Perception of Mandarin Chinese Tone 2 / Tone 3 and the Role of Creaky Voice

MISSING IMAGE

Material Information

Title:
Perception of Mandarin Chinese Tone 2 / Tone 3 and the Role of Creaky Voice
Physical Description:
1 online resource (99 p.)
Language:
english
Creator:
Cao, Rui
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Linguistics
Committee Chair:
Wayland, Ratree
Committee Co-Chair:
Kaan, Edith
Committee Members:
Cowles, Heidi
Fu, Danling

Subjects

Subjects / Keywords:
chinese -- creaky -- tones
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:
Research has shown that lexical tones, a suprasegmentalfeature, are processed by native speakers as linguistic elements just likeother segmental information. Among the four tones of Mandarin Chinese, inparticular, Tone 2 and Tone 3 are very similar in their pitch contour shapesand thus can be difficult to distinguish in native and nonnative perception. Aphonation type, creaky voice, has been reported to be associated withproduction of Tone 3. This study investigated the perception of Mandarin Tone 2and Tone 3 and the role of creaky voice in the perception of Tone 3 amongnative and nonnative listeners. The results showed that creaky voice does notserve as the major cue in perception of Mandarin Tone 3. In fact, the presenceof creaky voice seems to pyschoacoustically obscure the actual duration of thefalling portion in the tonal contour, making it shorter in perception amongboth native and non-native listeners.
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 Rui Cao.
Thesis:
Thesis (Ph.D.)--University of Florida, 2012.
Local:
Adviser: Wayland, Ratree.
Local:
Co-adviser: Kaan, Edith.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-02-28

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2012
System ID:
UFE0044096:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 PERCEPTION OF MANDARIN CHINESE TONE 2 / TONE 3 AND THE ROLE OF CREAKY VOICE By RUI CAO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGRE E OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012

PAGE 2

2 2012 Rui Cao

PAGE 3

3 To my Mother and late Father

PAGE 4

4 ACKNOWLEDGMENTS It has been a very long journey and there are many people I need to give my thanks to. First of all, I thank my advisor and the chair of my committee, Dr. Ratree Wayland, without whom this dissertation would not have been possible. I am thankful for her wisdom and guidance throughout my studies at the University of Florida. She has given me so many valuable insights and feedb ack on my research. I have grown professionally as well as personally following her as a mentor and role model. I would also like to thank my co chair, Dr. Edith Kaan, and my other committee members, Dr. Wind Cowles, and Dr. Danling Fu. They have given me very helpful suggestions on my study and provided valuable feedbacks. I thank Dr. Caroline Wiltshire, Dr. Diana Boxer, Dr. Eric Potsdam, Dr. Gary Miller, Dr. Virginia LoCastro, Dr. Theresa Antes, and Dr. Ann WehMeyer for teaching me all the wonderful subj ects in linguistics. I am also thankful to the English Language Institute, the Department of Languages, Literatures, and Cultures at the University of Florida and the Defense Language Institute in Monterey, California for giving me opportunities to teach English as a Second Language and Mandarin Chinese. I will always be grateful to the professor who offered me an assistantship ten years ago when I graduated from college and made it possible for me to come to the Unit ed States for graduate studies Dr. Dou glas Coleman, from the University of Toledo (Ohio). He is also a gator himself and it was him who introduced me to the Linguistics at UF I thank him for making my dream come true. I am also obliged to many of my friends, especially Ming Ren, Weilin Zhang, Yan Yang, Jing Li, Louisa Chang, Donruethai Laphasradakul Priyankoo Sa rmah, Mutsuo Nakamura, Jimmy Huang and Shengming Yang, for their friendship and moral support

PAGE 5

5 all these years. Special thanks should go to Tom Ratican, who has treated me like a daugh ter of his own. I am thankful for his genuine care for me, and for providing valuable feedback in editing my writing. I would like to express my greatest gratitude to my mother, Keru Pang, my husband, Taran Champagne, and other family members. Without the ir love, understanding, and encouragement, this wor k would not have been possible

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 11 Research Questions ................................ ................................ ............................... 11 Research Design ................................ ................................ ................................ .... 13 Out line ................................ ................................ ................................ .................... 14 2 BACKGROUND ................................ ................................ ................................ ...... 15 Cross Language Speech Perception ................................ ................................ ...... 15 Tone Languages ................................ ................................ ................................ ..... 23 Tone Production and Perception ................................ ................................ ...... 24 Mandarin Chinese ................................ ................................ ................................ ... 27 Native Perception and Production of Chinese ................................ .................. 27 Non Nativ e Perception and Production of Chinese and its influence on L2 acquisition of Tones ................................ ................................ ...................... 31 Categorical Perception and Mandarin Chinese Tones ................................ ..... 35 Creaky Voice and Mandarin Chinese ................................ ................................ ..... 40 Determination of Creaky Voice ................................ ................................ ......... 42 3 METHOD ................................ ................................ ................................ ................ 48 Participants ................................ ................................ ................................ ............. 49 Procedure and Stimuli ................................ ................................ ............................. 50 Tone Baseline Task ................................ ................................ .......................... 51 Stimuli ................................ ................................ ................................ ........ 51 Procedure ................................ ................................ ................................ .. 52 Tone Production Experiment ................................ ................................ ............ 52 Stimuli ................................ ................................ ................................ ........ 52 Procedure ................................ ................................ ................................ .. 53 Tone Categorization Experiment ................................ ................................ ...... 53 Stimuli ................................ ................................ ................................ ........ 53 Procedure ................................ ................................ ................................ .. 56 Debriefing ................................ ................................ ................................ ... 57

PAGE 7

7 4 RESULTS ................................ ................................ ................................ ............... 60 Results of the Baseline Experiment ................................ ................................ ........ 60 Results from the Production Experiment ................................ ................................ 63 Results from Perception Experiment ................................ ................................ ...... 64 Results of Tone Reponses for Each Stimulus ................................ .................. 64 Perceptual Boundaries of Tone 2 and Tone 3 ................................ .................. 66 Average Turning Point (ATP) ................................ ................................ ..... 67 Binary Logistic Regression for 50% Crossing Point (CP) ........................... 71 5 DISCUSSION AND CONCLUSION ................................ ................................ ........ 81 Research Question 1 ................................ ................................ .............................. 81 Research Question 2 ................................ ................................ .............................. 82 Research Question 3 ................................ ................................ .............................. 85 Conclusions, Limitations, and Fut ure Directions ................................ ..................... 91 APPENDIX: FORTY WORDS USED IN TONE BASELINE TASK (WITH PINYIN +TONE MARKS) ................................ ................................ ................................ ..... 93 LIST OF REFERENCES ................................ ................................ ............................... 94 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 99

PAGE 8

8 LIST OF TABLES Table page 3 1 Forty tokens of mono syllabic Chinese words used in tone production experiment ................................ ................................ ................................ .......... 58 4 1 Error rates for each tone category from Baseline task and mean scores ........... 77 4 2 Percentage of creaky voice present in four tones for NC ................................ ... 77 4 3 Percentages of four tone responses received on all stimuli from 3 groups. ........ 77 4 4 Two Average Turning Points (ATP) for Tone 2 responses (ms) for three proficiency groups ................................ ................................ .............................. 77 4 5 Two Average Turning Points (ATP) for Tone 2 responses for NC and NE (combined) ................................ ................................ ................................ .......... 78 4 6 Average Turning Point (ATP) for Tone 3 responses (ms) for three groups ........ 78 4 7 Average Turning Point (ATP) for Tone 3 resp onses (ms) for NC and NE (combined) ................................ ................................ ................................ .......... 78 4 8 NC and Combined NE of their Tone 2 and Tone 3 ATPs (ms) ........................... 78 4 9 SPSS output table of binary logistic regression on NC24 (Non creaky) ............. 78 4 10 Results of Binary logistic regression for three groups (predicted crossing point of perceptual boundary of Tone 2 and Tone 3). ................................ ......... 79

PAGE 9

9 LIST OF FIGURES Figure page 2 1 An example of modal, breathy, and creaky voice given from voiced vowels in Jalapa Mazatec w ords. ................................ ................................ ....................... 45 2 2 FFT spectra of modal breathy, and creaky vowel /a/ in three San Lucas Quizvini Zapotec words ................................ ................................ ...................... 45 2 3 Waveform of a To ne 3 [ma] produced with creaky voice. ................................ ... 46 2 4 Waveform of a Tone 2 [ma] produced without creaky voice. .............................. 46 2 5 FFT spectra of a slic e from the vowel [a] with creaky voice ................................ 46 2 6 FFT spectra of a slice from the vowel [a] without creaky voice ........................... 47 3 1 Natural uttera nce of [ma] with Tone 3 (non creaky), based on which stimuli in the categorization task were created ................................ ................................ .. 58 3 2 An example of an empty pitch tier in PRAAT with five pitch points add ed to create a n on creaky token ................................ ................................ ................. 58 3 3 An example of an empty pitch tier in PRAAT with five pitch points and a ................................ ................................ ................................ ....... 59 4 1 An example o f having Tone 2 responses again after 348 ms (from NC24, in Creaky condition). ................................ ................................ ............................... 79 4 2 Illustration of effect of creaky voice on tone responses. ................................ ..... 80 4 3 NC15 Tone 2/3 responses in Creaky condition ................................ .................. 8 0 4 4 NC21 Tone 2/3 responses in Creaky condition ................................ .................. 80 4 5 An exam ple of probability curve of a binary regression analysis. (B>0) ............. 80

PAGE 10

10 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 PERCEPTION OF MANDARIN CHINESE TONE 2 / TONE 3 AND THE ROLE OF CREAKY VOICE By Rui Cao August 2012 Chair: Ratree Wayland Cochair: Edith Kaan Major: Linguistics Research has shown that lexical tones, a suprasegmental feature, are processed by native speakers as linguistic elements just like other segmental informatio n. Among the four tones of Mandarin Chinese, in particular, Tone 2 and Tone 3 are very similar in their pitch contour shap es and thus can be difficult to distinguish in native and nonnative perception. A phonation type, creaky voice, has been reported to be associated with production of Tone 3. This study investigated the perception of Mandarin Tone 2 and Tone 3 and the role of creaky voice in the perception of Tone 3 among native and nonnative listeners. The r esults showed that creaky voice does not serve as the major cue in perception of Mandarin Tone 3. In fact, the presence of creaky voice seems to pyschoacoustically obscu re the actual duration of the falling portion in the tonal contour, making it shorter in perception among both native and non native listeners.

PAGE 11

11 CHAPTER 1 INTRODUCTION Research Questions This study aims to investigate the perception of Mandarin Tone 2 and Tone 3 and the role of creaky voice in the perception of these tones among native and nonnative listeners. suprasegmental level, an integral part attached to a syllable as an indispens able cue to phonological inventory: Tone 1 (high level), Tone 2 (high rising), Tone 3 (low dipping) and Tone 4 (high falling) (Chao, 1948). For instance, the syllable yi has f our different meanings depending on the tone; yi (high According to Abramson (1978), phonemic tones can be classified as static (or level) and dynamic (or contour) tones. In the three contour tones (Tone 2, Tone 3, and Tone 4) of Mandarin Chinese, Tone 2 and Tone 3 share a certain similarity in their physical fundamental frequency contours. Although Tone 2 is a high rising tone, there is a short initial falling in the tone contour before rising in its phonetic realization. This initial falling makes Tone 2, in reality, similar to a Tone 3 contour to a certain extent, a contour that also consists of a falling and rising pattern. The timing of the t urning point, then, seems to be critical in perceiving these two tones. What are the perceptual category boundaries? What cues do native listeners use to mark the boundaries? Do non native listeners perceive these two tones in the same way as natives?

PAGE 12

12 Na tive speakers of tone languages usually use one or more acoustic and perceptual cues in hearing tones, such as the fundamental frequency (F 0 ) (Howie 1972; Gandour 1984) amplitude contour (Chuang et al 1972), and sometimes duration (Chuang et al 1972). The t iming of the turning point plays a significant role in determining the shape of the tone contour (Jongman and Moore, 1997) Therefore, one of the focuses of the present study is the effect of timing of turning point on Tone 2 and Tone 3 perception. The current study not only investigates the range of the turning point that alters cue in Tone 2 and Tone 3 recognition creaky voice. Creaky voice is characterized b y reduced intensity in waveform, lower fundamental frequency, and less frequent pitch periods with irregular duration (Gordon & Ladefoged, 2001). This phonation type has been reported to be associate d with the production of Mandarin T one 3 ( Belotel Grenie, A. & Grenie, M. 1994 ) However, the role of creaky voice in Mandarin tone perception has not been studied extensively and it is worthwhile to examine it in finer detail especially its role in distinguishing Tone 2 and Tone 3 if any. The present study, therefore, explores the perceptual boundary between the two tones among native and non native listeners, as well as the role of creaky voice in perception of these two tones. The research questions that are addressed in the study are: Research Question 1 : Is creaky voice associated with a certain tone(s) produced by native Mandarin Chinese speakers ( NC ) ? Research Question 2 : How does the presence of creaky voice affect the perception of tones by native Chinese speakers (NC) and native English speaking L2 l earners of Chinese (NE), respectively?

PAGE 13

13 Research Question 3 : What is the perceptual boundary of the Tone 2 Tone 3 continuum perceived by native Chinese speakers (NC)? How does it differ from native English speaking L2 learners of Chinese (NE) with different proficiencies? What is the role of creaky voice in their perception? Research Design The overall design of this study involved testing two groups of listeners, native Mandarin Chinese (NC) speakers and native (American) English (NE) speakers who were seco nd language learners of Mandarin Chinese. The NE speakers were further divided into two subgroups based on their ability to identify Mandarin tones in a baseline tone identification task. The main task for both NC and NE listeners was a tone categorization task, designed to examine the perception of Tone 2 and Tone 3, as well as the role of creaky voice in the identification of Mandarin Tone 3. The frequency of occurrence of creaky voice in the production of Mandarin tones among NC speakers was also examine d. Since Belotel Grenie & Grenie ( 1994 ) conclusive picture of creaky voice with production of Tone 3, it is necessary to provide more evidence for this matter. Since one of the purposes of the perception experiment wa s to examine the role of creaky voice in the categorical perception of Tone 2 and Tone 3, the two sets of stimuli used in the experiment were consistent in all parameters except for the creaky part. Two sets of 34 stimuli were generated. The contour of the creaky toke ns was the same as for the non creaky stimuli, but with creakiness created at the dip of each stimulus. The prominent characteristics of creaky voice include irregular periodicity and a sudden decrease in fundamental frequency and intensity. The creakiness wa s generated by randomly assigning low pitch points at the dip to create a jitter Each of the participants did a baseline task and a perceptual task, while native Chinese speakers also

PAGE 14

14 performed a production task. In the perception experiment, the ton e tokens were presented by computer and the participants needed to make judgments regarding wh ich tone s they heard. The purpose of the production task was to decide whether there is an association between creaky voice and a certain tone in Mandarin Chinese Outline This dissertation is organized in the following manner. In Chapter 2 several cross language speech models are reviewed, as well as literature on Tone languages, on Mandarin Chinese tones, and phonation types. The detailed methods of the study a re described in Chapter 3. Results from production data and perception experiments are presented in Chapter 4, followed by a discussion of findings, suggestions for future research, and concluding remarks in C hapter 5

PAGE 15

15 CHAPTER 2 BACKGROUND In this chap ter we review the literature relevant to the present study. First, several cross language speech models are discussed, since the research reported herein includes comparisons between native and non native Chinese speakers. Perception in tone languages is t hen reviewed, particularly in Mandarin Chinese, (including native as well as non native perspectives). Following this we review phonation type including how creaky voice is described in previous research and how it is associated with Mandarin Chinese. Fina lly, a summary of the rationale behind the present study is given. Cross Language Speech Perception In the middle of the 19 th century, when the treaty of Nanjing was signed, ending the first Opium War between the British and Qing Empires, the five ports o f China were opened for trade. Shortly afterwards, the first English textbook in China, appeared in one of the port cities, Guangdong (Canton). The book collected some common English words and phrases with pronunciation written in Chinese cha racters. As amusing as it is, this anecdote has indeed, to some extent, depicted how people have made an effort to hear and speak foreign words. The difficulties in master ing a nonnative language have always been of interest to people and in recent decades researchers have begun to gain a better understanding of the basic perceptual mechanism of speech sounds. In recent years particularly, with advanced technology that has helped research, exploration of the neurological mechanism of speech perception has also emerged as a new area to complement previous behavioral studies.

PAGE 16

16 Though our understanding of the neural mechanism is still limited, there are some influential models proposed in phonetics research to predict or account for patterns of cross language speech perception, mostly focusing on segmental features. The three dominant models are reviewed in this section: 1) the Speech Learning Model ( SLM ) by James Flege, 2) the Perceptual Assimilation Model ( PAM ) by Catherine Best, and 3) the Native Language Magnet model ( NLM ) by Patricia Kuhl. In addition, these models will be extended to account for the acquisition of suprasegmental features lexical tones by non native speakers on the basis of which, predictions on phonological acquisition of Mandarin Chinese tones by native English speakers will be made. Speech Learning Model (SLM). Flege (1995, 2002) proposed the speech learning model in order to account for how individuals learn or fail to learn how to perceive and produce L2 consonants and vowels in a native like fashion. The major questions that SLM focuses on are whether or not certain L2 sounds are learnable, whether they are learnable only by children, and some issues i n the perception of speech sounds. This model was developed with the background of the Critical Period Contrastive Analysis (CA) hypothesis, which claims that L2 phonemes that are similar to L1 are easier than those that are different. target subjects described in SLM are bilinguals who have had some experience in speaking their second language seven hypotheses on second language sound acquisition.

PAGE 17

17 The model start s with a controversial postulate that the mechanism establishing the L1 sound system remains intact across the life span and can be accessible for L2 learning. Later Flege (2002) pointed out that this notion does not mean adult learners will ultimately achieve the same proficiency as child learners because other factors will interfere with the development of long term memory represe ntations of an L2 segment that are identical to a monolingual 's L1 Phonetic categories are also defined in the postulates as the specified language specific aspects of speech sounds in long term memory representations. The other two postulates state that L1 phonetic categories evolve over a that bilinguals maintain a contrast between L1 and L2 phonetic categories in a so T he two languages thus eventually influenc e each other. SLM (Flege 1995, 2002) propose s some important hypotheses to account for cross language speech perception and production, including: S ounds in L1 and L2 are perceptually related to each other at an allophonic level rather than at an abstract phonemic level; If an L2 sound is perceived as very phonetic ally different from the closest L1 sound, it is very likely that a new category will be established for the L2 sound; The likelihood for (2) to happen increases when the perceived dissimilarity increases; When the age of learning an L2 increases, the likelihood of discerning L1 and L2 sounds, as well as L2 sounds that are noncontrastive in the L1, will be decreased; SLM propose s the concept of equivalence classification, which may eventually pre vent creating new categories for certain L2 segments. This means that when equivalence classification happens, a single phonetic category will be formed to process perceptual ly linked L1 and L2 sounds (diaphones), which in production will resemble either L 1 or L2 sounds A bilingual may establish a phonetic category that differs from a monolingual s if L2 phonological space, or if the

PAGE 18

18 mo Production of a sound will ultimately correspond to its property presentation established in its phonetic category. This implies that without an accurate perceptual input, the production of a n L2 sound will be inaccurate. There are two speci fic mechanisms that SLM (Flege 1995) proposes through these hypotheses by which the phonetic categories that represent L1 and L2 phonetic systems interact: category assimilation and category dissimilation When a category is not formed for an L2 sound due to its great similarity to an L1 sound, then the long term memory representation that is used for an L1 phonetic category and a similar L2 phonetic category will assimilate, gradually leading to a in which he examined the production of /t/ in French and English words by English French and French English bilinguals both of whom were late L2 learners. The segment /t/ in English is usually realized with long lag VOT values, and French /t/, on the other hand, is usually realized with short lag VOT values. The English and French monolinguals production data confirmed this. Interestingly, French English bilinguals produced English /t/ with longer ever, the VOT of their English /t/ was not as long as that of English monolinguals. In contrast, the English production of English /t/; however, the VOT of their French /t/ w as longer than that of French monolinguals. Thus, the study showed that neither of the L2 groups established a new phonetic category for the L2 /t/; instead, the two sources of input information for

PAGE 19

19 /t/ from both L1 and L2 seemed to have caused them to mer ge their /t/ to somewhere between L1 and L2, reflecting both features of English /t/ and French /t/. The other mechanism, category dissimilation, happens when a new category is established for an L2 sound I t may differ from neighboring L1 and/or L2 sound s that are produced by a monolingual. Flege & Eefting (1987), for example, examined the production of /p t k/ in Spanish and English words by Spanish English bilinguals. Those segments are usually realized with short lag VOT in Spanish and long lag VOT in English. The results showed that both adult and child bilinguals produce Spanish /p t k/ with much shorter VOT than Spanish monol inguals This is consistent with the hypothesis that the bilinguals seemed to modify their L1 /p t k/ to make them more distinc t phonetically from those categories that they established in English. Perc eptual Assimilation Model (PAM). Another influential model in cross model focuses primarily o n the perception of a non native language by nave learners who have not had any learning experience with the language. The fundamental premise similarities to, and discrepanc ies from, the native segmental constellations that are in According to PAM, three patterns of perceptual assimilation of non native segments will happen: Assimilat ion to native category. This means a non native segment could be heard by a nave listener as either a good exemplar of his/her native category, an acceptable but not perfect exemplar of the native category, or a notably deviant exemplar.

PAGE 20

20 Assimilat ion as an uncategorizable speech sou nd. This means that if a non native sound cannot be categorized/ assimilated to any native categories, but still is heard as a speechlike sound, then it is likely to be assimilated within native phonological space. Not assimilated to speech. This happens w hen a non native sound cannot be assimilated to native phonological space and is thus perceived as a nonspeech sound. Although PAM does not directly address issues in cross language speech learning, it can be extended to apply to the second language learni ng process. In order to predict how nave listeners discriminate nonnative phonological contrasts, PAM also proposed patterns for how each phone in a contrasting nonnative pair is perceptually assimilated. For instance, very good discrimination is predicte d for Two Category (TC) assimilation; poor discrimination is predicted for Single Category (SC); etc. PAM takes into account both phonological and phonetic levels when explaining how the L1 system influences the perception of nonnative sounds (Best & Tyler 2007). One notion in which PAM differs from that of SLM i s that listeners extract information about articulatory gestures from the nonnative speech whereas the learners described in SLM try to form phonetic categories based on the acoustic cues they hear In a manner similar to SLM, PAM also agrees that listeners continue to refine their perception of speech gestures over their lifetime. Native Language Magnet model (NLM) Kuhl and her colleagues (1992, 1994) developed the Native Language Magnet model (N LM) with the goal of characteriz ing the developmental changes reflecting how infants reorganize their phonetic perception during first language acquisition. The basic claim of NLM is that infants have the ability to discriminat e any contrast in speech soun ds in the world even if they have not heard them before B y the end of their first year, this ability is gradually los t and their brain s

PAGE 21

21 are more tuned to only the native sounds to which they are exposed. In their experiments, both American adults and 6 mo nth old infants exhibited a perceptual magnet effect, such that if two tokens were within one category, their physical phonetic distance w ould be perceptually shrunk or ignored by listeners, and if the two tokens were perceived as in a different category, their physical phonetic distance w ould be perceptually stretched. The implications of NLM can be extended to both first language acquisition and adults learning a second language. For an infant, the shrinking and stretching of perceptual space can actuall y facilitate L1 acquisition since the irrelevant information will be left out in perceiving phones categorically. For adults learning a second perceptual space by reducin g sensitivity near phonetic prototypes, and these perceptual effects can be difficult to alter (Kuhl & is close to a native language prototype, NLM predicts that it will be very difficult for L2 learners to perceive the phonetic contrast in the second language. H ow then, can these models be extended to account for the acquisition of lexical tones by nonnative speakers and what predictions can we make? Those models deal mainly with the segmental features o f speech sound perception, but variations in discriminating nonnative speech go beyond vowels and consonants in many languages, for example, the tonal contrast in tone languages. like Manda rin Chinese, and an intonation language like English. Mandarin Chinese uses pitch information to discriminate lexical meanings, in addition to vowels and consonants, the

PAGE 22

22 three of which are all integra lly perceived by native speakers. On the other hand, in an intonation language, pitch is used to signal syntactic or emotional information at the phrase or sentence level. Therefore when a non native speaker starts to learn a tone language, great difficulty can be predicted simply because in their native phono logical system, tone as an integral cue for identifying a speech sound does not exist. Both SLM and PAM touch upon the relationship between the difficulty of perceiving an L2 sound contrast and the similarity/difference between L1 and L2 sounds. SLM claim s that the more an L2 sound is perceived differently from the closest L1, the more likely a new category for the L2 sound will be established. However, in the tone language case, this cannot be applied, because the phonetic distance her e is not at the same dimension since tones are used by non tonal speakers for pragmatic purposes, such as asking a question or express ing emotions. On the other hand, if the new cue is given much attention and can be successfully separated as another dimension of perception b y an L2 learner, then the SLM model would predict that it will be more likely for the L2 learner to produce the L2 sound in an accurate way. Moreover, both SLM and PAM agree that throughout the life span, learners will continue to refine their perception o f speech gestures. In fact, studies have shown that even for nave speakers of tone languages, the perception of tones is not totally non exist ent (Halle, Chang & Best, 2004). The NLM effect predicts that if an L2 sound falls within the psychoacoustic spa ce of an L1 sound, the L2 sound will be difficult to discriminate (Wayland, 2007). If this model is to be extended to tone acquisition, the n L2 learners will face great difficulties in

PAGE 23

23 extracting pitch contour information instead of only paying attention t o the averaged pitch. For instance, for Tone 1 in Mandarin Chinese, the phonological category does not change when the averaged pitch height changes, which might be easier for native English speakers, since there is no nonnative contrast in perceiving Tone 1. However, for Tone 2 and Tone 3, which are acoustically very similar and usually only differ in timing of the turning point, then English speakers have to learn to be more sensitive to the pitch contour instead of the absolute pitch value. In this latte r case, the nonnative t hus it will be especially difficult to learn. In conclusion, the three influential speech perception models have been reviewed in terms of their m ajor claims and hypotheses. They address the issues in speech perception o f segments from slight ly different point s of view and all ha ve implications in cross language speech learning. In the next few sections, background on perception of tone languages, p articularly Mandarin Chinese tones, and creaky voice is reviewed. Tone Languages According to Yip (2002), in approximately 60 determine meaning at a lexical level. Tone languages can be found in Asia, West Africa, and Eu population (Fromkin, 1978). Pitch variation is used in all the meaning, emotion, or intent at a sentential level (Burnham & Mattock, 2007). For instance a rising intonation of a sentence conveys a question, as in the English talk. These are different from using pitch variation or tone at a lexical level to distinguish lexi cal meaning. For example, in Thai, there are five phonologically distinctive tones;

PAGE 24

24 therefore for the same syllable /k h a:/ there will be different meanings depending on the tone, e.g. /k h h h a:/ h h Yip (2002) g ives a clear explanation of three important terms: fundamental frequency (F 0 ), pitch, and tone. According to Yip (2002), F 0 is a pure phonetic term which refers to the signal and how many pulses per second the signal contains, expressed in Hertz (Hz). Each pulse, in the case of speech, is produced by a single vibration of the vocal folds. Pitch, on the other hand, is more a perceptual term, erception is of a signal. Very often, pitch and fundamental frequency are used interchangeably; however, there is no one to one correspondence between these two (Jongman et al. 2007). Pitch can also consist of non speech signals, for example, in music tha t varies in pitch. The third term, tone, is a linguistic Tone Production and Perception How is tone produced? As previously mentioned, the fundamental freq uency of a sound, which is perceived as pitch, is a function of the rate of vocal fold vibration (Ohala, 1978). According to the description in Jongman et al. (2007), tone production is hanges in fundamental frequency (or in rate of vocal fold vibration) are made by manipulating tension in the vocal folds. In order to perceive a tone, a hearer must depend in whole or in part on pitch, and hence on fundamental frequency (Yip, 2002). The signal must contain a F 0 that fluctuates, which also needs to be large or prominent enough to be detected in order to have pitch differences. Other factors include duration and amplitude.

PAGE 25

25 Before tone perception, an important issue needs to be discussed, pitch detection. that th e minimal detectable differences for sounds with a level F 0 to 2Hz if the sound had F 0 with ramp or slope. Of course in human languages, tones are much further apart in tonal s pace (Yip, 2002). It was also shown in 1973 ) study that pitch is easier to discriminate on a steady state vowel than a non steady vowel, as in a diphthong. Therefore the duration is important for pitch detection as well, especially for contours. According to Greenberg & Zee (1979), if the syllable is less than 40 65ms long, contours cannot be perceived. Instead, they are mostly perceived as level. According to them, a real robust percept ion for the entire signal duration, though this is longer than some stop final syllables in many Chinese dialects (Yip 2002). If F 0 is important for pitch detection, it is therefore, reasonable to assume that it is also the primary cue for tone perception. This is true in some languages; however, Yip (2002) point s out that in some other tone languages, tones differ not only in F 0 but also in duration, amplitude, and voice quality or phonation type According to Gandour (1978), for many tone languages, including Thai, Mand arin Chinese, Yoruba, and Swedish, F 0 is an indispensable cue for tone recognition. Some studies (e.g. Fork 1974, Abramson 1978) have shown that when other cues are removed from the signal except for the fundamental frequency, native speakers can still dis criminate tones with a high degree of accuracy. In another study (Cao & Sarmah, 2007) investigating Mandarin Chinese Tone 2 and Tone 3, native speakers did not recognize the synthesized stimuli with only F 0 the same way as when all other cues

PAGE 26

26 were present. On the other hand, if the original fundamental frequencies of stimuli were removed and only cues like duration and amplitude remained intact, tonal discrimination was greatly impaired (e.g. Whalen & Xu, 1992; Fu & Zeng, 2000). Although fundamental freque ncy is very important in perception in tone languages, there are slight differences in the weighing of this cue in different cases. It has been shown that Thai listeners can easily identify the tones in monosyllabic words (Abramson, 1962). Though Thai has three level tones, high, mid, and low, the confusion of mid and low tones result s in only a very small number of errors, which are eventually eliminated when the stimuli are produced by one single speaker instead of ten (Abramson, 1976). T esting fundamenta l frequency alone can be a cue for identifying tones When Abramson (1962) for example, superimposed synthetic averaged F 0 contours on each of the natural speech monosyllabic words: naa the identification rate by native speakers was near ly perfect. The study suggested that fundamental frequency can override other cues like duration or amplitude that may be associated with a tone. However, the addition of amplitude can enhance the percepti on of tones, although this addition by itself is not sufficient enough for tone identification (Abramson, 1975). In the case of Yoruba, a Kwa language spoken in Nigeria, there are also three contrastive level tones: high, mid, and low. T here is a phenomen on in this language that on disyllabic words with high high, mid mid, and low low patterns, only the final low tone is markedly lower than its preceding tone. Studies (Hombert 1976 for example ) have investigated the relative importance of fundamental fre quency, amplitude, and duration in distinguishing a low tone from a mid tone in word final position. The results from that

PAGE 27

27 study indicated that if the duration or amplitude increases, it d oes not cause a shift in identification, which suggests that fundame ntal frequency is the principal acoustic correlate of Yoruba tones, except that the falling contour is the primary perceptual cue of a low tone in word final positions. In Mandarin Chinese there are other parameters to be used as perceptual cues besides f undamental frequency, such as intensity and duration. Perception and production of Mandarin Chinese is reviewed in the next section. Mandarin Chinese Native Perception and Production of Chinese Mandarin Chinese has four contrastive tones in its phonolog ical inventory: Tone 1 (high level), Tone 2 (high rising), Tone 3 (low dipping) and Tone 4 (high falling) (Chao, 1948). For instance, the syllable yi has four different meanings depending on the tone; yi (high F undamental frequency contours have been examined in many studies to show their primacy in the perception cues of Mandarin Chinese. For example, in one of Howie ents, synthetic tones were imposed on the real speech Tone 1 syllable bao producing four different tone stimuli N ative speakers demonstrated 95% accuracy in identification tasks. In his other study, a set of stimuli w as generated to form tokens that were minimally differentiated by tone s taken from real speech citation syllables and their fundamental frequenc ies were suppressed and replaced by a constant fundamental frequency of 128 Hz. It turned out that native speakers did very poorly in tone identifica tion. This suggests the primacy of fundamental frequency patterns in Mandarin Chinese tone identification.

PAGE 28

28 Gandour (1984) conducted a study to examine whether F 0 height or F 0 contour is more important as a perception cue. The results suggest that both of them are important, though native listeners seem to weigh F 0 contour slightly more important ly than F 0 height. In another study Massaro, Tseng & Cohen ( 1985) examined Tone 1 and Tone 2, and showed that both cues were used by native listeners, while neithe r F 0 height nor F 0 contour alone w ere sufficient enough for accurate identification. In the contour tones of Mandarin Chinese, the confusion between Tone 2 and Tone 3 has long been reported and studied. In physical fundamental frequency contours, these tw o tones share a great similarity. What, then, are the acoustic properties that can distinguish these two tones? Shen & Lin (1991) have reported that besides the F 0 height, which contributes to distinguish ing these two tones, there are two other parameters that are relevant These include: (i) the timing of the turning point, defined as the duration from the onset of the tone to the point of change in F 0 direction, and (ii) the decrease in F 0 from the onset of the tone to the turning point, which they call t 0 Shen & Lin (1991) found that Tone 2 usually has an earlier turning point and 0 than Tone 3. 0 is set at 30 Hz, native speakers of Mandarin Chinese perceive Tone 3 when the turning point is more than 40% of the total length of th e stimuli. W 0 is 15 Hz, Tone 3 is perceived when the turning point is more than 60 70% of the total length of the stimuli. However, Shen and Lin (1991) were not specific about the geographical affiliation of the Mandarin Chinese speakers who partici pated in their study. Marked idiosyncrasies among Mandarin Chinese speakers in producing Tone 3 can be safely considered as regional features; e.g. Yip (2002) reports

PAGE 29

29 that the Tianjin variety of Mandarin Chinese differs significantly from the Beijing varie ty in terms of the production of Tone 3. Other perception studies have also shown that these two parameters are very important cues to distinguishing Tone 2 and Tone 3. Moore & Jongman (1997) report ed the simultaneous effect of timing of the turning point 0 in Mandarin Chinese 0 systematically in isolated synthetic speech stimuli. They first created 12 stimuli having turning points at various times from 20 to 240 ms in 20 ms steps, then each stimul us 0 ranging from 10 to 70 Hz in 5 Hz steps. All 156 stimuli (12 X 13) were presented to native speakers of Mandarin Chinese in a forced choice perception te st where lexical entries ) were the choices. The results showed that when the turning point is earlier 0 is below 30 Hz, more Tone 2 is 0 is larger than 35Hz, subjects re port Tone 3 responses. They conclude d that 0 becomes crucial in the 0 needs to be more than 35 Hz. d the turning point in perceiving Mandarin tones. The results showed that when the timing of the turning point is between 42.5% and 72.5% of the total length of the stimuli, the stimuli will be associated with the ). Howev er, if the timing of the turning point is less than 42.5% of the total duration of the stimuli, the stimuli will be associated with the

PAGE 30

30 ). If the turning point occurs after 72.5% of the total length of the stimuli, the stim ). confusion between Tone 2 and Tone 3, but the pattern of confusion is asymmetrical. There were many more error s result ing from misidentifying Tone 3 as Tone 2, than from misidentifying Tone 2 as Tone 3. If this cannot be easily explained by the cues discussed above it could, perhaps, be related to a phonological rule that Mandarin Chinese has, known as tone sandh i (Chao, 1948) a Tone 3 becomes Tone 2 when it precedes a Tone 3. For instance, the two compounds, fen chang (Tone 2 Tone 3) chang (Tone 3 connected speech. Besides the primary cue, F 0 in Man darin Chinese tone perception, there are other cues, such as duration and amplitude. From production data, it has been shown that Mandarin tones also differ in their overall duration (e.g. Chuang et al. 1972). Tone 2 and Tone 3 seem to be longer than the other two tones, and Tone 4 is the shortest, although this may change if they are in different positions and serve different functions in a sentence. Blicher et al (1990) reported that lengthening the duration of stimuli which are ambiguous between Tone 2 and Tone 3 will make the identification more often Tone 3. In Cao & Sarmah (2007), even though the total duration of ambiguous Tone 2 and Tone 3 was kept constant, when duration of turning point and/or the timing of turning point changes, duration alone could not be a distinctive cue for the perception of Tone 3 as other factors w ould still cause native speakers to ma k e judgments favoring Tone 2.

PAGE 31

31 Studies have also investigated how amplitude affects tone perception. For instance, Chuang et al. (1972) showe d that overall Tone 4 has the highest amplitude and Tone 3 has the lowest. Although amplitude is considered to have little effect on tone perception, there is some evidence that it can be used alone as a perceptual cue. When Whalen & Xu (1992) for exampl e, removed the F 0 and formant structure from a natural speech signal and only the amplitude information remained, native listeners were successful in tone identification for all but Tone 1 tokens. From the studies above, we have seen that the primary acou stic and perceptual cue for Mandarin tone identification is fundamental frequency. Besides this most important cue, F 0, turning point, duration, and amplitude all contribute to the perception of Mandarin tones. Non Native Perception and Production of Ch inese and its influence on L2 acquisition of Tones We have seen that native speakers of Mandarin Chinese identify tones using various acoustic cues. Studies using neuroimaging techniques also suggest that Mandarin tones, for native speakers, are lateralize d in the left hemisphere, which suggests that lexical tones are processed as linguistic elements in the same way that other segmental properties are processed (Wang, Jongman, & Sereno, 2001). Will nonnative speakers process tones the same as native tone sp eakers? Will they process tones linguistically or auditorily? How will the similarity and difference influence their learning of Chinese tones? From various studies, it has been shown that since the F 0 and pitch pattern associated with the lexicon is unfam iliar to speakers whose native language is nontonal, tone is very difficult for them to learn (e.g. Shen 1989).

PAGE 32

32 First, tonal errors made by nonnative speakers have been investigated in some studies. For example, in Shen (1989), American learners who had l earned Chinese for four months were tested for production of tones. The results showed that the learners made errors on all tones with Tone 4 having the highest error rate (55.6%). Shen speculated that Tone 4 is more likely to be subject to first language interference, because it is prosodically less marked for English learners. Another study by Miracle (1989) also looked at tonal errors and reported an overall rate of 42.9% of errors by second year American learners of Chinese. He classified the tone error s as tonal register, meaning either too high or too low, or tonal contour errors. What is interesting in that study is that the two types of errors were found evenly distributed among tones and tone errors. Specifically, the register errors for Tone 1 w ere produced when the high level tone s were too low in tonal space; the contour errors for Tone 1 w ere realized by a falling contour instead of a level contour. The register error for Tone 2 was too high in tonal space and the contour error occurred when a le vel or falling contour was produced instead of a correct rising contour. The same register error occurred with Tone 3 and the contour error occurred when it was realize d as a rising tone. The register error for Tone 4 was realized in the mid low register a nd contour error s were mainly level one s instead of falling contours Second ly there seem to be great differences between the pitch range of native Chinese speakers and English speakers, which might influence L2 learning. Chen (1974) conducted an experim ent comparing pitch range between Chinese and English speakers with sentences in both languages. He found that the pitch range of Chinese speakers speaking Chinese was 1.5 times wider than for English speakers speaking

PAGE 33

33 English. Even when English speakers s p oke Chinese, their pitch range increased, but not to the same degree as that of native speakers. Therefore the implication for American English learners of Chinese is to make pitch range much wider. There are also differences in perceiving Chinese tones between native and non native speakers. Fundamental frequency is the primary perceptual cue for native Chinese speakers T he two parameter s of height and contour, however, have been shown to be weighted differently, depending on the linguistic exp erience of listeners (Wang 2006, as tested by listeners of both tonal languages and non tonal languages. The four types of tonal language listeners were Mandarin, Cantonese, Taiwanese, and Thai, and th e nontonal listeners were English. By using multidimensional scaling, Gandour found that English listeners placed much more importance on F 0 height, and less attention was given to the contour, compared to listeners of most of the tonal languages. The auth or speculated that since there are no contrastive tones in English and pitch variation is usually used at sentence level, English speakers only direct their attention to F 0 height. Duration is another perceptual cue that is used by listeners. Tone 3 is us ually longer than Tone 2 i n citation form; however, there seems to be a difference in its role among native and non native speakers. In Change (2011), duration normalized and non normalized stimuli of Tone 2 and Tone 3 were compared in a perceptual experim ent by eight native Chinese and eight non native listeners. Their results showed that for native listeners, the absence of the duration cue did not affect the accuracy of the tone recognition but longer reaction time was needed; on the other hand, non nati ve listeners suffered greatly in their perceptual accuracy as well as reaction time. The

PAGE 34

34 author concluded that syllable duration may facilitate Tone 2 Tone 3 distinction for native speakers and serves as a primary cue besides F 0 for non natives. Another difference lies in the context of tone presentation and can be attributed to first language interference. Broselow et al. (1987) compared the perception of tones presented in isolation and in the context of two or three syllables by American learners of Ma ndarin. The interesting finding from this study is mostly on the nonnative that Tone 4 was the most easily identified tone when presented in isolation, as well as in th e final position of a two/three syllable word. The errors of Tone 4 only became wors e when the tone was presented in non final position. Broselow et al. (1987) discussed the possible rationale for this phenomenon and concluded that it might have resulted f rom the interference of English intonation pattern s In English, falling intonation occurs at the end of a declarative sentence, which is acoustically similar to that of Tone 4 in Mandarin. Th at being the case, it is p ossible to explain why English speaker s identified Tone 4 more accurately in final position by virtue of the fact that they are very familiar with this pattern. In addition, the study showed that Tone 4 was more likely to be misidentified as Tone 1 when in utterance final position. The authors also explained this as due to first language interference. Because both Tone 4 and Tone 1 start with a high pitch, English listeners are more likely to pay attention to the high part and ignore what follo ws as is the case at the end of an English sentenc showed first language influence on tone perception. He found that English listeners tend to identify the high tones as stressed and low tone s like Tone 3, as unstressed I n fact

PAGE 35

3 5 however, stress in Mandarin is not realized fro m fundamental frequency but from duration and amplitude. In conclusion, we have discussed phonetic implementation in production and perception of lexical tone s in tone languages in general and in Mandarin Chinese in particular Native and nonnative percep tion and production of Mandarin Chinese were also compared, with the result that linguistic experience does influence the acquisition of tones, either positively or negatively. Though different patterns o f tone processing have been seen in nonnative speake rs, as well as in hemisphere lateralization (Klein et al., 2001), the performance of tone learning can be improved through training, even after a short period of time (Wang et al., 2000). Moreover, cortical involvement when processing tones c an also be mod ified as proficiency improves (Wang et al., 2000). Categorical Perception and Mandarin Chinese Tones Tones are perceived linguistically by native speakers just like vowels and consonants are This has been supported by a number of studies. One of the earli est studies (Van Lancker & Fromkin, 1973) found in native Thai speakers a right ear advantage in hearing tones in words rather segmental information. Similarly, Wang et al. (2001) found the same patterns in Mandarin Chin ese by comparing the tone errors of dichotically presented tone pairs by Chinese and English native listeners Chinese listeners showed a significant right ear advantage whereas English speakers showed no ear preference. In recent years, studies using n eur oimaging techniques have also yield ed results that support this claim. For instance, Gandour et al. (2000) used PET to examine tone processing in Native Thai speakers and found that they used mainly the left hemisphere to process tones as

PAGE 36

36 other linguistic elements, whereas native English speakers did not show any lateralization If tones are perceived as a linguistic segment, are they perceived categorically? Categorical perception, according to Liberman (1957, 1967), refers to the phenomenon steps along an acoustic continuum will produce perceptible differences when they occur between phonetic categories, but not when they occur within a her colleagues prop that speech sound perception is strongly influenced by category goodness it is more difficult to discriminate a prototype (good exemplars) from its variants than a nonprototype from its va riants. This phenomenon has been widely studied in consonants and vowels in a number of languages. However, very limited work has been done on categorical perception of the suprasegmental feature, tones. One of the early studies by Abramson (1976) examine d the native perception of three Thai tones: the three static low, mid, and high tones. A continuum of 16 level tones with constant F 0 contours was used on the syllable [ k h a :] between 92 and 152 Hz. The results showed that most Thai listeners were able to place the continuum into their corresponding three tone categories. However, their discrimination performance was shown. Halle et al. (2004) speculated that the static tones might be expected to yield lower categoricity and perhaps the perception of dynamic tones is more categorical in nature. Wang (1976) also reported a study on categorical perception in Mandarin Chinese. There w as a n 11 step continuum of the syllable [i], s tarting with a 135Hz level

PAGE 37

37 tone and tones that r o se in a linear fashion from 105 132 Hz to 135Hz. From the perception results, native Mandarin Chinese speakers did show a pattern of categoricity, which was absent for English listeners. Based on the study, Wang (1976) concluded that tone perception in Mandarin Chinese is categorical. However, there were very limited number s in that study with two Chinese participants and three English speakers. Another study by Stagray & Downs (1993) also provided some evid ence for this claim. They present ed a continuum of level tones with pitch variations to Mandarin Chinese listeners and English listeners. It was revealed that Mandarin listeners were not as sensitive as English listeners to small F 0 contour variations when doing a same different discrimination task. Chang & Halle (2000) claimed to have the first study using tone continua for the issue of tone categorization with Taiwan ese Mandarin listeners. Their study reported a gradient in categoricity, in the sense that tones are perceived in a roughly similar degree of categoricity compared to vowels. 1976 in which Zue reported a study on the categorical perception of Mandarin Chinese tones ( Gandour, 1978). The study used a continuum that had nine variants with a possible intended target of between Tone 2 and Tone 3. Those nine tones were superimposed to a synthetic syllable bao with a rising linear fundamental frequency contour of 100 to 160 Hz. What caused these nine tokens to differ was the duration of the portion before the point of ris ing & Lin, 1991), which varied from 0 400 ms with a 50ms step. Interestingly, both Chinese listeners and En glish listeners showed a sharp category boundary at Variant 5 ( in which the point of rising was in the middle of the stimuli) and a peak in discrimination at the

PAGE 38

38 category boundary. This suggest s that both native and nonnative listeners treat Variant 1 4 as similarity between native and nonnative perception in categorizing these variants might be due to the limited number of subjects and tonal stimuli as well. Cao and Sarmah (2007 ) also conducted a study, similarly but on a finer scale, to examine the role of the shape of the pitch contour in the perception of Mandarin Chinese Tone 3 and categorical perception of tones. A set of stimuli was constructed by varying a recorded Tone 3 of syllable ma in two conditions: (1) varying the duration of the dip (or turning point) and (2) varying the timing of the turning point (duration of the slope). T here were thus 40 stimuli total in a continuum with a constant duration of 400ms. Those varia tions of dur ation of dip or timing of the turning point were made incremental at a 10ms step. The manipulated stimuli were presented to native speakers of Mandarin Chinese in two sets: (a) a set of speech stimuli and (b) a set of non speech stimuli. The r e sults showed that to be perceived as Tone 3, the duration of the dip should not be more than 67.5% of the total length of the stimuli. Once the dip is more than 67.5% of the total length of the stimuli, Mandarin Chinese speakers perceive the stimuli as Ton e 1 I n addition, a Tone 3 is perceived if the turning point occurs between 42.5% and 72.5% of the total length of the stimuli. If it occurs before 42.5%, it is most likely to be perceived as a Tone 2 and if it occurs after 72.5% of the total length of the stimuli, the stimuli is perceived as Tone 4. This study provided evidence of a clear categorical perception o f tones by native speakers, though the non speech stimuli that were devoid of consonantal and vocalic information did make it difficult for native speakers in accurate identification and thus they perceiv ed them categorically. Halle et al. (2004) conducted a tone continuum

PAGE 39

39 perception experiment on Taiwanese Mandarin listeners and French listeners. The stimuli consisted of three types of continuum: T one 1 Tone 2, Tone 2 Tone 4, and Tone 3 Tone 4. The Mandarin listeners show ed a categorical perception at the tone boundaries, wh ereas the French listeners showed no increase d sensitivity near category boundaries, suggesting the influence of a language experience effect on speech perception. On the other hand, the performance of nonnatives was not t hat bad, according to the results T he authors speculated th at this was due to the sensitivity to intonation contours by the French listeners; or, according t o the PAM model, tones have fair to good performance depending on the perceived salience. Another issue is the influence of speaker F 0 range. Tones can be perceived using acoustic cues, but in natural speech they are also perceived relative to other tones. Listeners have to pay attention to a speaker's F 0 range in order to distinguish tones that differ only in F 0 height. A few studies have examined the role of the extrinsi c F 0 in tone perception. For instance, Moore and Jongman (1997) investigated speaker normalization in the perception of Tone 2 and Tone 3 in Mandarin Chinese by use of F 0 range as a cue to speaker identity. There were two speakers with different F 0 ranges so that Tone 2 of the low pitched speakers and Tone 3 of the high pitched speakers occurred approximately at the equivalent F 0 height. The three tone co 0, or both, and both attached to a natural precursor phrase from each of the two speakers. The results from the study showed that identification shifted such that the same stimuli were identified as low tones in th e high precursor condition, and as high tones in the low precursor condition. The

PAGE 40

40 study suggested that tone identification is influenced by F 0 change and listeners use that cue as a reference to interpret ambiguous tones. The present study focuses on one of the important perceptual cues for Mandarin tones: turning point, and particularly the timing of the turning point in the tone continuum of Tone 2 and Tone 3. We hope to discover the perceptual boundary of these two tones. In addition, we examine voice q uality (creaky voice) to determine its role in Tone 2 and Tone 3 perception. Creaky Voice and Mandarin Chinese In the production of Mandarin tones, there is one acoustic cue that is of interest here and which has received less attention, namely phonation q uality. Phonation quality, such as creaky voice, breathy voice, and modal voice, has generally been a less studied area in language research. In some languages, there is vowel or consonant phonation contrast; therefore, in those languages, phonation qualit y may be associated with a perception cue. In Mandarin Chinese, these phonation qualities do not form segmental contrasts. Creaky voice is characterized by reduced intensity in waveform, lower fundamental frequency, and less frequent pitch periods with ir regular duration (Gordon & Ladefoged, 2001). According to Keating (2006), creaky voice, or a constricted glottis, is associated with aperiodic glottal pulses. The irregular pitch periods can be visually tical striations reflected pitch & Ladefoged, 2001, p.387). Although creaky voice does not differentiate words at their semantic level, it has been studied in its association with production and percep tion of Mandarin Chinese tones.

PAGE 41

41 Creaky voice has been reported as being associated with Tone 3 and Tone 4 in Mandarin Chinese (Belotel Grenie, A. & Grenie, M., 1994). In that study, production data from 7 native speakers (4 M, 3 F) of 31 words were analyz ed, and the vowel in those words was /a/. The results showed that for all four male speakers, creaky voice was always associated with Tone 3 (8 times for 8 words); for the three female speakers; however, the occurrence of creaky voice was 4/5, 1/8, and 3/6 In the study that the same authors conducted later (1995), similar results were reported based on data from a male speaker reading monosyllabic Chinese words with no initials but all kinds of vowels (low vowel, high front vowels, and high back vowels) we re included. The data from that speaker showed the percentage of words produced with creaky voice to be 45.8% in Tone 3 and 10.5% in Tone 4, with no occurrences of creaky voice in Tone 1 and Tone 2. High front vowels were also more associated with creaky v oice than high back vowels. However, because of the very limited number of participants in these production studies, it is difficult to generalize this result into a solid conclusion. What is the role of creaky voice in the perception of Mandarin Tones t hen? There have been a limited number of studies in this area. In Belotel Grenie, A. & Grenie, M (1997), it was determined that the recognition point was earlier for a creaky Tone 3 than for a non creaky Tone 3. In the experiment reported in that study, fo ur words containing the vowel /a/ and consonant initials -/m/, /n/, /l/, and /d/ were recorded by two speakers (one female and one male). The male speaker produced creaky voice and the female did not (it is not clear, however, whether the creaky voice was present for all or some of the tones reported therein). Those stimuli were subsequently partitioned into 30%, 40% segments, up to 100% of the duration of the tone for the perception experiment. Ten

PAGE 42

42 native speakers participated in the identification task. T he results indicated that a Tone 3 produced with creaky voice is recognized more quickly (at 60% of the duration) than a Tone 3 without creakiness (at 70%). The authors asserted that creaky voice is a secondary cue for Tone 3 perception, based on the resul ts of their study. Creaky voice has also been studied in other tonal languages and dialects. For example, creaky voice in Cantonese has been reported as being associated with the production of the lowest tone (mid s study reported speech data from eight Mandarin speakers and eight Cantonese speakers which showed that 68% of Mandarin Tone 3 was found to have creaky voice, but only 25% overall in Cantonese Tone 4. These results notwithstanding, there are some limita tions in these studies. The number of participants from whom production data was collected, for example, was small. Also, the number of vowels and consonants included in the speech data was limited. Furthermore, the issue of how creaky voice interacts wit h categorical perception of Mandarin tones not only among native speakers but also among non native speakers was not addressed. Therefore, the current study is an attempt to fill these gaps and provide a more comprehensive understanding of the role of crea ky voice in the production and perception of Mandarin tones, specifically, Tone 2 and Tone 3. Determination of Creaky Voice Figure 2 1 is an example of modal, breathy, and creaky voice given in Gordon and Ladefoged (2001, p.390) from voiced vowels in Jala pa Mazatec words. In their paper, Gordon and Ladefoged discuss a number of phonetic properties that are associated with distinguishing creaky voice from breathy phonation and modal voices, including

PAGE 43

43 periodicity, acoustic intensity, spectral tilt, fundament al frequency, formant frequencies, duration, and airflow. Although different languages do not have a uniform measurement for these parameters, some general contrasts in these phonetic properties can be found. In the present study, aural and visual inspec tion of the waveform was the primary method used to determine the presence of creaky voice. There were two raters who listened and inspected the waveforms of the production data. There are other measurements, however, that can also be used to detect creaky voice, such as running a Fast Fourier Transform ( FFT ). According to Gordon and Ladefoged (2001), the spectral tilt (difference between the amplitude of the second harmonic to fundamental frequency, H2 F 0 ) is most steeply positive for creaky vowels (Figur e 2 2.). The snapshots of waveform (Figure 2 3, Figure 2 4) were taken from two native Chinese speakers who participated in the production experiment pronouncing the syllable [ma] with Tone 3 and Tone 2. Creaky voice can be seen in the circled area in the top waveform (Figure 2 3), while the bottom waveform (Figure 2 4) has no creaky voice present. In the FFT spectra of a slice from the vowel [a], there is a positive slope from F 0 to H2 in the top spectrum (Figure 2 5), indicating the presence of creaky v oice, which is absent in the second FFT spectrum (Figure 2 6). If the production data from the study above was conclusive that creaky voice is highly associated with production of T one 3 in Mandarin Chinese, then how important i s i t in perceiving T one 3? Will the absence or presence of creaky voice influence the categorical perception of T one 3 and Tone 2 among native listeners and nonnative listeners? Do native listeners and nonnative listeners pay attention to this cue in the

PAGE 44

44 same way? What is the diffe rence in the relative importance between phonation cues and pitch contour in tonal identity? The goal of the present study is to answer those questions by exploring the categorical perception of Mandarin Tone 2 and Tone 3 and the role of creaky voice in the perception of these two tones, among native and non native speakers. This research was guided by the following questions: Research Question 1 : Is creaky voice associated with a certain tone(s) produced by native Mandarin Chinese speakers ( NC ) ? Resea rch Question 2 : How does the presence of creaky voice affect the perception of tones by native Chinese speakers (NC) and L2 learners of Chinese (NE), respectively? Research Question 3 : What is the perceptual boundary of the Tone 2 Tone 3 continuum perceive d by native Chinese speakers (NC)? How does it differ from L2 learners of Chinese (NE) with different proficiencies? What is the role of creaky voice in their perception? Chapter 3 describes the methodology used in the present study. Results and a discussi on are presented thereafter.

PAGE 45

45 Figure 2 1. An example of modal (top left) breathy (top left) and creaky voice (bottom) given in Gordon and Ladefoged (2001, p.390) from voiced vowels in Jalapa Mazatec words. Figure 2 2. FFT spectra of modal (top le ft), breathy (top right) and creaky (bottom) vowel /a/ in three San Lucas Quizvini Zapotec words (Gordon & Ladefoged, 2001)

PAGE 46

46 Figure 2 3. Waveform of a Tone 3 [ma] produced with creaky voice. Figure 2 4. Waveform of a Tone 2 [ma] produced without crea ky voice. F 0 H2 Figure 2 5. FFT spectra of a slice from the vowel [a] with creaky voice

PAGE 47

47 F 0 H2 Figure 2 6. FFT spectra of a slice from the vowel [a] without creaky voice

PAGE 48

48 CHAPTER 3 METHOD The two major goals of this dissertation are t o investigate the use of creaky voice phonation in native production of Mandarin Chinese tones, and to examine the effect of creaky voice in the categorization of Mandarin tones 2 and 3 among native and non native speakers of Mandarin. The study was guided by the following research questions: Research Question 1 : Is creaky voice associated with a certain tone(s) produced by native Mandarin Chinese speakers ( NC )? Research Question 2 : How does the presence of creaky voice affect the perception of Tones by n ative Chinese speakers (NC) and L2 learners of Chinese (NE), respectively? Research Question 3 : What is the perceptual boundary of the Tone 2 Tone 3 continuum perceived by native Chinese speakers (NC)? How does it differ from L2 learners of Chinese (NE) w ith different proficiencies? What is the role of creaky voice in their perception? The experiments consisted of three parts a tone baseline task, a tone production experiment, and a tone categorization experiment. First, native speakers of Mandarin Chines e and L2 learners of Chinese did a tone baseline task for the purpose of measuring proficiency of Mandarin tone perception and to divide the NE up into two groups Each of the native Chinese participants was then recorded producing utterances with target mono syllabic Chinese words with all four tones and the presence of creaky voice was examined in the production data. In the third task, categorization of tone 2 and tone 3 among both NC and NE was examined. Two sets of tone continuum were created such tha t creaky phonation was present during the dip duration in the set. All other parameters were the same. Both sets of stimuli were randomly presented to both NC and NE for

PAGE 49

49 tone categorization. Categorization differences, particularly for tone 2 and tone 3 among the two groups were examined. In addition, the effect of L2 proficiency on tone categorization among L2 learners of Chinese (NE) was investigated. Participants Group 1 (Native Chinese Spe akers) : Thirty three native speakers of Chinese, age 22 40 (Mean= 28.5, S D =5.3 ), with no reported hearing or speaking problems, participated in the experiment; however, one participant was disqualified as a result of failing the tone baseline task (less tha n 50 % correct) perceptual data were lost due to a technical problem. Therefore, data from 30 participants were included in the analysis Among the thirty native Chinese speakers, fifteen (7 females and 8 males) were born and gre least 18. The other fifteen (9 females and 6 males) were from other areas of China, including ten provinces ( Heilongjiang, Liaoning, Shandong, Hebei, Henan, Xinjiang, Jiangsu, Anhui, Hubei, and Jiangxi ) and one municipalit y ( Shanghai ). All of the Chinese participants could speak standard Mandarin. Among the fifteen non Beijingers, four (two from Jiangsu one from Jiangxi (male), and one from Shanghai ) also spoke a variety of Wu dialect, and the others were from regions where mainly Mandarin is spoken. The thirty native speakers of Chinese were either studying or working at the University of Florida, and had been in the United States for 7 months to 12 years (Mean= 2.9 yrs, S D .= 2.4 ). Among the Chinese participants, eleven of them reported having musical training experience from 1 to 16 years (Mean=2.5 yrs, SD= 4.4) Two are left handed. They were all paid for participating in the study.

PAGE 50

50 Group 2 (English Learners of Chinese) : F orty two native speakers of American English, age 18 23 ( M ean = 20.9, S D = 1.4) 14 females and 28 males, participated in the experiment. All of them except one (first year graduate student) were undergraduate students at the University of Florida. None rep orted any speech or hearing problems. Four indicated left handedness. Seventeen of the English speakers were paid for participation, and the other twenty five received extra credit for their Chinese classes. All the native English speakers had completed, or almost completed, at least two semesters of Chinese at the University of Florida or another institution. Except for one participant, who started learning Chinese with a tutor at age 6 for four years and continued in college from age 18 21 (thus had been excluded from data analysis) all other native English speakers were late learners of Chinese (AoA later than age 16) and the length of their Chinese learning ranged from two semesters to four years. Eighteen of them had study aboard experience in China f or at least three months in Beijing, Shanghai, Chengdu, or Taiwan. There is another speaker who did not pass the baseline task thus was eliminated from the study. Therefore, there are 40 NE speakers (14 Females, 26 M ales) in the analysis of the current stu dy. Among the 4 0 English speakers, 28 had played a musical instrument for 2 18 years (starting age 5 17). One, who did not play any musical instrument, had been in a chorus for one year. Four English participants reported themselves to be left handed. Pro cedure and Stimuli All experiments were conducted at the linguistics laboratory of the University of Florida. Befor e the experiments began, participants were briefed as to what to expect in the study and informed consent forms were signed. A questionnaire was then filled out by each of the participants, including information on their education/language

PAGE 51

51 background and music background. After that, the following three experiments were conducted in the following order: 1) Tone Baseline Task, 2) Production Expe riment, and 3) Categorization Experiment, followed by a debriefing session. The entire experiment took about one hour for each participant. The stimuli used and procedures in each of the tasks are explained below. Tone Baseline Task Stimuli The stimuli u sed in the tone baseline task consisted of 40 mono syllabic words of Mandarin Chinese, with 10 Consonant+Vowel combinations produced with all four tones by two native Chinese speakers from Beijing, one female and one male, resulting in 80 target words tota l. The 40 words were chosen from the first year Chinese textbook ( Integrated Chinese I ) used by the Department of Languages, Literatures, and Cultures of the University of Florida, so that all L2 learners of Chinese should have encountered all the words pr ior to participating in the study. (See Appendix for the complete list of 40 words). Each of the target words appeared at the final position in a frame sentence (in please read this word out loud -_______. the soundproof booth in the L inguistics lab, using a digital recorder ( Marantz PMD660 ) and a head mounted microphone (Shure SM 10A) at a sampling rate of 44.1 kHz The sound files were then transferred to a computer and saved as .WAV files. Each of the 40 targ carrier sentence and saved as 80 individual sound files (.wav) using PRAAT. All tokens were normalized at 98% peak intensity with the UAB software developed by Steve Smith at the Un iversity of Alabama in Birmingham.

PAGE 52

52 Procedure The tone baseline bask was carried out in a quiet room located in the Linguisti cs lab by means of a computer. The presentation of the stimuli in the tone baseline task was controlled by the UAB software. The 8 0 stimuli were presented in two blocks, one for the 40 words read by the female, and the other for the 40 words read by the male. The 40 target words in each block were randomized Presentation of the two blocks of words was counter balanced for all partic ipants. There was a break after the first 40 words. Participants were tested individually. After a participant entered the room and sat in front of the computer, s/he was instructed (in her/his native language) to identify the tone by clicking one of the f our buttons marked word through a pair of headphones connected to the computer. There was no time constraint in this task; that is to say, participants were allowed to take as much time as they needed to make a choice. There was no feedback provided throughout the task, except that a red dot would appear on top of the tone square a participant clicked, as an acknowledgement of a response to a stimulus. Tone Production Experi ment Stimuli Materials used in the tone production task consisted of 40 different monosyllabic Chinese words, with 10 Consonant + V owe l combinations for ea ch of the four tones, with each repeated three times, resulting in 120 words in total. These monosylla bic words were constructed with initials [ m, n, p h p, k h k, t] followed by finals of all possible monophthongal vowels in Mandarin Chinese : front vowel /i/ c entral vowel /a/ and back vowel s / u/ These consonants and vowels were chosen to ensure th at they could be

PAGE 53

53 produced with all four Chinese tones to generate real Chinese words (Table 3 1 ). Notice that these words are different from those used in the baseline task The 120 words were all randomized and each was placed at the final position of a c arrier sentence in To make sure that all the words were pronounced with their correct tones, pinyin was written next to each of the target words. Procedure The tone production experiment was conducted in th e soundproof booth in the linguistics lab at the University of Florida. A head mounted microphone (Shure SM 10A) and a digital recorder (Marantz PMD660) were used to record the production of Chinese words. Each of the participants read the same list of 120 sentences in Chinese with target words at the end of each sentence. The same order was used for every participant. These sound files were transferred to a computer and saved as .WAV files. Each of the target words was examined for the presence of a creaky voice by two raters using audio and visual inspection. Tone Categorization Experiment Stimuli There were two goals in the tone perception (categorization) experiment : one was to examine the categorial boundary in the perception of Chinese Tone2 Tone3 b y native and non native (L2) listeners, and the other was to investigate whether the presence of creaky voice affects the perceptual boundary between the two tones among both native speakers of Chinese and native speakers of American English who are learni ng Chinese. Therefore, two identical sets of tone stimuli were created for this experiment, except that creaky voice was present in one set, but was absent in the other.

PAGE 54

54 The first set of tone stimuli consisted of 34 different tokens with no creaky voice g enerated from a single token of a ) syllable produced by a female native speaker of Mandarin Chinese from the Beijing area. This word was recorded using a head mounted microphone (Shure SM 10A) and a digital recorder (Marantz PMD660) at a sampling frequency of 44.1 kHz and was saved as a .WAV sound file to a computer. An examination of the spectrogram generated by PRAAT software (Paul Boersma & David Weenink) indicated that this word was produced with no creaky voice. In addition, the total duration of this syllable was 460 milliseconds (ms) long with F 0 at syllable onset of 203.8 Hz and offset of 253.5 Hz. All 34 tokens of the non creaky stimuli were created from this naturally produced model [ma] syllable by manipulating pitch contours using the PRAAT software. In t his original iteration of [ma] (Figure 3 1), four pitch heights were observed: syllable onset F 0 = 203.8Hz, onset of vowel [a] F 0 = 203.8Hz (starting at 57ms), minimal F 0 (turning point) = 170.5Hz, and the syllable offset F 0 = 253.5Hz. To generate the 34 t okens of creaky stimuli, another [ma] syllable produced with tone 3 by the same female speaker was used as the model for examining the natural parameters of creakiness. Examination with PRAAT indicated that this utterance was produced with creakiness at th e dip of the Tone3 pitch contour with duration of 68ms. Therefore, the dip of the non creaky tokens in the first group was also set to 68ms, with the minimal observed F 0 of 170.5 Hz from the non creaky [ma] utterance. S timulus group 1 non creaky tokens : For the first group of non creaky tokens, an empty pitch tier (length=460ms) was created in PRAAT for each token, then 5 pitch points were added to each empty pitch tier (Figure 3 2),1) syllable onset (0ms, F 0 =203.8

PAGE 55

55 Hz), 2) starting of vowel (57ms, F 0 =203 .8 Hz), 3) starting point of the dip (starting from 58ms to 388ms, in an increment of 10ms, F 0 =170.5 Hz), 4) ending point of the dip (starting from 126ms up to 456ms, in an increment of 10ms, F 0 =170.5 Hz), and 5) syllable offset (460ms, F 0 =253.5 Hz). There fore, without manipulating the total length of the syllable, the length of the dip was kept constant at 68ms same as in the observed value from naturally produced syllable, only the timing of the turning point (dip) was altered; in other words, the time t hat the turning point (dip) occurred in each token differed by 10ms, resulting in 34 different pitch tiers The modal [ma] token was then synthesized with these 34 different pitch tiers, generating 34 non creaky tone stimuli to be used in the tone percepti on experiment. S timulus group 2: creaky tokens : The second group of perception stimuli consisted of 34 tokens that were identical to the 34 tokens in the first group, except that tokens in this group were manipulated again with PRAAT so that each of them h ad creaky voice at the dip. Just as in the first group, an empty pitch tier (length=460ms) was created in PRAAT for each token, and the same five pitch points were added. In addition, the 68ms dip was manipulated again to create a creaky voice effect. Re call that the prominent characteristics of creaky voice include irregular periodicity and sudden decrease in fundamental frequency and intensity. Therefore, creakiness was generated by randomly adding extremely low and irregular pitch points at a level fa r below the non ure 3 3). The creakiness sounded quite close to a naturally produced creakiness according to two native Chinese speakers Thus the or all creaky

PAGE 56

56 tokens. Each of the pitch tiers was then synthesized with the naturally produced [ma], generating 34 creaky tone tokens for the perception experiment. Altogether, 68 different tokens were created for the tone categorization experiment. T he presentation of these stimuli was again operated with UAB software. They were first normalized at 98% peak intensity with the UAB program. Then the two groups of creaky and non creaky tokens were mixed together and randomized. The stimuli were presented to each participant in three blocks, with two breaks to reduce the fatigue effect. In each block, the mixed 68 tokens were repeated two times (68*2=136 stimuli). The order of the mixed 68 tokens was different in each block. The presentation of the three b locks was counter balanced across the participants in the following way: 123, 132, 213, 231, 321, and 312 thus resulting in six possible orders of presentations. In total, there were 68 tokens*2 repetitions *3 blocks=408 stimuli in the tone perception exp eriment for each participant. Procedure The tone categorization experiment was also conducted in the same room where the tone baseline task was carried out. Participants were individually tested. Each participant sat in front of a computer, wearing headp hones, and was instructed (in their native language) to identify the tone after hearing each stimulus by clicking one of the on each of them. Although responses were expecte d to be mostly Tone 2 and Tone 3, the purpose of having four tone responses was to make the task more natural and to find out whether some stimuli were perceived as other tones. Responses had to be made within 3 seconds; otherwise, the following stimulus w ould be presented. There was no feedback provided throughout the experiment, except that a red dot would Creak y jitter

PAGE 57

57 appear on top of the button to acknowledge the response. The participants were also instructed that they should respond according to their first intuit ion as soon as possible and that they should take two breaks by taking off the earphones and resting for a few minutes. Debriefing After finishing all three tasks tone baseline, tone production, and tone categorization experiments, each of the participa nts was interviewed. The interviews debriefing were to find out whether any of the participants was aware of creaky voice before and during the experiments and what strategies/c riteria were used in the tone categorization task.

PAGE 58

58 Table 3 1 Forty tokens of mono syllabic Chinese words used in tone production experiment T1 T2 T 3 T4 [mi] [ni] [pi] [p h i] [ma] [pa] [ta] [p h u] [k h ] [k ] Figure 3 1. Natural u tterance of [ma] with Tone 3 (non creaky), based on which stimuli in the categorization task were created Fig ure 3 2. An example of an empty pitch tier in PRAAT with five pitch points added to create a non creaky token: 1) syllable onset (0ms, F 0 =203.8 Hz), 2) starting of vowel (57ms, F 0 =203.8 Hz), 3) starting point of the dip ( in this example: 118ms, F 0 =170.5 Hz), 4) ending point of the dip (186ms, F 0 =170.5 Hz), and 5) the syllable offset (460ms, F 0 =253.5 Hz).

PAGE 59

59 Fig ure 3 3 An example of an empty pitch tier in PRAAT with five pitch points added. Then randomly added extremely low and irregular pitch points below the non creaky dip to create a jitter jitter

PAGE 60

60 CHAPTER 4 RESULTS The perception experiments examined two research questions on the catego rical perceptual boundary of the Tone2 Tone3 continuum and the role of creaky voice in boundary shift by native Chinese speakers and L2 learners of Chinese. The production experiment examined the presence of creaky voice with Mandarin Chinese tones, as we ll as its correlation with the perception result. Results from the baseline task are discussed first in order to divide native English speaker participants into subgroups based on their Mandarin proficiency level. Then research question 1 is answered thr ough analysis of production data. Finally, r esearch questions 2 and 3 are then elaborated based on the results from the perception experiments. Research Question 1: Is creaky voice associated with a certain tone(s) produced by native Mandarin Chinese spea kers ( NC ) ? Research Question 2: How does the presence of creaky voice affect the perception of Tones by native Chinese speakers (NC) and L2 learners of Chinese (NE), respectively? Research Question 3: What is the perceptual boundary of the Tone 2 Tone 3 c ontinuum perceived by native Chinese speakers (NC)? How does it differ from L2 learners of Chinese (NE) with different proficiencies? What is the role of creaky voice in their perception? Results of the Baseline Experiment In order to answer the first two questions, the Chinese tone proficiencies of all the participants need to be reported here from the baseline experiment. In the baseline experiment, participants listened to 80 Chinese words in isolation followed by a judgment task of what tone they heard after each token. There were 20 words for each of the four tones. The baseline score of each participant was the percentage of the total number of correct answers divided by 80.

PAGE 61

61 (mean=98.9, SD=1.9). Among the 42 native English speakers, there were two participants who were eliminated from the study: one had an age of acquisition (AoA) of SD lower than the group mean (mean=87.1, SD=10.5). The range of the 40 remaining Since there was noticeable variance among the scores, the native English group needed to be divided into two subgroups: NE Hi gh and NE Low. The two groups were divided such that the number of participants in each group was as equal as possible and that the mean scores between the two groups were significantly different. To meet these two criteria, the best cut off point to divi de the NE group was determined to be 90% In this way, one group had a score range of 91.25% to 100%, which was the same as the native Chinese group, and the numbers in each group were a s close to each other as possible plus the means of the two groups we re significant ly different from each other. The NE High group included 18 participants whose scores ranged from 91.25% to 100% (mean=94.9, SD=2.9). The NE Low group had 22 participants and their scores ranged from 61.25% to 90% (mean=80.7, SD=10.1). T test s were run for the two pairs: NE High vs. NE Low and NE High vs. NC. The r esults showed that the two English groups were significantly different from each other in their mean scores (p=.00), while the NE High and NC were not (p=.90). However, for the purpo se of the current research, it is necessary to regard the NE High as a different group from the native Chinese speakers. Therefore, for all data analyses reported in this chapter, all

PAGE 62

62 participants are considered to be in one of three proficiency groups: NC NE High, and NE Low. Table 4 1 shows the error rates that participants made in each of the tone categories (number of errors in each tone category divided by the total number of the expected number of correct response s ) A O ne Way ANOVA was run for the error rates within each of the groups The results showed that tone category had a significant effect on error rates in both of the NE groups (NE H: [F (3, 68)=5.934, p=.001]; NE L: [F(3, 84)=2.93, p=.038]), but not o n NC ([F(3,116)=1.289, p=.282]). The post hoc test (Bonferroni) for the two NE groups suggested that NE High had the two highest error rates in Tone 2 and Tone 3 (8.1% in both) and the NE L group had the highest error rate in Tone 2 (27.5%) compare d to the other tones category, which had about the same rate (p>.05). In addition to calculating the mean scores for tone proficiency, Pearson Correlation tests were also performed to ascertain tone proficiency scor e s (total) and their age, gender, right hand ed ness, length of Chinese study (# of semesters), length of time studying abroad (# of months), and music experience (# of years). The results showed that none of the above mentioned factors had a ny significant c (p>.001). The same test was also run for Chinese speakers and the results showed that there was no correlation between their tone scores and their age, gender, right hand ed ness, years in the U.S ., whether or not they were from Beijing, nor their music experience (p>.001).

PAGE 63

63 In the following sections, data from the perception and production experiments is presented. The goals of the analysis are to report 1) how creaky voice affects perceptions of t ones among native and non native speakers, 2) what the categorical perceptual boundaries for Tone 2 and Tone 3 are, and what role creaky voice plays in production. Results from t he Production Experiment In the production part of the experiment, all native Chinese participants were recorded reading 120 target words all monosyllabic Chinese words, with 10 Consonant + V owe l combinations for ea ch of the 4 tones, and with each repeated t hree times. The recordings of each target word from each participant were carefully examined for the presence of creaky voice by two raters. The raters listened to each of the target words and made a judgment as to the presence of creaky voice. After the first rating, the two raters compared their ratings. There were a number of words for which the raters had different judgments. A second rating was then performed by both raters until agreement was reached for each target word by both raters. Samples of the stimuli were chosen to be visually inspected. The results agreed with the rating of creakiness. Table 4 2 summarizes the percentages of creakiness detected in all the target words in four tone categories. There were 30 native Chinese speakers and each of them read 120 target words (30 words for each tone), resulting in 900 (30*30) tokens for each tone category. A Chi Square test was run on the numbers of words that were produced with creaky voice, and the results showed there is a strong relationship be tween creakiness and tone category [ X 2 (3, N=1046) =1952.8, p=.00]. As can be

PAGE 64

64 clearly seen, Tone 3 had the most occurrences of creaky voice (97.6%) compared to the other three tones (Tone 1 1.9%, Tone 2 11.6%, and Tone 4 5.2%). This is in accordance with pr evious research suggesting that creaky voice is associated with Mandarin Tone 3. Moreover, the results of the present study suggest that creaky voice is highly associated with Mandarin Tone 3 production. Results from Perception Experiment Results of Ton e Reponses for Each Stimulus P ercentages of four tone responses : There were 34 stimuli (differ ing in their turning point locations, with 6 repetitions for each), 4 types of tone responses, 3 groups (NC=2, NE High=1, NE Low=0), and 2 conditions (Non creak y=0, Creaky=1). The p ercentages of the responses for each tone category were calculated by running crosstab and a Chi square test of the data, and Table 4 3 was generated to show the overall picture from the three groups. Chi Square tests on all four tone responses of the three groups in two conditions were performed. The r esults showed that there was a creaky: X 2 ( 8 N= 14280 ) = 580 p=.00; Creaky: X 2 ( 8 N= 14280 ) = 773.5 p=. 00). In order to ascertain whether those differences in tone response percentages are significant, six Chi Square tests with numbers of tone responses were run to compare Tone 2 vs. Tone 3 response for each of the responses within each group. The results showed that e xcept for the NC Tone 2 vs. Tone 3 response in Creaky condition which did not exhibit a significant difference ([X 2 (2, N=5120) =3.51, p=.06>.01), all other comparison s yielded a significant result: Non creaky condition NC Tone 2 vs. Tone 3: X 2 (1, N=5282) =896.4, p=.00

PAGE 65

65 NE H Tone 2 vs. Tone 3: X 2 (1, N=3010) =603.7, p=.00 NE L Tone 2 vs. Tone 3: X 2 (1, N=3428) =142.9, p=.00 Creaky condition NE H Tone 2 vs. Tone 3: X 2 (1, N=2932) =237.2, p=.00 NE L Tone 2 vs. Tone 3: X 2 (1, N=3250) =287.1, p=.00 The r esults are summarized as follows: 1. In Non creaky condition, all three groups had more Tone 3 responses than Tone 2 responses. Among the three groups, NE Low had the most Tone 2 and the least Tone 3 responses. 2. In Creaky condition, both NE groups had l ess Tone 3 responses than Tone 2, but the NC group had the same for each tone category. Among the three groups, NC had less Tone 2 and more Tone 3 than NEs. 3. Across the two conditions, all three groups had more Tone 2 and less Tone 3 responses when there w as creaky voice present in the stimuli. The pattern seems to be very clear from the results above, namely that the presence of creakiness leads to increased perception of Tone 2 and less perception of Tone 3 regardless of language proficiency. This sugges ts that creakiness is not the primary cue for identification of Tone 3; otherwise, there would have been more Tone 3 responses in the Creaky condition, given that all other factors were the same. When looking at the change in Tone 3 responses across condi tions, it is found that the NC group show ed the least amount of decrease from Non creaky (60.9%) to Creaky (42.9%), when compared to both NE High and NE Low (59.3 % to 28.6%, 46% to 25.4%). This suggests that English speakers are more affected by the prese nce of creaky voice than native Mandarin listeners when it comes to Tone 3 perception. Given the fact that the two NE groups had the same number of Tone 2 response s and Tone 3 response s only in Creaky condition we can surmise that creaky voice makes Engl ish speakers with different Tone proficienc ies behave in a similar manner

PAGE 66

66 Perceptual Boundaries of Tone 2 and Tone 3 In order to investigate the questions of perceptual boundary from Tone 2 to Tone 3 and how creaky voice makes a difference, if any, tw o major statistical analyses were performed. First, the average turning points (ATP) were calculated from those stimuli that received only Tone 2 or Tone 3 responses. This was done in order to determine the actual turning point at which perception shifts f rom Tone 2 to Tone 3 under each condition across the three groups. Second, binary logistic regression tests were run for % category crossover point for each participant. The crossing point s between two groups (NC and NE with two proficiency levels: Low and High) were then compared in both conditions (non creaky and creaky). Before running the two tests, Tone 1 and Tone 4 responses had to be eliminated from the data for two reasons. In the first place, the current research focuses only on Tone 2 and Tone 3. The response buttons designed for the perception experiment, which included all four tone responses, were for the purpose of making the task as natural as possible since there are four to nes in Mandarin Chinese. The other reason for the elimination of the Tone 1 and 4 responses was that for the analysis of binary logistic regression, only data with a binary outcome could be used. In order to determine whether the trials with Tone 1 and To ne 4 responses could be eliminated, a Chi Square test was run on the combined responses of Tone 2+Tone 3 and Tone 1 +Tone 4 among the three groups, in both conditions. The results showed that for all groups and under both conditions, the Tone 2 and Tone 3 responses were significantly greater than the Tone 1 and Tone 4 responses (p=.00). In the following analysis, only stimuli with responses of Tone 2 and Tone 3 are included.

PAGE 67

67 Average Turning Point (ATP) In order to locate the perceptual boundary between T one 2 and Tone 3, the average turning point (turning point refers to the timing of the start of the low dip in F 0 in milliseconds in the current study) was calculated from stimuli that received 100% Tone 2 or Tone 3 responses. This means that what was incl uded were those stimuli that either a) received six Tone 2 or Tone 3 responses out of 6 repetitions or b) received five Tone 2 or Tone 3 responses out of five repetitions when there was a missing response. The turning point across these stimuli was then a veraged for each participant. The ATP of tone 2 responses : In the examination of Tone 2 responses, it was found that besides the expected peak of Tone 2 responses on the stimuli with turning points which occurred around the middle of a syllable, most parti cipants showed a very consistent Tone 2 perception on stimuli for which the turning point started at or later than 348ms (75% of the syllable length). Figure 4 1 is an example of this pattern from a native Chinese participant in Creak y condition (NC24). T he x axis was the timing of the turning point, and the y axis displays the numbers of the tone responses. As can be seen, the number of Tone 2 responses was quite high when turning points occurred earlier in a syllable, while it decreased when Tone 3 respo nses r o se. This is not surprising since Tone 2 is associated with earlier turning points compared to Tone 3. Towards the end, however, there is a very abrupt increase of Tone 2 response s (almost exclusively) on stimuli with turning points occurring after 3 48ms. This phenomenon was found for 80% of NC, 62.5% of NE in Non creaky condition and 100% of NC, plus 80% of NE in Creaky condition. This may be due to the fact that when the turning point occurs very late in a syllable, the ending slope tends to be very sharp.

PAGE 68

68 Given the above the ATP s of Tone 2 responses were divided into two parts. The first part consisted of all stimuli with turning points occurring before 348ms and receiving 100% Tone 2 response, while the second part include d stimuli with turning po ints occurring at or later than 348 ms and receiving 100% Tone 2 responses. T he ATP was then calculated by averaging the turning points from those stimuli in each part. The average turning point s for Tone 2 responses for the NC and the two NE groups (NE H igh, NE Low) are shown in Table 4 4 As can be seen in this table, the data suggests that for all three groups, the average turning points for Tone 2 responses were earlier when creaky voice was absent. A p aired sample T t est was run on the data and the r first average turning point was 10.2 ms earlier in Non creaky (73.9 ms) condition compared to Creaky condition (84.1 ms, p=.003), and there is no difference in the 2 nd ATP for NC (both at 368.6ms). Although descriptively, the data seem to hold true for both the NE High and NE Low groups, the data, nonetheless, show no significant difference s in both of the ATPs within each group going from Non creaky to Creaky. When these two groups are combined (see Table 4 5 ), however, the di fferences (9 ms and 8 ms ) in both ATPs are significant ( [t( 42)= .845 p=.003]; t( 66)= 3.748, p=.00 ). In other words, both of the average turning points of those stimuli that received 100% Tone 2 responses were 9 and 8 ms later when creaky voice was present for Native English speakers. In summary, then, both native and non native speakers of Chinese exhibit later turning points for Tone 2 responses in Creaky condition.

PAGE 69

69 The fact that Tone 2 perception occurs at a later turning point for creaky stimuli sugges ts that the presence of creaky voice may have perceptually shortened the initial falling portion of the tonal contour before it rises (i.e., the turning point) among listeners. The ATP of tone 3 responses : The average turning points for Tone 3 responses fo r the NC and the two NE groups (NE High, NE Low) are shown in Table 4 6 As can be seen in this table, the data suggests that for all three groups, the average turning points for Tone 3 responses are also earlier in the Non creaky condition. A p aired sam ple T t est was run for the ATPs within each group and the results were similar to the results from the ATPs of Tone 2 responses. The NC group showed a significant difference ([t(55)= 2.59, p=.012) in Tone 3 ATPs across conditions 18.2 ms earlier in Non crea ky (193.2ms) than in Creaky (211.4 ms ). Neither of the NE groups showed a significant difference in their Tone 3 ATPs across conditions (p=.43 for NE High; p=.39 for NE Low); however, when these two groups were combined (see Table 4 7 ), the difference (18.4 ms) in their Tone 3 ATPs was significant ( Non Creaky ATP for NE: 175.6 ms; Creaky ATP for NE: 1 94.0 ms. [t(59)= 1.835, p=.02] ). In other words, disregarding tone proficiencies, both native and non native speakers of Chinese show a later turning point for Tone 3 response in Creaky condition. These results, similar to those from the Tone 2 ATPs, again suggest that the perceptual boundary seems to occur at a later point when creaky voice is added. Table 4 8 summarizes the results from this section of average turning points. Notice that the data for NE here is a combined result from the NE High and NE Low groups. The differences in ATPs of Tone 2 and Tone 3 responses across conditions are all significant, and they suggest that the presence of creaky voice does affect Tone 2

PAGE 70

70 and Tone 3 perceptions for both native and non native listeners. In particular, the duration of the initial falling of a tone contour seems to be perceptually shorter when there is creaky voice present. It is unclear why the differences in To ne 2 and Tone 3 ATPs between the two groups (NC and NE) are so close to each other (Tone2: NC 10.2ms, NE 9ms; Tone 3: NC 18.2ms, NE 18.4ms). One interpretation could be that the extent of how much creaky voice shortens perception is similar among both nati ve and non native listeners. The analysis of the average turning point shows how creaky voice plays a role in the perception of Tone 2 and Tone 3 among both native and non native Chinese speakers. Previous research has shown longer duration of a syllable may yield more Tone 3 identification compared to Tone 2 (Blicher et al 1990) and smaller F 0 has often been found with Tone 2 rather than Tone 3 (Shen & Lin, 1991). Since the stimuli in the current study did not vary in their duration, nor F 0 what we h ave surmised from the results ( i.e., that creaky voice makes the turning point occur at a later point) indicates that the presence of creaky voice may have perceptually shortened the initial falling portion of the tonal contour before it rises (i.e., the t urning point) among both native Chinese and English listeners. Figure 4 2 describes the effect of adding creaky voice to the stimuli. The x axis represents the timing of turning point s and the two tone contours could be either Tone 2 or Tone 3 in both con ditions. Regardless of whether the tone is Tone 2 or Tone 3 the responses were triggered with a later turning point when creaky voice was present.

PAGE 71

71 In the following section, a data analysis of binary logistic regression is presented. This test was used to ascertain the predicted 50% crossing point (CP) of Tone 2 and Tone 3 categorial perception among listeners. Binary Logistic Regression for 50% Crossing Point (CP) occurrence when there are two outcomes (e.g. pass/fail, life/death, either ordinal or nominal responses). The purpose of the test is to determine how a response (outcome) changes according to one or more independent variables. For example, if we wanted to find out h ow hours of training affect test results -pass or fail, then the independent variable would be hours of training, and the two outcome values would be pass or fail. For the research reported herein, for each participant, the tone response outcome has two v alues, either Tone 2 or Tone 3 (after the elimination of Tone 1 and Tone 4 responses). Logistic regression works better here because our participants only listened to the same stimulus six times. If the method of drawing lines across the 50% had been used it would have created some uncertain situations. For instance, as shown in the graph s (Figure 4 3 and Figure 4 4), it can be very clearly seen that for participant NC15 (Figure 4 3), in Creaky condition, the 50% crossing point where Tone 2 responses swit ched to Tone 3 was at stimulus 3 and vice versa at stimulus 30. Those two stimuli can then be converted to the millisecond of the turning point: 78ms and 348ms. This means that for this participant, if the turning point of the dip occurs between 17% (78/46 0 ms) and 75.7% (348/460 ms) of the total syllable length, the stimulus is heard as Tone 3. If the turning point occurs before 17% or after 75.7%, it is heard as Tone 2. However, if we look at Figure 4 4, which shows the data table from participant NC21 i n Creaky

PAGE 72

72 condition, it is very difficult to decide where exactly the 50% categorial crossover line is located 3 among stimuli 6 to 8. Therefore, the drawing method cannot solve t he problem for the current study. Using binary logistic regression can allow us to predict where participants would have a crossover point if they heard stimuli an indefinite number of times. Thus, binary logistic regression was run for each participant i n both conditions. Since the 34 stimuli in each condition differ only in their turning points, the timing of those turning points is the predictor of responses. In this case, for the two outcomes possible, for Tone 2 it would be 0 and for Tone 3 it would be 1. Logistic regression then creates a curve for each participant, in each condition, describing the probability that Tone 3 or Tone 2 is heard at any point on the continuum. Recall that when examining the ATP of tone response s in the last section, many participants showed a very abrupt rise in Tone 2 response s with stimuli that have very late turning points (after 348ms) F or the purpose of the binary regression, those stimuli were temporarily excluded here. After each curve was created, the predicted 5 0% crossing point (CP) was then calculated by using the formula x= & Finlay 2009, p. 485). X stands for the timing of turning points (58 368 ms), constant and B value. Figure 4 5 is an example of a binary regression curve. The x axis is the timing of the turning point, and the y axis displays the two outcomes 1 (Tone 3) and 0 (Tone 2). A logit of 0 (i.e. where the y axis is crossed) means 50%, and this is how we calculate where this point is: 0= X; therefore, we have the formula

PAGE 73

73 mentioned above: x= perceived as Tone 3 increases as the location of the turning point increases. When B<0, the probability of the stimulus being perceive d as Tone 3 decreases as the location of the turning point i ncreases (in other words, Tone 2 is more likely to be perceived on stimu li with earlier turning points). To give an example, Table 4 9 shows the results from the logistic regression run on parti cipant NC24 in Non creaky condition. The significant P value (0.003, <.05) suggests that turning point is a strong predictor for this participant in responding with either Tone 2 or Tone 3. The constant ( ) here is X= ( P(y=1) increases or decreases as X increases (see Agresti & Finlay 2009, p.484). When 0, y decreases as X increases. This means that for participant NC24, when there is no creakiness in the stimuli, it is predicted that when the turning point occurs before 12.5% (57.6/460ms) of the total syllable length, the stimulus will most likely be hea rd as a Tone 3; when the turning point occurs after 12.5% of the syllable, it will be heard as Tone 2. Let us now take a look at the same person in the creaky condition. Using the same calculation with the above, the X here = ( 5.1)/.0043=118.6 ms, which is at 25.8% of the syllable. If we compare the predicted CPs in these two conditions for this participant (12.5% in Non creaky; 25.8% in Creaky), the results suggest that by adding creakiness, the predicted crossing point where perception shifts from Tone 2 to Tone 3 seems to come later, as compared to conditions without creakiness. In other words, it can be

PAGE 74

74 inferred that, when there is creaky voice, the actual length of the falling part in the token seems shorter to the ears. The P value for each partic ipant was examined and it was found that not all p values from the binary regression are significant. In Non creaky condition, 90% of NC, 83% of NE High, and 50% of NE Low had significant p value (P<0.05), which means that for those participants, turning point is a predictor of their tone response to a certain stimulus. In Creaky condition, the significant portions are 100% of NC, 72.2% of NE High, and 50% of NE Low. The final average perceptual boundaries (X) were calculated only with those that had a sig nificant p value. As previously shape s 3 increases as the location of the turning point increases, which means the X is the boundary of Ton stimulus being perceived as Tone 3 decreases as the location of the turning point increases, which means the X is the boundary of Tone 3 shifting to Tone 2 perception. Therefore, it is nec essary to analyze the binary result in two different categories 1 0 summarizes the results of Binary logistic regressi on averaged from participants in the three groups (only those using turning points as a predictor of tone response s). In Table 4 1 0 T tests and one way ANOVA were performed to see whether all the differences were significant. The results from one way ANOVA showed that there is no significant difference among the three groups in each of the column s ; therefore, an ave rage is calculated at the bottom of each column in Table 4 1 0 In Non creaky

PAGE 75

75 to Tone 3 occurred at 76.6ms (16.7%) and the rest (38%) of the participants shift ed their p erception from Tone 3 to Tone 2 at the turning point of 361.1ms (78.5%); in Creaky condition, the participants average perceptual boundary of Tone 2 to Tone 3 was at 160.7ms (35%). Within each of the groups, T tests showed that the differences of the Ton e 2 to Tone 3 boundary in Non creaky condition are all significant for each of the groups (NC: p=.00, NE High: p=.039, and NE L: p=.009). In summary, for native and nonnative listeners, the predicted perceptual boundary from Tone 2 to Tone 3 occurs l ater when creaky voice is present (from 76.6ms to 160.7ms). Since some of the stimuli that have turning points later than 348ms were excluded from the binary logistic regression analysis, it is necessary to include them now. In Non creaky condition, we s ee that the predicted perceptual boundary of Tone 3 to Tone 2 for some participants occurs at 361.1ms (78.5%) I t falls right in the middle of 348 ms to 388 ms, as has been shown from the ATP analysis. In addition, creaky voice also made those few particip ants (from all three groups) have more consistent Tone 2 responses when the turning points occur red later than 348ms. This means that for a listener (Chinese or English), if the turning point of the syllable occurs before 76.7 ms (or 16.7% of the syllable length) when there is no creaky voice, it will most likely be perceived as a Tone 2. If the turning point occurs after 76.7 ms (16.7%) and before 361.1ms (78.5%), it will most likely be heard as a Tone 3. When the turning point occurs after 361.1ms (78.5% ), it will be heard as a Tone 2 again. However, when creaky voice is present in the syllable, the perceptual boundary of Tone

PAGE 76

76 2 to Tone 3 is predicted to be at 154.7 ms (or 33.6% of the syllable length). Stimuli with turning point s before that will be hea rd as Tone 2, otherwise as Tone 3. These results clearly show that for both native and non native listeners, the perceptual boundaries at which Tone 2 switches to Tone 3 seem to come later when there is creaky voice, which is consistent with the findings f rom analysis of the average turning points of Tone 2 and Tone 3 responses from the previous section. Binary logistic regression was also run for each participant to see whether the stimulus type i.e. whether there is creakiness or not, is a predictor of Tone 3 responses. stimulus type was the independent variable. Not surprisingly, the results showed a significant result for 90% of the Native Chinese and 85% of the Native English spea kers (p<.05). This means that for 90% of Chinese and 85% of English speakers, the likelihood of a Tone 3 response decre ases with the presence of creaky voice. In other words, the pre sence of creaky voice is a predictor of a smaller number of Tone 3 respons es In Chapter 5 the results from production experiments and perception are reviewed in relation to previous research and findings. Limitations and future direction are also addressed.

PAGE 77

77 Table 4 1. Error rates for each tone category from Baseline task and mean scores Error Rate ( %) #of sub Mean Score (%) Tone 1 Tone2 Tone 3 Tone 4 sig.? NC 30 98.9 0.7 1.8 1.3 0.7 N NE High 18 94.9 2.5 8.1 8.1 1.7 Y NE Low 22 80.7 14.1 27.5 17.7 18 Y NE(combined) 40 87.1 8.9 18.8 13.4 10.6 Y Table 4 2. Percentage of creaky voice present in four tones for NC With Creaky (out of 900) Percentage Tone 1 17 1.90% Tone2 104 11.60% Tone3 878 97.60% Tone4 47 5.20% Table 4 3 Percentages of four tone responses received on all stimuli fr om 3 groups. Non c reaky Creaky T1 (%) T2 (%) T3 (%) T4 (%) T1 (%) T2 (%) T3 (%) T4(%) NC 7.7 25.4 60.9 5.4 9.4 40.7 42.9 6.4 NE H 2.4 22.6 59.3 13.4 3.1 51.3 28.6 14.9 NE L 8.4 30.4 46 13.6 11.3 47 25.4 14.3 Average 6.6 26.2 55.8 10 8.4 45.4 33.7 11.1 Table 4 4 Two Average Turning Point s (ATP) for Tone 2 responses (ms) for three proficiency groups 1st ATP of Tone 2 2nd ATP of Tone 2 Non Creaky Creaky Change Non Creaky Creaky Change NC 73.9 84.1 10.2 368.6 368.6 0 NE High 90.3 101.3 9 359.1 367.4 7.6 NE Low 97.4 105.5 8.1 360.4 368 8.3

PAGE 78

78 Table 4 5 Two Average Turning Point s (ATP) for Tone 2 responses for NC and NE (combined) 1st ATP of Tone 2 2nd ATP of Tone 2 Non Creaky Creaky Change Non Creaky C reaky Change NC 73.9 (16.1%) 84.1 (18.3%) 10.2 (2.2%) 368.6 (80.1%) 368.6 (80.1%) 0 NE 94.2 (20.5%) 103.2 (22.4%) 9 (2%) 359.7 (78.2%) 367.7 (79.9%) 8 (1.7%) Table 4 6. Average Turning Point (ATP) for Tone 3 responses (ms) for three groups Non creaky T3 Creaky T3 NC 193.2 211.4 NE High 178.6 187.3 NE Low 188.5 201.3 Table 4 7. Average Turning Point (ATP) for Tone 3 responses (ms) for NC and NE (combined) ATP of Tone 3 Non Creaky Creaky Change NC 193.2 211.4 18.2ms NE 175.6 194 18.4ms Table 4 8 NC and Combined NE of their Tone 2 and Tone 3 ATPs (ms) 1st ATP of Tone 2 2nd ATP of Tone 2 ATP of Tone 3 Non Creaky Creaky Change Non Creaky Creaky Change Non Creaky Creaky Change NC 73.9 (16.1%) 84.1 (18.3%) 10.2 (2.2%) 368.6 (80.1%) 368.6 (80.1%) 0 193.2 (42%) 211.4 (46%) 18.2 (4%) NE 94.2 (20.5%) 103.2 (22.4%) 9 (2%) 359.7 (78.2%) 367.7 (79.9%) 8 (1.7%) 175.6 (38.2) 194 (42.2 ) 18.4 (4%) Table 4 9 SPSS output table of binary logistic regression on NC 24 (Non creaky) Variables in the Equation B S.E. Wald df Sig. Exp(B) Step 1 a TurningPoint 0.054 0.108 8.768 1 0.003 1.056 Constant 3.108 1.521 4.177 1 0.041 0.045 a. Variable(s) entered on step 1: TurningPoint

PAGE 79

79 Table 4 10. Results of Binary l ogistic regression for three groups (predicted crossing point of perceptual boundary of Tone 2 and Tone 3). Non creaky Creaky (B>0) (B<0) (B>0) (B<0) NC (N=30) 79.4ms (17%) (N=21) 362.5ms (18.8 %) (N=6) 143.9ms (31.3%) (N=30) (N=0) (90% sig.) (100% sig.) NE High (N=18) 63.9ms (13.9%) (N=8) 344.5ms (75%) (N=7) 160ms (34.8%) (N=13) (N=0) (83% sig.) (72.2% sig.) NE Low (N=22) 84.4ms (18.4%) (N=4) 376.5ms (82%) (N =7) 178.2ms (38.7%) (N=11) (N=0) (50% sig.) (50% sig.) Average 76.7ms (16.7%) 361.1ms (78.5%) 160.7ms (35%) Figure 4 1. An example of having Tone 2 responses again after 348 ms (from NC24, in Creaky condition).

PAGE 80

80 Figure 4 2. Illustr ation of effect of creaky voice on tone responses. Figure 4 3. NC15 Tone 2/3 responses in Creaky condition Figure 4 4. NC21 Tone 2/3 responses in Creaky condition Figure 4 5. An example of probability curve of a binary regression analysis. (B> 0)

PAGE 81

81 CHAPTER 5 DISCUSSION AND CONCLUSION The goal of this research has been to explore the perception of Mandarin Chinese Tone 2 and Tone 3 and the role of creaky voice among native speakers of Chinese (NC) and native speakers of English (NE) who are also Chinese language learners. In order to take L2 language proficiency into consideration, the non native speakers were divided into two subgroups: NE High and NE Low. Thus, the participants were all placed in three groups: 30 native Chinese (NC) speakers, 18 high proficiency native English speakers (NE High), and 22 low proficiency native English speakers (NE Low). This chapter discusses the results of the current study and provides interpretations responding to each of the three research questions. Finall y, the limitations inherent in a study like this are addressed and the future direction of research is suggested. Research Question 1 Is creaky voice associated with a certain tone(s) produced by native Mandarin Chinese speakers ( NC ) ? In the present study, production data from 30 native Chinese speakers (16 females and 14 males) w ere collected. The number of participants exceeded that of studies conducted previously (e.g. Belotel Grenie, A & Grenie, M. 1994, 1995, Yu 2010). Each of the recordings included 1 20 words with 40 different words, with each word repeated three times. In the 40 target words, six initials [ m, n, p h p, k h k, t] and four vowels (front vowel /i/ c entral vowel /a/ and back vowel s / u/) were used and there were 10 words for each tone category. Recordings of all the target words (120*30=3600) were examined for the presence of creakiness. The results confirmed that creaky voice is extremely common in the production of Mandarin Tone 3 (97.6%). This supports previous research that creaky

PAGE 82

82 voice is indeed associated with Tone 3, but we observed a much higher frequency in the present study. This is probably due to the fact that the number of the participants in the current study was much greater than previous studies and that more combinatio ns of initials and finals were included in the target words. In addition, we determined that the second highest tone category that has creaky voice present is Tone 2 11.6%. This is different from the results of other studies such as those reported in Greni study in which the second highest category was Tone 4 (10.5%). However, due to the would seem to have more validity. Research Question 2 How does the presence of creaky voice affect the perception of Tones by native Chinese speakers (NC) and L2 learners of Chinese (NE), respectively? To answer this question, the percentages of different tone responses received for each stimulus from the three groups in each condit ion were calculated. The results (see Table 4 4 Chapter 4) show that when there is no creaky voice, all three of the participating groups (NC, NE High, NE Low) had more Tone 3 than Tone 2 responses. On the other hand, when creaky voice is present, all thr ee groups showed an increase in their Tone 2 responses and a decrease in their Tone 3 responses. This change was particularly noticeable in the two NE groups in that they had more Tone 2 responses than Tone 3 in Creaky condition At the same time, the res ponses in the two categories were the same in the NC group. When we look at the differences across conditions, within each of the three groups, the presence of creakiness resulted in more Tone 2 and fewer Tone 3 responses. This holds true for all native a nd non native speakers and may indicate that

PAGE 83

83 creaky voice is not a primary cue for identification of Tone 3 for either native or non native listeners. If it were a primary cue then the presence of creaky voice would have led to more Tone 3 response s Whi ch group was affected the most by the presence of creaky voice? If we examine the percentages of tone response change s from Non creaky to C reaky condition we can see that the NE high group is affected the most by the presence of creaky voice, resulting i n more Tone 2 (+12 7 %) and less Tone 3 ( 30.7%) responses, and NC is the least affected (Tone 2: 15.3% increase; Tone 3: 18% decrease). This suggests that native English speakers are more affected by creaky voice than native Mandarin speakers in their perc eption of Tone 3. In particular, it appears that native English listeners do not use creaky voice to facilitate their identification of Tone 3 This is, in fact, somewhat different from what previous research might have predicted. In (1997 ) study ten native Chinese speakers participated in a perception test with partial ly presented words with Tone 3 with a nd without creaky voice. The results showed that a Tone 3 produced with creaky voice was more quickly recognized (beginning at 60% of the durat ion of a syllable) than a Tone 3 without creakiness (70% of the duration). Their study suggested that creaky voice is a secondary indicator of Tone 3. It would thus be a reasonable prediction that creaky voice would facilitate tone 3 identification; howeve r, this is not supported by the results from the current research. Several reasons may contribute to this discrepancy. First, the methodology in the current study is different from that in the previous study. In the current study, the stimuli consisted of complete sy llables when they were presented and they differ ed not only in

PAGE 84

84 their creakiness but also in the timing of the turning points. On the other hand, the study (1997) were partially presented to see how fast an accurate Tone 3 rec ognition occurs. The two experimental designs were different from each other because their research goals were not the same. The current study aims to determine whether stimuli with creaky voice will generate more Tone 3 responses while the other study att empted to find out the earliest recognition point before a co mplete syllable unfolds itself. Second ly the stimuli from the previous study were from naturally recorded pro duction from native speakers. Th erefore for each stimulus (either partially or fully I n the current study, by contrast, all the stimuli were manipulated in their timing of turning poin t and presence of creaky voice. Thus, there were no right or wrong answer s for each tone respons e. It appears that there might have been some interaction between the effect of phonation type and tone contour such that when the tone contour is fully presented with among non nati tested only native Mandarin speakers; therefore, whether creakiness would make the Tone 3 recognition point early remains unknown. Jongman & native Mandarin listeners could use their language background to help in distinguishing tone contrast s that had variations in F 0 and speaker rate, but native English listeners still used only acoustic variation as a cue to discriminate phonemic contrast. These results seem to be supported by the c urrent study in that the creaky voice was not used as a helpful cue but rather took on the role of a limited perceptual resource for listeners.

PAGE 85

85 If creaky voice is indeed a primary perceptual cue, then it should make the Tone 3 identification easier with t his extra cue and produce more Tone 3 responses. Contrary to that expectation however, the presence of creaky voice actually reduced the total number of Tone 3 responses and more Tone 2 responses were generated. Does this mean creaky voice is a redundant cue in Tone 3 perception? Does it mean that adding creaky voice will only confuse listeners? Although this may appear to be the case, this phenomenon is not as easily explicable as it may seem. To understand the role of creaky voice, we need to take a clos er look at the perceptual boundary of the tone continuum in the next question. Research Question 3 What is the perceptual boundary of the Tone 2 Tone 3 continuum perceived by native Chinese speakers (NC)? How does it differ from L2 learners of Chinese (NE) with different proficiencies? What is the role of cr eaky voice in their perception? From the results obtained from calculating Average Turning Points (ATP) for both tone responses, we found that for both native and non native speakers, the average locatio n of the turning point is later when creaky voice is present. This holds true for b oth Tone 2 and Tone 3 responses. There was no significant difference in the two native English groups; therefore, their L2 proficiencies do not seem to make a difference in the average location of the turning points for Tone 2 and Tone 3 perception. In addition, the current study also found that when the turning point occurs late in a stimulus (on or after 348ms in the current experiment), the stimulus will be identified as a robust Tone 2 from most listeners, regardless of their native languages. Since the dip in all stimuli in the current research was kept at 68ms, the real rising of those tone

PAGE 86

86 contour s actually occurred at 416 ms (348+68), which was about 90% of the duratio n. This steep rising appears to override all other parameters in a stimulus, including the presence of creakiness. This is also different from the findings of a previous study conducted by Cao & Sarmah (2007), in which stimuli with turning points later tha n 72.5% of the syllable were identified as Tone 4. An explanation of this would be the different parameters used in these two studies. In the 2007 study, the F 0 s of stimulus onset, turning point, and offset were 186Hz, 163 Hz, and 211Hz, and the syllable d uration was 400ms. I n the current study, however, the corresponding F 0 s are 203Hz, 170.5Hz, and 253 Hz, with the duration of syllable at 460ms. In fact, the angle at which the rising occurred in the present study was much steeper compared to the previous o ne. Therefore, the results do not contradict the previous finding in that study. These data were calculated from the tokens that received 100% either Tone 2 or Tone 3 responses. Thus, these average turning points indicate the average value of the location of turning point from those stimuli that are perceived as a certain tone. For instance, the ATP for native Chinese speakers in Non creaky condition is 16.1%. This means that when there is no creaky voice, a typical stimulus that will be perceived as a Tone 2 is the one that has its turning point at 16.1% of the syllable. H owever, if creaky voice is present, the turning point of this typical stimulus will move to 18.3% of the syllable. This change for a Native English listener would be from 20.5% to 22.4%. Similarly, when creaky voice is absent, a typical stimulus that will be heard as a Tone 3 by Native Chinese listeners is the one with a turning point at 42% W hen there is creakiness, the turning point shifts to 46%. This shifting pattern for Native Englis h listeners would be from 38.2% to 42.2%.

PAGE 87

87 To summarize, the presence of creaky voice makes the average turning point appear later regardless of the listeners In other words, creaky voice makes the duration of the initial falling of a tone contour seem to be perceptually shorter to both native and non native ears. This finding is particularly enlightening because none of the previous studies we have seen appear to make similar discoveries. To predict the perceptual boundary of the Tone 2 Tone 3 continu um, Binary Logistic Regression was used in the present study. This method of calculating is different from that use d previously in similar research (e.g. Cao & Sarmah 2007, Yang 2011). It is, in fact, a better approach. In perception experiments, the same stimulus is usually repeated only a limited number of times (6 times in t he present study). As shown in C hapter 4 ( figure 4 4 ), using the traditional 50% categorical crossover line to see the boundary might not be applicable to all participants due to the limited number of responses to one stimulus. Binary logistic regression is a more suitable method here because it provides us with a predicted value of that 50% categorical crossover line if there is an indefinite number of repetitions of a stimulus and a n indefinite number of responses. This is more accurate and better applies to our data. The results from binary logistic regression show that for both native and non native listeners, the perceptual boundar y of Tone 2 Tone 3 shifts to a later point when cr eaky voice is added. There is also no significant difference between native and non native speakers on those boundaries. When crea ky voice is absent, the boundary of shifting Tone 2 to Tone 3 perception happens at 16.7% of the syllable When creaky voice i s added, the perceptual boundary shifts to 35% of the syllable.

PAGE 88

88 The results here seem to be different from the previous study (Cao & Sarmah, 2007) that the crossing point of the Tone 2 Tone3 perception happened at 42.5% of the syllable. This can be attri buted to the following factors First, the parameters of the stimuli are different in these two experiments. As previously mentioned, the shape of the tone contours in the stimuli used in the two studies are not the same, which would result in different an gle s of rising/falling contours even when the turning points are the same. Second ly the methods for calculating perceptual boundaries are very different in these studies. The previous study used the traditional method of drawing a 50% crossing point, whic h could only represent the actual value collected from those participants listening to limited numbers of stimuli from a certain study T he current research on the other hand, utilized a better approach binary logistic regression -to calculate the predic ted boundary as if participants were listening to an unlimited number of stimuli. In summary, the role of creaky voice in Mandarin Tone 2 and Tone 3 perception is neither critical nor redundant. Creaky voice is not a primary cue when it comes to Tone 2 and Tone 3 distinction because its presence does not predict more Tone 3 responses. However, creaky voice is not a redundant perceptual cue because it does affect the perception of certain tone contours when the turning point falls within a certain range. Non native listeners seem to be more influenced by this perceptual cue. In addition, the presence of creaky voice seems to shorten the perceptual duration of the initial falling of a syllable in both native and non native listeners. The psychoacoustic effect of creaky voice has not been studied intensively in previous research; however, it has been suggested that non modal vowels are associated with longer duration phonetically compared to their modal counterpart (Gordon 1998), such

PAGE 89

89 as in Kedang and Jalapa Ma zatec. This may, perhaps, help to explain the perceptual shortening found in the current research. If in some languages, a creaky vowel has to be longer than a clear vowel, it implies that listeners perceive the vowel with creakiness as the same vowel with out it. Therefore, the fact that the turning point occurs at a later time when creaky voice is present does not see m surprising at all. To ascertain whether this is also true with for Mandarin tone production, a sample of Tone 3 tokens was taken from the in the current study. Two sets of words were chosen produced with and without creaky voice. They were compared in terms of their syllable duration and timing of the turning points. The results did not conform to previous r esearch. The durations of the words produced with creaky voice tend ed to be shorter (413ms) than those that are clear (567ms, p=.00) The average timing of the turning points in those creaky words occurred at 50% of the syllable, later than that in clear w ords (40.5%, p=.00) This was also different from what we found in the perception experiment in the current study, where the falling part of a tone contour seems shorter when creaky voice is added. It has been established that perception precedes productio n and phonetic perception is not necessarily equivalent to phonological interpretation (Dinnsen 1985) The discrepancy found in the current study provides additional evidence for this. In addition, in future research, it would be more beneficial to examine their perception. The current study also has some implications for the speech mod e ls that were reviewed in Chapter 2. Those models can be extended to account for the suprasegmental feature s and phonation types as explored in the current research. For

PAGE 90

90 example, the Speech Learning Mod e l predicts that when an L2 sound is very different from the closest L1 sound, a new categor y will likely to be established. According to the SLM, if native Engl ish speakers can perceive how Mandarin Tone 3 is different from Tone 2, it is very likely that the two different tone categories will be built in their L2 phonological inventory. However, previous research (e.g. Wang 2006) and the current study both sugges t that L2 learners of Mandarin tones do not use perceptual cues in the same way that native Chinese speakers do. The fact that t he L2 speakers in the current study are more affected by the pre sence of creaky voice suggests they do not use pitch contour as the primary cue for tone recognition, or to the same extent as native speakers. At least with their current L2 proficiency levels, creaky voice seems to have a negative influence on their ability to fully perceive pitch contour. One of the predictions by the Perceptual Assimilation Model is that a non native segment could be assimilated to the native category if it is heard as a good, acceptable, or deviant exemplar from the native category. tone category. The closest thing that tone can be related to is the intonation that is used at sentence level. Lexical tones would fall out of their native phonological space and thus they are predicted to pose potential difficulty for English speakers. I n this case, degree of perceptual difficulty is predicted by PAM to depend on the phonetic saliency between the two tones. As both SLM and PAM agree L2 listeners will continue to refine their perception, and when their L2 proficiency is more native like, the differences between Mandarin Tone 2 a nd Tone 3 will be more salient. PAM would then predict that an L2 learner will perceive s tones with a moderate to good level eventually The Native Language Magnet model predicts that if an L2 sound is very close t o a native language

PAGE 91

91 prototype, it will be difficult for L2 learners to perceive the phonetic contrast. Si nce our L2 learners do not have lexical tone prototypes in their native language, it is difficult for them to extract pitch contour information to perc eive tones. Depending on the psycho acoustic distances between tones, the N LM would predict that Tone 2 and Tone 3 are difficult to distinguish, since they are acoustically very similar to each other. Tones have to be mapped differently from their current phonological space in order to be perceived in the same way as native speakers of Mandarin. Conclusions, Limitations, and Future Directions The present study explored Mandarin Tone 2 and Tone 3 perception among native and non native speakers and the role of creaky voice in their perception of these tones. The location of the turning points that mark the perceptual boundaries between Tone 2 and Tone 3 are 16.7% (no creaky) and 35 % (with creaky) for native Chinese and English listeners. When the turning poi nt occurs after 90% of the syllable, perception reverts to Tone 2. Creaky voice is not a primary cue for Tone 3 perception (although it is frequently found in production of Tone 3) but it is not redundant. It shortens the perceptual length of the initial falling of a tone contour, making the perceptual boundary shift earlier for both native and non native listeners. There are some limitations in the present study that are relevant for future research. In the perception experiment, for example, stimuli wit h only vowel /a/ are included. Future studies could consider having more vowels in the perceptual stimuli to see if they affect perceptual boundaries. In addition, a secondary confirmation of the creaky voice inspection in the production data, such as exam ining the FFT of each word, would allow for stronger claims.

PAGE 92

92 In conclusion, our study not only explored the perception of Mandarin Tone 2 and Tone 3 to a finer extent, but also examined the role of creaky voice in the perception of tones among both native and non native listeners. The results provided some interesting findings in the psychoacoustic domain showing that voice quality could interact with perception of certain sounds. It is hoped that the research reported herein contributes to a better unders tanding of the perceptual cues used in determining Mandarin tones for both native speakers and L2 learners. With regard to the teaching of Chinese to L2 learners, perhaps the findings here will be of use in the development of teaching plans or chapters in Chinese L2 textbooks that focus on how to listen more effectively for Chinese tones.

PAGE 93

93 APPENDIX FORTY WORDS USED IN TONE BASELIN E TASK (WITH PINYIN +TONE MARKS ) ba1 bei1 che1 chi1 dong1 duo1 fen1 he1 jia1 kai1 lai2 men2 qian2 ren2 shi2 tai2 wan2 xue2 shei2 bie2 bi3 wo3 da3 dong3 gei3 hao3 ji3 ke3 li3 qing3 si4 wen4 xia4 xing4 you4 zai4 zhe4 ke4 jin4 hui4

PAGE 94

94 LIST OF REFERENCES Abramson, A. (1962). The vowels and tones of Standard Thai: Acoustical measurements and experiments. International Journal of American Linguistics, 28 (2). Bloomington: Indiana University Research Center in Anthropol ogy, Folklore and Linguistics. Abramson, A. (1975). The tones of central Thai: Some perceptual experiments. In J.G. Harris & J. Chamberlain (Eds). Studies in T h ai linguistics. Bangkok: Central Institute of English Language. pp. 1 16. Abramson, A. (1976). Thai tones as reference system. In T. Gething, J. Harris, and P. Kullavanijaya (Eds.), Tai linguistics in honor of Fang Kuei Li Bangkok: Chulalongkorn University Press, pp, 1 12. Abramson, A. (1978). Static and dynamic acoustic cues in distinctive tones. Language and Speech 21 : 319 25. Belotel Grenie, A. & Grenie, M.. (1994). Phonation types analysis in Standard Chinese. Japan, pp. 343 346. Belotel Grenie, A. & Grenie, M. (1995). Consonants and Vowels Influence on Phonati on types in isolated words in Standard Chinese Proceedings of XIIIth ICPhS, Stockholm, Sweden, pp. 400 403. Belotel Grenie, A. & Grenie, M.. (1997). Types de phonation et tons en chinois standard. Cahiers de Linguistique Asie Orientale, 26(2): 249 279. 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 York Press: Baltimore, pp. 171 203. Best, C. T. and Tyler, M. D. (2007) Nonna tive and second language speech perception: commonalities and complementarities. In the Language Experience in Second Language Speech Learning: In honor of James Emil Flege ( edited by Bohn O S and Munro, M ), pp. 213 234. Blicher, D. L., Diehl, R., and Coh en, L. B. ( 1990 ) Effects of sylla ble duration on the perception of the Mandarin Tone 2/Tone 3 distinction: evidence of auditory enhancement. Journal of Phonetics 18, 37 49. Broselow, E. Hurtig, R.R. and Ringen, C. (1987). The perception of second languag e prosody In G. Ioup and S.H. Weinberger (eds.). Inter language Phonology, the Acquisition of Second language Sound System. Cambridge: Newbury House Publishers, pp. 350 361.

PAGE 95

95 Burnham, D. & Mattock, K. (2007) The perception of tones and phones. In the Lan guage Experience in Second Language Speech Learning: In honor of James Emil Flege ( edited by Bohn O S and Munro,M ), pp. 259 279. Cao, R. & Sarmah, P. (2007). A Perception Study on the Third Tone in Mandarin Chinese UTA Working Papers in Linguistics (2007 ), 2, 50 66. Chao, Y. R. (1948). Mandarin primer. Cambridge: Harvard University Press. Chen, G. T. (1974). The pitch range of English and Chinese speakers. Journal of Chinese Linguistics 2, 159 171. Chuang et al. (1972). The acoustic features and perce ptual cues of the four tones of standar colloquial Chinese In Proceedings of the 7 th International Congress of Acoustics (Vol. 3). Budapest: Akademial Kiado. pp. 297 300. Chang, Y.S. (2011). Distinction between Mandarin Tone 2 and Tone 3 for L1 and L2 lis teners. Proceedings of the 23 rd North American Conference on Chinese Linguistics (NACCL 23), 1 84 96. Davison, D.S. (1986). An acoustic study of so called creaky voice in Tianjin Mandarin. Working Papers in Phonetics, UCLA, 78, 50 57. Dinnsen, D. A. (19 85). A re examination of phonological neutralization. Journal of Linguistics, 21, 265 279. Flege J. & Eetfing W. (1987). The production and perception of English stops by Spanish speakers of English. Journal of Phonetics 15: 67 83. Flege J. & Eetfing W. ( 1988). Imitation of a VOT continuum by native speakers of English and Spanish: Evidence for phonetic category information. Journal of the Acoustic Society of America 83 : 729 740. Flege, J.E. (1995). Second language speech learning: Theory, Findings, and Problems. In W. Strange (Ed.) Speech Perception and Linguistic Experience: Issues in Cross Language research York Press: Baltimore, pp. 233 277. Flege, J. E. (2002). Interaction be tween the native and second language phonetic systems. In An integrated view of language development. (edited by Burmeister, P. et al.) pp. 217 243. Fork, C.Y. Y. (1974). A perceptual study of tones in Cantonese. Centre of Asian Studies, University of Hon g Kong. Fromkin, V.A. (1978). Tone: a linguistic survey. NY: Academic Press. Inc. Fu, Q.J. and Zeng, F.G.(2000). Identification of temporal envelop cues in Chinese tone recognition. Asia Pacific Journal of Speech, Language and Hearing 5 45 57.

PAGE 96

96 Gandour, J. (1978). The perception of tone. In Tone: a linguistic survey (Fromkin, 1978) NY: Academic Press. Inc. Gandour, J. T. (1983). Tone dissimilarity judgments by Chinese listeners. Journal of Chinese Linguistics, 12, 235 261. Gandour, J. et al. (2000). A cross linguistic PET study of tone perception. Journal of Cognitive Neuroscience, 12(1), 207 222. Gordon, Matthew. ( 1998 ) The phonetics and phonology of non modal vowels: a cross linguistic perspective. Berkeley Linguistics Society 24. 93 105 Hombert, J. M. (1976). Consonant types, vowel weight and tone in Yoruba. UCLA Working Papers in Phonetics, 33, pp. 40 54. Joogman, A. and Moore, C. (2000). The role of language experience in speaker and rate normalization process. Proceedings of the 6 th international conference on Spoken Language Processing, I. 62 65. Jongman, A., Wang, Y., Moore, C.B., & Sereno, J. A. (2007). Perception and production of Mandarin Chinese tones. In the Language Experience in Second Language Speech Learning: In honor of James Emil Fleg e ( edited by Bohn O S and Munro,M ), pp. 209 217. Halle, P.A, Chang, Y.C, Best, C.T. (2003) Identification and discrimination of Mandarin Chinese tones by Mandarin Chinese vs. French listeners Journal of Phonetics 32, 395 421. Halle, P.A, Chang, Y.C, Best, C.T. (2004) Identification and discrimination of Mandarin Chinese tones by Mandarin Chinese vs. French listeners Journal of Phonetics Volume 32, Issue 3, July 2004, pp.395 421. Keating, P.A. & Esposito, C. (2006) Linguistic Voice Quality. UCLA Working Papers in Phonetics, No. 105 85 91. Klatt, D. (1973). Discrimination of fundamental frequency contours in synthe tic speech duplications for models of pitch perception Journal of the Acoustical Society of America 53 8 16. Klein, D. et al. (2001). A cross linguistic PET study of tone perception in Mandarin Chinese and English speakers. NeuroImage 13 646 653. Kuhl P. (1991). Human for the prototypes of speech categories, monkeys do not. Perception and Psychophysics, 50, 93 107. Kuhl, Patricia K. (1992). "Infants' perception and representation of speech: d evelopment of a new theory", In ICSLP 1992, 449 456.

PAGE 97

97 Kuhl, P. K. and Iverson, P. (1994). Mapping the perceptual magnet effect for speech using signal detection theory and multidimensional scaling. Acoustic Society of America, 553 562. Massaro, D.W. et al. (1985). The evaluation and integration of pitch height and pitch contour in lexical tone perception in Mandarin Chinese. Journal of Chinese Linguistics 13, 267 289. Miracle, W.C. (1989). Tone production of American students of Chinese: A preliminary acou stic study. Journal of Chinese language teachers association 24, 49 65. Moore, C.B., and Jongman, A. (1997). Speaker normalization in the perception of Mandarin Chinese tones. Journal of the Acoustical Society of America 102, 1864 1877. Liberman, A.M. (1 957). Some results of research on speech perception. Journal of the Acoustical Society of America 29, 117 123. Liberman, A.M. Cooper, F.S., Shankweiler, D. P., & Studdert Kennedy, M. (1967). Perception of the speech code. Psychological Review, 74, 431 46 1. Ohala, J. J. (1978). Production of tone. In Tone: a linguistic survey (Fromkin, 1978) NY: Academic Press. Inc. Sebastian Galles N. (2005). Cross language speech perception. In the Handbook of Speech Perception (Pisoni, D. & Remez, R. editors) pp. 54 6 566. London: Blackwell Publishing. Shen X. S. (1989). Toward a register approach in teaching Mandarin tones. Journal of Chinese Language Teachers Association 24, 27 47. Shen, X., and M. Lin. ( 1991 ) A perceptual study of Mandarin Tones 2 and 3. Langua ge Speech 34, 145 156. Strange, W. (2007). Cross language phonetic similarity of vowels: theoretical and methodological issues. In the Language Experience in Second Language Speech Learning: In honor of James Emil Flege ( edited by Bohn O S and Munro,M ) Stagray, J. & Downs, D. (1993). Differential sensitivity for frequency among speakers of a tone and nontone language. Journal of Chinese Linguistics 143 163. Evide nce from Thai. Journal of Phonetics, 1, 101 109. Vance, T.J. (1977). Tonal distinctions in Cantonese. Phonetica 34, 93 107. Wang, W. S. Y. (1976). Language change. Annals of the New York Academy of Sciences, 28, 61 72.

PAGE 98

98 Wang, Y., Sereno, J. A., Jongman A., and Hirsch, J. (2000) Coritcal reorganization associated with the acquisition of Mandarin tones by American learners: An fMRI study. Proceedings of the 6 th International conference on Spoken Language Processing II. 511 514. Wang, Y., Jongman, A. and Sereno, J. A. (2006). L2 acquisition and Processing of Mandarin Tone. In Li,P. et al. (eds.) Handbook of Chinese Psycholinguistics. Cambridge University Press. Wang, Y., Jongman, A., and Sereno, J.A. (2001). Dichotic perception of Mandarin tones by Chine se and American listeners. Brain and Language 78, pp.332 348. Wayland, P. R. (2007). The relationship between identification and discrimination in cross language perception. In the Language Experience in Second Language Speech Learning: In honor of James E mil Flege ( edited by Bohn O S and Munro,M ), pp. 201 217. Whalen, D. H. and Y. Xu. (1992). Information for Mandarin tones in the amplitude contour and in brief segments. Phonetica, 49 25 47. White, C. M. (1981). Tonal perception errors and interference fr om English intonation. Journal of Chinese Language Teachers Association 16, 27 56. Yang, R. X. (2011). The phonation factor in the categorical perception of Mandarin Tones. ICPhS XVII 2204 2207. Yip, M. (2002). Tone Cambridge: Cambridge University Press Yu, K. M. (2010). Laryngealization and features for Chinese tonal recognition In Interspeech 2010 ,1529 1532. Zue, V. (1976). Some perceptual experiments on the Mandarin tones. Paper presented at the 29 th Meeting of the Acoustical Society of America, San Diego, California.

PAGE 99

99 BIOGRAPHICAL SKETCH Rui Cao was born and grew up in Tianjin, China. She went to NanKai University in her hometown and received her BA in English for foreign t rade in 2001 She worked at Tianjin IEILTS Training Center for a year as an English teacher after graduation from college. Rui came to the US in 2002. She earned her MA in English (Teaching English as a Second Language) from the University of Toledo, where she also worked as a Teaching Assistant teaching college ESL courses. In 2004, Rui started her PhD study in l inguistics a t the University of Florida (UF) Her interests include phonetics, tones, second language acquisition, foreign language teaching and learning, and neurolinguisitcs. During her PhD study, Rui taught Englis h at the English Language Institute (ELI) at the University of Florida from 2005 2007, 2011 2012, and undergrad course of Mandarin Chinese at UF from 2008 2009. She then worked as an Assistant Professor in the Chinese Department of the Defense Language Ins titute (DLI) for two years in Monterey, California. In spring 2012, she was awarded the CLAS Dissertation Fellowship from the University of Florida. She received her PhD in linguistics from the University of Florida in summer 2012. After graduation, Rui is going to move to Texas to join her husband.