A., & Sereno, J. A. (2003a). Acoustic and perceptual evaluation of Mandarin tone productions before and after perceptual training. Journal of the Ac oustical Society of America, 113, 1033 1043. Wang, Y., Sereno, J.A., Jongman, A., & Hirsch, J. (2003b). fMRI evidence for corticalmodification during learning of Mandarin lexical tone. Journal of Cognitive Neuroscience, 15(7), 1 9. Wayland, R. (1997) Non native production of Thai: acoustic measurements and accentedness ratings, Applied Linguistics, 18, 345 373. Wayland, R. & Guion, S.G. (2003). Perceptual discrimination of Thai tones by nave and experienced learners of Thai. Applied Psycholinguistics, 2 4, 113 129. Wayland, R. & Guion, S.G. (2004). Training English and Chinese listeners to perceive Thai tones: A Preliminary Report. Language Learning, 54 (4), 681 712. Wayland, R, Kaan E, Bao, M. & Barkley, CM ( 2007 (2007). Effects of native language and training on lexical tone perception: An event related potential study. Brain Research, 1148 113 122.

PAGE 109

109 Werker, J. F., & Tees, R. C. (1984). Phonemic and phonetic factors in adult cross language speech perception. Journal of the Acoustical Society of Americ a, 75, 1866 1878. Wittayasakpan, S. (2002). Bap rian pasaThai buengton nai boribot Thai sueksa samrap chao tangchat. Bangkok : University Bureau. Wittayasakpan, S. (2005) Teaching Thai as a Foreign Language Wannawitat, 5, 215 261. Yip, M. (2002) Tone New York: Cambridge University Press.

PAGE 110

110 BIOGRAPHICAL SKETCH Donruethai Laphasradakul, daughter of Saner Khaodeedech and Be nchaphorn s from the Thai department of Chulalongkorn University, Thailand She completed her undergraduat e educat ion with a Top Scores Achievement Award in Thai Courses and Second Class honors. She was awarded a fellowship for her MA studies and then became a teaching assistant at the department While studying for her MA, s he st arted to work as a lecturer at the Department of Th ai Language for Communication, t he University of the Thai Chamber of Co mmerce (UTCC). She also won a full scholarship to the Ph.D. program in Linguistics at the University of Florida from UTCC and she will r eturn to UTCC after graduation. As a Thai language teacher who has enthusiastically been teaching Thai for eight years, she has always been fascinated by the use of Thai as a second language. The Seminar on Teaching Thai as a Foreign Language 2002, at Ma hasarakham University, Thailand, inspired her chiefly to search for new aspects of Second Language Acquisition to strengthen her capability in teaching Thai to foreigners. Therefore, her aim is doing research on teaching Thai as a second language, focusing on Thai tones.