xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001670337
007 cr mnu|||uuuuu
008 051116s2005 flu sbm s000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001198
Doty, Astrid Zerla.
The effects first language use phonological difficulty perception foreign accented speech [sic]
h [electronic resource] /
by Astrid Zerla Doty.
[Tampa, Fla.] :
b University of South Florida,
Thesis (M.A.)--University of South Florida, 2005.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 98 pages.
ABSTRACT: Listener perception of accentedness has been shown to be influenced by experience with L2 (measured by length of residence in US). However, frequency of L1 use and degree of phonological complexity (defined by the number of non-native phonetic features targeted) may provide more insight into the role of experience in the perception of accentedness.Three groups of listeners (monolingual English and Spanish [L1] speakers divided into two groups of high and low use of English [L2]) rated the accentedness of bilingual speakers who spoke with varying degrees of accentedness. The speakers read sentences adapted from Magan (1998) to include phonological aspects likely to be difficult for native Spanish speakers.Listeners performed similarly in rating speakers degree of accent.
Adviser: Ruth Huntley Bahr.
Co-adviser: Judith Becker Bryant
Second language acquisition.
t USF Electronic Theses and Dissertations.
The Effects First Language Use Phonological Difficulty Perception Foreign Accented Speech by Astrid Zerla Doty A thesis submitted in partial fulfillment of the requirements for the degree of Master of Arts Department of Psychology College of Arts and Sciences University of South Florida Co-Major Professor: Ruth Huntley Bahr, Ph.D. Co-Major Professor: Judith Becker Bryant, Ph.D. Catherine Rogers, Ph.D. Doug Nelson, Ph.D. Date of Approval: February 4, 2005 Keywords: second language acquisition, bilingual, phonology Copyright 2005, Astrid Zerla Doty
Dedication To my family who believed in me so much, they left no room for my own doubts. All that I am, and all that I h ope to be, I owe to them. Â“Somewhere, something incredible is waiting to happen.Â” Carl Sagan
Acknowledgements I wish to express my gratitude for the scholarly advise and guidance of Dr. Ruth Bahr. Her encouragement and support helped through many times. I also wish to extend special thanks to Dr Judy Bryant, Dr. Cathy Rogers, and Dr. Doug Nelson who were always available for assistance and assurance. Thanks also to my family and friends w ho believed in and helped me in so many invaluable ways.
iTable of Contents List of Tables iii List of Figures iv Abstract v Introduction 1 Critical Period Hypothesis 3 Equivalence Classification Hypothesis and the Single System Hypothesis 6 ECH Rests Upon Single System Assumptions 7 How Attitude Influences L2 Learning and Foreign Accent 9 Social Factors That Influence L2 Learning 13 Speaker Variables 13 Age of Learning 13 Amount of Native Language Use 17 Experience 18 Speaker and Listener Variables 19 Characteristics of the Stimuli 21 Temporal and Acoustic Properties of the Stimuli 22 Speaking Rate 22 Length of Stimuli 23 Rating Scales 23 Linguistic Variables 25 Phonological Characteristics of the L1 and the L2 25 Purpose of the Proposed Study 28 Method 31 Design 31 Participants 31 Speakers 31 Listeners 32 Materials 34 Demographic questionnaire for the speakers 34 Demographic questionnaire for the listeners 34 Practice sentences 36 Stimulus Recording 37 Recording of the stimulus sentences 38 Stimulus preparation 40 Procedure 41
ii Data Reduction 43 Results 43 Intra-rater Reliability 45 Degree of Perceived Accent 45 Analysis of Individual Speakers Ratings 50 Analysis of SpeakersÂ’ Production of Stimuli Targets 53 Summary 62 Discussion 64 Differences Betw een Listener Groups 65 Characteristics of the Stimuli 71 Future Directions 74 Conclusions 76 References 78 Appendices 83 Appendix A: Study Advertisement 84 Appendix B: Practice Sentences 85 Appendix C: Stimulus Sentences 86 Appendix D: Inst ructions to Speakers 88 Appendix E: Instructions to Listeners 89
iiiList of Tables Table 1 Speaker Characteristics 32 Table 2 Listener Characteristics 34 Table 3 Distribution of Targets for Less and More Difficult Sentences 55
ivList of Figures Figure 1 Mean Accentedness Ratings by Listener and Speaker 47 Figure 2 Mean Ratings of Accent by Both Speaker and Listener Groups 49 Figure 3 Mean Accentedness Ratings of Each Speaker 51 Figure 4 Mean Ratings by Listener and Heavily and Moderately Accented Speaker Group 52 Figure 5 Mean Accent Ratings for Indi vidual Speakers for Less and More Difficult Sentences 53 Figure 6 Mean Percent Target Errors by Individual Speakers and Sentence Type 54 Figure 7 Mean Percent Target Errors for Less and More Difficult Sentences 56 Figure 8 Mean Percent Errors by I ndividual Speakers for Less and More Difficult Sentences 61
vFirst Language Use and Phonological Difficulty on the Perception of Foreign Accented Speech Astrid Zerla Doty ABSTRACT Listener perception of accentedness ha s been shown to be influenced by experience with L2 (measured by length of resi dence in US). However, frequency of L1 use and degree of phonological difficulty (defin ed by the number of non-native phonetic features targeted) may provide more insight in to the role of experience in the perception of accentedness. Three groups of listeners (monolingual Engl ish and Spanish [L1] speakers divided into two groups of high and low use of Eng lish [L2]) rated the accentedness of bilingual speakers who spoke with varying degrees of accentedness. The speakers read sentences adapted from Magan (1998) to include phonolog ical aspects likely to be difficult for native Spanish speakers. Listeners performed similarl y in rating speakersÂ’ degree of accent. Amount of daily L1 use only influenced the ratings of the slightly accented group; the high-use bilingual group rated these speakers as more accented than the native English group, regardless of level of phonological difficulty. These results suggest that the high-use groupsÂ’ lack of L2 experience made them le ss perceptually sensitive to certain phonetic features of English. Because speaker s did not make the predicted target errors, the listener groups may have based their ra tings on features not target ed in this investigation
1 Chapter 1 Introduction Evidence suggests that in the domain of phonology, the younger a person learns a second language, the more likely he or she wi ll be able to pass as a native speaker in that language, but the reasons why this happens have been debated for some time. The theories used to explain the effects of ag e on the learning of a second language have evolved from those that are more neurologi cally based to those that build upon these neurological models with th e addition of sociological considerations. Two major theoretical approaches dominate the literature as explanations of how age-related factors influence the learning of a second language : the critical period hypothesis (CPH) (Lenneberg, 1967) and equivalence classi fication hypothesis (ECH) (Bohn & Flege, 1990; Flege & Eefting, 1987). In this paper, the ECH will be considered within the framework of the single-system hypothesis. The earlier of the two theories, the CPH, states that there are neurological and maturational constraints that influence the l earning of a second language. However, Flege contends that the ability to learn a second language remain s intact across the lifespan. The ECH deals with the inter action of the phonological systems of the first and second languages in predicting areas of difficulty fo r the second language le arner and recognizes sociocultural and sociolingistic factors as in fluential in the producti on of foreign accent.
2These are areas the critical pe riod hypothesis ignores, yet have been shown to be highly related to foreign accent, regardless of the age of learning. Numerous investigations have doc umented patterns of second language acquisition in children and a dults. Many have credited the age differences in L2 acquisition to changes in the physiology of the brain alone, w ithout probing more deeply as to how these changes speci fically relate to the learni ng of phonology in the L1 and the L2. Certainly, the age at which one learns a second language is much more complex than previously thought. In the initial section of this paper, the CPH is explained and its limitations described. Following this, the ECH and th e single-system hypothesis are considered because they represent an evolution of the cr itical period hypothesis and a paradigm shift in the understanding of second language phonologi cal learning. In the present study, the degree of perceived foreign accent was evalua ted in speakers with varying degrees of accentedness by listeners who differed in their amount of L1 use. The speakers read sentences that varied in phonological diffi culty, which was manipulated by creating sentences that include varying numbers of ta rgets that are deemed difficult for Spanish speakers who speak English as a second la nguage to produce. Those evaluating the sentences were also bilingual and differed in their amount of daily L1 use. Critical Period Hypothesis The older and more traditi onal view of second language learning stems from Eric LennebergÂ’s (1967) argument that a critical pe riod exists for the acquisition of a first language (L1). Lenneberg proposed that this period extends from about two years of age through the end of puberty, which he marked at age 14. One criticism of this hypothesis
3is that Lenneberg only considered first la nguage acquisition; there is no definitive or widely accepted theory regarding a critic al period for the acquisition of a second language. Further, if one accepts the notion that a critical peri od exists for second language learning, the question remains as to what the boundaries are for this period. LennebergÂ’s critical period hypothesis was fo rmed at a time when little evidence was available to test it directly. That is there existed no credib le reports of normal children who had been deprived of exposure to a first language. Therefore, he based his hypothesis on indirect evidence, such as differe nces in recovery from aphasia for children versus adults, and differences in the progr ess of language acquisition before and after puberty in children who were mentally reta rded. He claimed that neurological underpinnings were responsible for the ma turational changes observed in language learning abilities. Lenneberg suggested that after puberty, the brain loses the plasticity and organizational capacities necessary for ac quiring language. The implication is that any language acquisition that takes place after puberty will be qualitatively different from that involved in first language acquisition. By extension, any language learning that occurs after the age of pube rty will be more laborious and less successful (Lenneberg, 1967). There are degrees to which research ers in second language learning have subscribed to the CPH, based mainly on the extent to which the theory accounts for exceptional cases of adult sec ond language learning. One interpretation of the CPH is the strong version (Neufeld, 1979). Briefly stated, the assumptions of this position are that there are biological constrai nts upon second language learning in adults, that these constraints are inevitable and irreversible and that no one beyond puberty can hope to
4lose his or her foreign accent in the second la nguage. The soft or weak version, to which Lenneberg ascribed, states that most adults will be incapable of native-like speech in the second language. In addition to the strong a nd weak versions of the CPH, there exist further variations. The term Â“sensitive periodÂ”, which is sim ilar to the weak version, refers to the notion that the age limitation on second language acquisition is not absolute in the sense proposed by the critical period hypothesis (Patkowski, 1980). Rather, the approach suggests that is possible to acquire a sec ond language after the sensitive period, but it would not be possible to attain native-like proficiency. Patkowski (1980) suggested that the term Â“critical periodÂ” be reserved for ca ses of first language acquisition, while the term Â“sensitive periodÂ” be used in the cas e of second language acquisition, because the limitation is on the ability to acquire comple te native-like proficiency in the foreign language. For those in the critical period camp, th ere is a difference of opinion as to the range on maturational constraints on second la nguage learning. Some argue that the range of age-related constraints is lim ited only to phonology, while others contend it extends into other domains of language, su ch as syntax, morphology, and semantics. Adults may have a better ability to think a bout language and use for their learning of an L2 some of the same skills they acquired in learning and mastering their L1. Yet, for reasons not fully understood, adu lts apparently ini tially acquire a second language faster than young children, yet the child learners eventu ally achieve more native-like mastery of the L2 that adults rarely experience (Long, 1990; Snow & Hoefnagel-Hohle, 1978). Additionally, the initial advantag es that adults experience dur ing L2 learning seem not to
5involve the domain of phonology, but are restricted to other do mains of language such as syntax, morphology, and the lexicon. Equivalence Classification Hypothesi s and the Single System Hypothesis The CPH is limited in that it does not fu lly consider how maturational constraints interact with sociocultural variables. A mo re useful explanation is termed the single system hypothesis, (Flege, Freida, & Nozawa, 1997) which asserts that bilinguals have a single phonological system in which both thei r languages reside and that they cannot fully isolate either phonetic system (Guion, Fleg e & Loftin, 2000). It further predicts the loss or attenuation of L1 through disuse. In other words, the less L1 there is, the smaller will be its influence on the L2 (Flege et al., 1997). According to Grosjean (1992), the L1 phonetic system influences that of the L2, a nd the nature of this influence depends on several variables, including the amount and t ype of use of each language. Generally, the more individuals speak their native language, the stronger will be their accent in their second language (Flege et al ., 1997; Guion et al., 2000). Fu rthermore, this relationship seems to be asymmetrical; although L1 use has an effect on accent in the L2, the L2 has little effect on L1 production. The single system hypothesis also states that the loss of L1 may reduce the degree of perceived foreign accent in an L2. The single system hypothesis, which make s predictions based on the amount of use of the L1 and L2, is enhanced when one considers the ECH. This hypothesis distinguishes between identic al, similar, and new sounds in a cross-language context (Bohn & Flege, 1990; Flege & Eefting, 1987). Firs t, consider the perceptual assimilation model (PAM), which asserts that during L1 speech acquisition, non-native segments tend to be perceived according to their similarities to and/or differences from the closest native
6speech segment (Best, 1995). According to this model, listeners will perceptually assimilate non-native phone s to native categories. The equivalence classification hypothesis, on the other hand, deal s more specifically with the perceptual assimilation of second language phones to native categories. The predictions of the hypothesis are that identical sounds in two languages (e.g., the Ge rman and English /b/) are unlikely to cause a problem, but similar sounds, like the English and German /u/, might offer persistent although subtle problems for the second language learner in acquisition. Similar sounds, therefore, should be most difficult because th ey will probably be substituted by the sound from the first language, even after extended L2 exposure. Sounds that are completely new, in the sense that they are not equiva lent or even similar to sounds from the individualÂ’s L1, will be established into a new category as a result of phonetic learning that is not hampered by equiva lency classification. As the am ount of experience with the L2 increases, individuals will produce second language vowels more like natives (Flege et al., 1997). However, production varies as a function of the rela tionship between the native and second language phonolo gy. Therefore, the single system hypothesis seems to work in conjunction with the ECH in that bo th consider experience with the L2 and the interactions of the L1 and L2 phonetic systems. ECH Rests Upon Single System Assumptions Opinion among researchers as to the exis tence of maturational constraints in second language learning is sharply di vided, with both sides offering supporting evidence. Several researchers have shown what they considered to be such evidence of maturation constraints operating (Bir dsong, 1992; Johnson & Newport, 1989, 1991; Patkowski, 1980; Tahta, Wood, & Loewenthal, 1981). Others have claimed that their
7findings suggest an advantage for older learne rs and rejected the CPH altogether, even with respect to pronuncia tion or phonology (Hill, 1970; Neufeld, 1979; Snow & Hoefnagel-Hohle, 1978). Generally, the l iterature thus far has supported three generalizations: adults proceed through ear ly stages of morphological and syntactic development more quickly than children do, older children acquire these domains more quickly than younger children, and child lear ners outperform post-pube scent learners in the long run (Long, 1990). To test the CPH with specific attent ion to rate of ac quisition, Snow and Hoefnagel-Hohle (1978) conduc ted a longitudinal study of the natural acquisition of Dutch by English speakers of different ages. The authors tested two groups English speakers: monolingual English speakers who were just starting to learn Dutch and English speakers who had been living in the Netherlands and speaking Dutch for at least 18 months. The beginning learne rs were tested three time s at four to five month intervals. The advanced learners were test ed only once. The beginning learners wree distributed into the following age groups: three to five year-olds, six-seven year-olds, eight-ten year-olds, 12-15 yea r-olds, and adults. The advanc ed learners were distributed into the following age groups: six-seven year-o lds, eight-ten year-olds, 12-15 year-olds, and adults. Participants we re assessed in the areas of imitative and spontaneous pronunciation, auditory disc rimination, morphology, senten ce repetition, sentence translation, sentence judgment vocabulary, story comprehens ion, and storytelling. The results of this study point to faster initial learni ng in the older subject s relative to the younger ones, but not differences in ultimate at tainment. Interestingly, there were also differences noted within the group of na tive-speakers on morphology and auditory
8discrimination tasks. The authors contended that differences noted within the nativespeaker group are important to the asse ssment of the CPH. If native-speakers demonstrate a range of skills in their firs t language, then it seems logical that, by extension, post-adolescent second language learne rs will not achieve equal skills in their second language. The cause for the range of skills demonstrated by second-language learners who began L2 acquisition at the same age is not addressed by the CPH. One of the variables believed to contribute to a speakerÂ’s degr ee of foreign accent, regardless of age of acquisition, is attitude (Anderson & Koehler, 1988; August & Hakut a, 1998; Bresnahan, Ohashi, Nebashi, Liu, & Shearman, 2002; Cummins, 2000; Hill, 1970; Zecker, 2004). Attitude towards oneÂ’s second language may indire ctly affect oneÂ’s foreign accent in that it determines, to a large extent the amount of daily L1 use, a variable found to contribute significantly to accent under the EC and the willingness to lose the accent. How Attitude Influences L2 Learning and Foreign Accent Disputes over the CPH stem from researchers who contend that age constraints are not only due to neurological changes but may reflect social factors. They insist that the disparity between child and adult performa nce can better be explained by social and psychological factors that are independent of psycholinguistic abilities but dependent on cultural tradition (Anderson & Koehler, 1988; August & Hakuta, 1998; Bresnahan, Ohashi, Nebashi, Liu, & Shearman, 2002; Cummins, 2000; Hill, 1970; Zecker, 2004). These factors include status of the first and second languages, motiva tion or the extent to which one needs to learn the second language in order to function in the target or second culture, and the cognitive demands of learning a second language.
9There is some reason to doubt that the advantages children seem to have in attaining mastery of second languages are uni form across cultures. Hill (1970) explored the influence of the social and cultural aspects of language, as well as attitudes surrounding second language use. For example, she pointed out that adults have opinions about the negative and positive qualities of a second language. These ideas are certainly not inborn, but are the result of their cultu ral traditions (August & Hakuta, 1998). For example, Hill (1970) cited studies that examin ed the role and nature of second language learning in American-Indian and Australian New Guinea cultures. In these studies, adults acquired new languages because of the roles mu ltilinguilism played in political activities. Thus, a motivational factor has been iden tified as contributing to language learning success. Hill noted that most of these ear ly ethnographic studies did not examine the question of language proficienc y or whether adults master foreign languages as well as children in communities where there is intense so cial and political pressure for adults to learn another language. Li kewise, August and Hakuta (1998) found that the extreme importance of learning English in order to succeed in American society overrode immigrantsÂ’ negative attitudes towards Englis h. Additionally, motiva tion is an important factor in AmericansÂ’ apparent lack of bili ngualism. There seems little reason to learn a second language when English is consider ed by many Americans to be a Â“world languageÂ” because anywhere they are in the world, someone will speak English. Another attitude potentially affecting the learning of a second language is that multilingualism is bad for children (August & Hakuta, 1998; Hill, 1970; Zecker, 2004). The contention is that children exposed to more than one language will not perform as well on intelligence tests compared to monoli ngual children. However, Zecker found that
10Englishand Spanish-dominant children pla ced in a two-way immersion classroom (one in which instruction was in both Spanish a nd English) actually performed considerably better on English literacy achievement measur es than did English speakers in regular monolingual English classrooms. Unfortunately, children who w ould naturally be expected to be bilingual, such as American-Indian and Mexicanor CubanAmericans, may find their bilingualism discouraged in the school setting, partly becau se the cultures with which these languages are associated are considered by many teach ers to be lower class (August & Hakuta, 1998; Hill, 1970). Therefore, these childre n may experience the loss of their first language, which subsequently would affect th eir degree of foreign accent. Conversely, adult bilinguals are often considered to be exceptionally intelligen t, but again, only if their language is associated with a high-status culture. Thus, these individuals may be motivated to maintain their fo reign accent. The high-status given to some languages and not others also may be responsible for reinforc ing the idea that adults can never lose their foreign accent. This distinction between the linguistic majority and minority was discussed by August and Hakuta (1998), who desc ribed the effects of societal variables, such as prestige and status of the languages, involved in bilingualism. Their conclusion was that immigrants whose language was not valued in the United States experienced erosion of their first langua ge, including its phonology, whic h influenced their degree of foreign accent. Although Hill (1970) made her observations more than 30 years ago, more recent studies lend credence to her a ssertions. Bresnahan et al. ( 2002) evaluated attitudinal and affective responses toward accented English as a function of speakersÂ’ identity and
11intelligibility. The authors also sought to determine whether participantsÂ’ level of ethnic identity had any relationship to their att itude towards accented speech. Native English speakers from various ethnic backgrounds liste ned to recorded messages in one of six conditions: intelligible foreign friend, intelligib le foreign teaching assistant, unintelligible foreign friend, unintelligible foreign teaching assistant, intelligible American friend, and intelligible American teaching assistant. Af ter listening to the recordings, participants then completed a 101-item questionnaire which assessed attitude, affective response, and ethnic identity. American English was the preferred accent, followed by intelligible foreign accent, with unintelligible foreign accent the least preferred. Role also influenced participantsÂ’ attitude: friends evoked more pos itive responses than teaching assistants in all conditions. These responses may have b een a result of confounds in stimuli used. The script for the friend was a narrative descri bing a trip to visit a roommateÂ’s family, whereas the teaching assistant script was a lecture on human communication. It is possible that the friendÂ’s script was inherently more desirable to listen to. Interestingly, participants with strong ethnic identity deem ed American English, which reflected their ethnicity, to be more pleasing and have hi gher status than the unintelligible foreign accent, whereas those with weak ethnic iden tity found unintelligibl e foreign accent more pleasing and attributed higher status to it than to Amer ican English. The authors explained that people with str ong ethnic identity may be more attached to their ethnic group and, therefore, will be more likely to recognize a foreign accent as representing someone in an out-group and have more nega tive attitudes toward s those with foreign accents.
12The attitudes found in Bresnahan et al. (2002) echo those described by Hill (1970) and may contribute to Ameri cansÂ’ lack of tolerance for some foreign accents. It would have been helpful if Bresnahan et al. had liste d the first languages of the speakers in their study, because this would have addressed Hill Â’s assertions that some accents are more prestigious than others. Thes e limitations call for the need to examine foreign accent in terms of the ECH, which takes these so cial factors into consideration. Social factors that influence L2 learning Investigators have also ev aluated the CPH using other measures, such as age of L2 learning (AOL), age of arrival (AOA) in the target country, foreign language experience, amount of native la nguage use, length of residence, familiarity, and speaking rate (Bohn & Flege, 1990; Flege, 1988; Flege, Bohn, & Jang, 1997; Flege, Frieda, & Nozawa, 1997; Flege, MacKay, & Meador, 199 9; Flege, Yeni-Komshian & Liu, 1999b; Gass & Varonis, 1984; Guion, Flege, & Lofti n, 2000; Matsura, Chiba, & Fujieda, 1999; Munro & Derwing, 1998). All of these variab les have been found to have measurable effects on foreign accent. For example, the la ter individuals arrive in the country of the second language, the stronger their accent as judged by native listeners (Flege et al., 1999a; Flege et al., 1999b). Speaker Variables Age of learning. An individualÂ’s degree of foreign accent depends on the age at which second language learning begins (F lege, 1988; Flege & Fletcher, 1992; Munro, Flege, & MacKay, 1996; Tahta et al., 1981). Fo r example, Tahta et al. (1981) looked at predictors of transfer of accent from the fi rst language (L1) to the second language (L2) in a group of people whose acquisition of Eng lish as an L2 had begun at ages ranging
13from 6 to 15 years. The L1Â’s of the speakers were: Arabic, Armenian, Cantonese, Konkan, Czech, Dutch, French, German, Guja rati, Greek, Hindi, Japanese, Persian, Polish, Portuguese, Spanish, Swahili, Swedis h, Urdu, and Serbo-Croatian. Speakers read a paragraph of English into a tape-recorder th at was later listened to by three independent native-English speaking judges. The recordi ngs were rated for degree of accent and assigned a score of either 0 (no foreign accen t), 1 (slight foreign accent), or 2 (marked foreign accent). There was a significant effect of age of acquisition of English as an L2 on whether there was a foreign accent. S ubjects who had learned the L2 by age 6-7 showed no foreign accent. The authors contended that those who commenced leaning the L2 during their 7th to 9th years tended to show very slight, if any, foreign accent. However, the chances of speaking the L2 accent free dropped significantly for those who commenced learning the L2 between the ages of nine and 11 years. Although Tahta et al.Â’s study (1981) s upports the CPH, some methodological issues are worth noting. First, the stimuli consis ted of a paragraph from an airline leaflet, with no concern for the contributions and in teractions of the two language systems under consideration. The interaction between the L1 and the L2 may have been especially important for this study, which included subjec ts with over 20 different native languages. Certain speech sounds that ex ist in the second language may not exist in the first language, so it would be expected that these sounds might have contributed to the degree of accent. Therefore, if stimuli included speech sounds that were easier for some participants than others, then these speak ers may have been judged as less accented compared to that of speakers whose native language phonology differed significantly from language of the stimuli. The degree of accentedness would therefore have been
14partially due to the inte raction of the languages, rather th an the age of L2 learning alone. The authors only controlled for age at whic h L2 learning began, but did not control for amount of language use, age of a rrival in the target culture, or other variables found to be significant influences on foreign accent. Thus conclusions regarding transfer of accent from the L1 to the L2 must be drawn with caution. Additionally, the rating scale used consisted of only a three-point scale, thus reducing potential variability and categorizing subjects together who might actually have very different degrees of accented L2. Similar results involving the age of l earning (AOL) have been found in studies with better methodologies and more contro lled participant groups (Flege, 1988; Flege & Fletcher, 1992; Munro et al., 1996). Flege and Fletcher ( 1992) found that native Spanishspeaking participants who commenced learning E nglish at the age of five years or earlier could produce English without a detectable accent as meas ured by the ratings given by native English speakers. In contrast, Chines e subjects with an age of learning (AOL) of 7.6 years spoke with a measurable accent (F lege, 1988). Although the authors of these studies observed that the adultsÂ’ pronunc iation of the L2 improved over time, they concluded that a sensitive period for speech le arning is reached long before the age of 12 years. In other words, the age at which a foreign accent first becomes perceptible occurs long before puberty. They further argued that L2 learners of all ages remain remarkably able to establish new phonetic categories for L2 sounds that do not exist in their L1. However, the ability to establish categories for sounds that are similar between the L1 and L2 seems to decrease after the age of five to six years. Instead, learners tend to perceive and produce these similar L2 sounds as the corresponding L1 sounds because they ignore the acoustic differences that distinguish the pairs.
15Abundant evidence exists that individuals cannot achieve a native-like accent in a second language unless they are exposed to it at an early age, but some researchers have argued that adult learners can do so and show evidence. Neufeld (1979) sought to determine the extent to which adults could repr oduce prosodic and arti culatory features of a new language and ultimately demonstrated that high levels of accuracy in pronunciation and intonation are achievable by adu lt second language learners. In NeufeldÂ’s (1979) first study, after receiving 18 hours of intensive training in Japanese and Chinese phonology, 20 adult native speakers of E nglish practiced five times and then recorded ten phrases of four to eigh t syllables in length in both languages. The tapes were rated for degree of foreign accen t by native speakers of each language. Of the 20 participants, three earned a native speaker rating in one language and one did so in both languages. Believing that this eviden ce was not enough to refute the strong version of the CPH, Neufeld investigated individuals who learned their second la nguage as adults and could pass as native speakers of that langua ge. For this second study, 150 French words were prepared that included phonemes and phoneme clusters that were known to be especially difficult for English language lear ners. Three native French speakers and seven nonnative French speakers recorded th e words for judgement by native French speakers. Some nonnative speakers were good enough to be classified as native by some of the judges. NeufeldÂ’s findings led hi m to claim that accent-free second language performance is possible in adulthood and, theref ore, there is no critical period for second language acquisition.
16Perhaps Neufeld (1979) overstated hi s case. According to Long (1990), NeufeldÂ’s studies suffer from some importa nt limitations and possible methodological flaws. For example, in the study on Fren ch (1990), Long raised the question of population validity or generalizability. Th e nonnative speakers, drawn from a bilingual environment, considered themselves highly pr oficient and survived an initial screening interview for Â“accentedness.Â” Therefore, they may not have been representative of the population at large. Long argued that just because these cases were rare does not preclude them as potential test cases for the CPH, but severe ly limits any generalizations about typical adult second language abilities. It could be argued, however, that Neufeld (1979) was not trying to discount the hypothesis with one stu dy. Rather, his position was that Â‘nonuniversalityÂ’ constitutes the principl e flaw in the maturational constraints argumentÂ” (p. 236). Although Long concluded th at NeufeldÂ’s findings do not constitute counterevidence to the idea that there is a sensitive period for second language acquisition, one must recall that the CPH stat es that no one who commences learning a second language as an adult will be able to speak the L2 without a foreign accent, whereas Neufeld has shown that it may be rare but achievable. Amount of native language use. A variable related to oneÂ’s ability to produce target language sounds is the amount of na tive-language use. Flege et al. (1997) examined the effect of L1 use on producti on of an L2. Two groups of native Italian subjects who immigrated to Canada betw een the ages of 2.6 and 9.6 years and whose average residence in Canada was 18 years we re used. The groups varied according to their self-reported daily use of Italian. The participants read and recorded sentences in English that were later rated for degree of foreign accent by native English speakers. The
17researchers found that, generally, the more i ndividuals spoke their native language, the stronger their accent in their second language (Flege et al., 1997). Guion et al. (2000) attempted to replicate th e findings of Flege et al. (1997). This time, Spanish sentences recorded by b ilingual Spanish-Quichua speakers and monolingual Spanish speakers were presente d to native Spanish listeners. Likewise, Quichua sentences produced by Quichua-S panish bilinguals and near-monolingual Quichua speakers were presented to near-m onolingual Quichua listeners. In both cases, the listeners were instructed to rate degr ee of foreign accent; in the case of the Spanish sentences, they were asked to rate Quic hua accent, and in the case of the Quichua sentences, they were asked to rate Spanish ac cent. As in the Flege et al. (1997) study, the more the L1 was reportedly used, the greater the perceived foreign accent in the L2. Moreover, although L1 use was related to degree of accent in the L2, it was not related to L1 production. Therefore, an asymmetrical re lationship exists between the L1 and the L2 sound systems that may have some bearing on the pronunciation of an L2, but the L2 sound system does not seem to be related to the production of the L1 (Guion et al., 2000). This evidence is counter to GrosjeanÂ’s argument which claims that the influence of the L1 and the L2 is bi-directional. Flege et al. (1997) argued, as did Grosjean (1992), that bilingualsÂ’ degree of L1 activation, or how much the L1 is used on a daily basis, influences their L2 production accuracy, rather than only neurological matu ration at the time L2 learning commences. This single system hypothesis, as discussed ear lier, contends that bilinguals have a single phonological system in which sounds from their L1 and L2 reside, making it difficult for them to isolate either system fully.
18Experience. The amount of experience an individual has with the second language has also been found to relate to the speakerÂ’s production and perception of the L2. Bohn and Flege (1990) found that the pe rception of English vow els by adult native speakers of German improved somewhat linea rly with the length of time spent in an English-speaking environment. For example, L2 experience did not affect perception of vowels that were similar in German and Eng lish. However, for the English vowels that were new to native German speakers, the Germans more experienced with English more closely resembled the native English speaker s than did the inexperienced Germans. Likewise, Flege, Bohn, and Jang (1997) found th at, as the amount of experience with the L2 increased, participants produced and percei ved English vowels more like natives. In this case, Â“experienceÂ” was defined as the length of residence in the U.S. In both of the above studies, production and perception varied as a function of th e relationship between the native language and English phonol ogy. Although Flege (1988) found that the amount of second-language experience is relate d to participantsÂ’ ab ility to detect a foreign accent, it was not related to adultsÂ’ L2 production accuracy. Rather, it was the age of learning that was found to be more st rongly related to the degree of perceived foreign accent. Speaker and Listener Variables So far, only speaker characteristics have b een considered as they relate to the degree of perceived foreign accent. However, some factors believed to contribute to the amount of perceived foreign accent are presen t in both the speaker and the listener and seem to interact. These factors are the sp eaker and listener relationship, and familiarity with various aspects of the language and message.
19Wijngaarden, Steeneken, and Houtgast ( 2002) also found that the degree of perceived foreign accent decreased with experience with the second language, and determined that a shared L1 facilitated intelligibility for nonnative speakers. They examined the intelligibility of Dutch, Eng lish, and German sentences produced by native and non-native speakers for trilingual listeners. Listeners in this st udy differed in their amount of experience with and proficiency of German and English. Specifically, the listener groups differed in their experience with German, with one group reporting weekly use and the other group reporting only yearly use. German proficiency differed between the listener groups as well; although both groups were fairly equal in English proficiency, the group who spoke German more often was more profic ient than the group who rarely spoke German. To measure senten ce intelligibil ity, the SRT method was used which is an adaptive method that measures th e speech-to-noise ratio at which 50% of the tested sentences are perceive d correctly. For this study, af ter the presentation of each sentence, listeners orally responds by repea ting the sentence to the experimenter. The listeners who were highly proficient in E nglish found English sentences spoken by native German speakers less intelligible than thos e spoken by native English speakers. For the same listeners, who were less proficient in German, the German sentences were found to be more intelligible when produced by non-native German speakers versus native talkers. The authors concluded that highly proficient listeners were able to use subtle phonetic cues present in native speech, for the less pr oficient listeners, thes e cues were not as helpful. This was because the less proficient L2 listeners were not able to categorize the L2 phonemes as natives. Instead, they percei ved L2 speech as more intelligible if the sounds were matched to their L1 phonemes, as would be done by non-native speakers of
20their same L1. This ability to categorize the L2 phonemes like nativ e speakers increases with the amount of experience one has w ith the target language (Bohn & Flege, 1990; Flege, Bohn, & Jang, 1997). Bent and Bradlow (2003) also investigated the relations hip between speakersÂ’ and listenersÂ’ native language background and speec h intelligibility. They found that a shared native language between a speaker and a listener facilitated speech intelligibility. In their study, native Chinese, Korean, and English sp eakers recorded simple English sentences for presentation to listeners from those same L1s and other native language backgrounds. The listenersÂ’ task was to listen to the senten ce and write down whatever she or he heard. In general, non-native listeners found nativ e speakers to be more intelligible than speakers from other first language backgrounds. Interestingly, for non-native listeners, the speech of non-native speakers from the same L1 was found to be as intelligible as that of native speakers. The authorsÂ’ conclusion wa s consistent with that of Wijngaarden et al. (2002): non-native speech perception is as sociated with the relationship of shared speaker and listener L1. L2 speakers with the same native language share linguistic knowledge of both the L1 and the L2. In c ontrast, L2 speakers who differ in native language background share only linguistic know ledge of their target language. Therefore, a non-native listener is better able to interpret the L2 speech of a speaker with the same L1 compared to a speaker with a di fferent L1, even if the speech differs greatly from the target language norm. Characteristics of the st imuli and scaling methods Thus far, speaker and listener variables ha ve been considered as they influence the degree of perceived foreign accent and produc tion in an L2. However, characteristics
21beyond those of the speaker and listener influe nce the ease with wh ich one is able to produce a second language and the degree to wh ich one is judged to have a foreign accent. Specifically, there exist certain characteristics of the speech signal and differences in the types of scales used that influence the degree of perceived foreign accent. Temporal and Acoustic Properties of Stimuli. The kind of stimuli used no doubt affects the degree to which accent is perceived and the way in which a second language is produced. Gottfried and Beddor (1988) presented spect ral and temporal manipulations of French vowels to French and English listeners who were asked to listen to and identify the vowel in each syllable. The vowels were identified differently by the two language groups. In contrast to the French listeners, native English listeners were influenced by vowel duration in their categoriz ation of the vowels, rather th an only spectral cues, which is consistent with the prominent role of duration in the English vowel system. The researchers concluded that how one perceives a given vowel cont rast in terms of spectral or temporal cues does not simply follow fr om experience with t hose vowels through their use. Rather, how one perceives a foreign language sound depends on the extent to which its acoustic properties correlate within oneÂ’s L1 phonetic system. This argument is consistent with the single system hypothe sis proposed by Flege and Eefting (1987). Speaking Rate. Manipulations of various aspect s of the speech signal have been shown to affect listenersÂ’ ev aluations of native and foreign accented speech in somewhat counterintuitive ways. Increasing the speaking rate, for ex ample, resulted in more nativelike accent ratings of Mandarin speakers (M unro & Derwing, 1998). The ideal rate of nonnative speech was found to be somewhat slower than that of native speech, but faster
22than what nonnative speakers typically pr oduce (Anderson & Kohler, 1988). Munro and Derwing (2001) found that there is a point beyond which an increase in rate decreases accent ratings. Length of Stimuli. Although speaking rate certainly affects the accent ratings of foreign speech, other manipulations, such as duration of the stimuli, have yielded similarly interesting results. Flege (1984) isolated progressive ly shorter units of English speech to determine whether a French foreign accent was detectable in English. In these experiments, listeners simply had to identi fy each item as having been spoken by a native or non-native speaker. First, he used sent ence-length stimuli a nd found that listeners were able to detect a French accent. Then, he digitally manipulated the stimuli, isolating only the first word from the sentence and then only the first syllable from an utterance and, again, both trained and untrained listeners were able to detect a French accent. Finally, Flege found that even in short bursts of 30 ms, obtained from the digitally edited /tu/ syllable, native English listeners were able to detect the presence of a French accent. Rating Scales Different types of scales have been used in experiments involving the rating of foreign accent (Flege, 1988; Flege, 1995, Flege, Yeni-Komshian & Liu, 1999; Guion & Flege, 2000; Magen, 1998; M eador et al., 2000; Munro & Derwing, 1998, 2001; Piske, MacKay, & Flege, 2001; Southw ood & Flege, 1999). In fact, Flege has spoken of the appropriateness and benefits of using an equal appearing interval scale (EAI) for the rating of foreign accents. He compared it to direct magnitude estimation and concluded that the EAI scale was prefe rred. Further, he has argued that the range should be nine points. However, he never ex plicitly explained why it is a better choice than the direct magnitude. A personal communication with James Flege (May 27, 2004)
23revealed his reasons for using this type of scale rather than the direct magnitude estimation: "Parsimony. How many different degrees of accentedness can listeners reliably discern? In the Southwood/Flege paper, we provide preliminary ev idence that a sevenpoint scale under utili zes listeners' ability, whereas a ninepoint scale does not. Of course, as you will have noted in the Piske et al. paper that a nine point scale and a continuous scale yield much the same results." The use of the EAI scale is also consistent with the rating scales used in much of the foreign accent literature. A review of the methodologies of studies that had listeners rate the degree of foreign accent they percei ved in speakers reveal ed that, with the exception of one (Flege, 1988), all used equal appearing interval sc ales (Flege, YeniKomshian & Liu, 1999; Guionet al., 2000; Magan, 1998; Munro & Derwing, 1998, 2001; Piske, MacKay, & Flege, 2001; Southwood & Flege, 1999). In fact, two of them specifically discussed the methodologies in te rms of type of rati ng scale and concluded that listeners were able to partition L2 fore ign accents into equal intervals, so it is appropriate to use an EAI scale in fore ign accent studies. Other researchers have examined various types of scales and the results support the use of the EAI scale. With regard to the range of the scale vales that are needed to exploit listeners' full range of sensitivity, Southwood and Flege (1999 ) found that "a nine-point rating scale should be used to rate L2 speech samples for degree of foreign accent" (Piske, MacKay, & Flege, 2001, p. 195). This happens because, when rating foreign accents, a potential ceiling effect could occur due to the number of scale intervals used. Southwood and Flege noted that, although seven-point scales are frequently used, they may not be
24sensitive enough for all listeners to discrimi nate among the stimuli, in this case, the speakers' sentences. Additionally, these authors examined the use of the fiveand sevenand nine-point scales for native and nonnative sentences. The five-poi nt scale failed to yield a significant between-group difference, whereas the ninepoint scale did. Linguistic variables Additional factors spec ific to the languages under inve stigation also contribute to a speakerÂ’s foreign language production and degree of foreign accent. As Bent and Bradlow (2003) noted, there ar e specific linguistic contribu tions of the L1 during the production of the L2 which serve to mediat e intelligibility betw een L2 speakers and listeners. Some of these specific interactions are discussed next. Phonological similarities and differ ences of the L1 and the L2. Specific and unique interactions exist between the L1 and L2 phonological systems that may predict areas of pronunciation difficulty for the second language learner (Bohn & Flege, 1990; Flege & Eefting, 1987; Guion et al., 2000; M unro & Derwing, 1998). To better elucidate the difficulties encountered by native Spanis h speakers learning English, a discussion follows that explores the unique contributi ons of the Spanish and English phonological systems to pronunciation. This discussi on is relevant to the present study. The Spanish and English phonological systems have specific differences that may create difficulty for a native Spanish speaker le arning English. Only the more salient and those most relevant to the present study ar e discussed. It should be noted that the following observations are generalizations and that not all Spanish and English dialects exhibit these characteristics.
25The differences between the English and Spanish phonological systems include syllable structure, vowel quality, consonants, and stress. One of the factors affecting syllable structure is the in sertion of the initial an d non-initial epenthetic / / by Spanish speakers before an English cluster (Magan, 1998). For example, Spanish speakers may say / staemp/ instead of /stamp/ because the Span ish language has differe nt clusters than English. The insertion of the epenthetic schwa facilitates production of the English clusters by breaking the clus ter into two syllables: / s/ and /taemp/. Vowels are another area in which Spanis h and English differ. Whereas English has 14 vowels, Spanish has only five. Eng lish speakers tend to reduce the vowels in unstressed syllables, whereas Spanish speaker s are more likely to produce them fully, e.g., seas [o] ns for seas [ ]ns. Also, in Spanish accented speech, English lax vowels tend to be produced as tense vowels, e.g., ch [i] p for ch [I] p (Magen, 1998). Even during rapid speech, vowel length is maintained, and Span ish speakers tend to delete syllables or consonants rather than shorten the vowel (Iglesias & Anderson, 1993). Spanish and English also have diffe rent consonants a nd phonological rules. Spanish speakers tend to drop word final /s/ or /z/ to simplify final clusters not allowed in Spanish. In Spanish, plurality is marked by redundancy across the verb phrase, so one can delete the final /s/ and still convey plura lity by marking it in the verb (Iglesias & Anderson, 1993). Spanish speakers learning English may therefore say I saw three cat instead of I saw three cats. Also, Spanish does not indicate possession by using an apostrophe /s/, /z/ or / z/, so, instead of saying the girlÂ’s dog Spanish speakers would say the dog of the girl So an English sentence such as I saw the girlÂ’s dog may be produced as I saw the girl dog Spanish has only one affricate, /t /, which occurs in word-initial
26and intervocalic positions and is commonl y substituted for the English fricative / /. There are voicing distinctions be tween intervocalic /s/ and /z/ in English that are typically not produced by Spanish speakers, leading to substitutions such as free [s] er for free [z] er Finally, there are subtle stress pattern differences between English and Spanish. In terms of lexical stress, Spanish typically places the stress on the final syllable in multisyllabic words, whereas English stress pa tterns call for the stress on the penultimate or antepenultimate syllable. Therefore, Span ish speakers tend to stress the final syllable of English multisyllabic words, e.g. combination for combination Also, Spanish has two degrees of stress (weak and st rong), whereas English has th ree (strong, medial, and weak) (Stockwell & Bowen, 1965). Further, the vow el qualities differ between those found in weak-stress syllables versus those found in sy llables with greater stress. In English, word-level stress patterns can be used to differentiate nouns from verbs, such as in nsult and ins lt (Stockwell & Bowen, 1965). In Spanish however, word stress is less varied and of different kinds. Verb endings such as o first person singular, and third singular past, exemplify stress contrasts that cha nge meaning in Spanish. For example, hblo means I am speaking while habl means he spoke (Stockwell & Bowen, 1965). Finally, English has a relatively regular alteration between syllables of weak stress and those of strong stress, whereas Spanish has fairly long sequences of weak-stress syllables. This can be seen in the differences in the stress patterns between English pertion and Spanish operacin (Stockwell & Bowen, 1965). The present study attempted to dete rmine whether increased phonological difficulty, defined by the number of target phonemes deemed most difficult for Spanish
27speakers learning English, would be related to an increase in percei ved foreign accent. Manipulation of this variable is une xplored in the literature thus far. Purpose of the Present Study Despite the differential outcomes of va rious studies on ag e-related language learning constraints, several generalizations can be made (Long, 1990). Adults proceed through early stages of morphological and syntac tic development more quickly than children. This differential rate of acquisition may be the re sult of older le arnersÂ’ more advanced metalinguistic skills. In other words, adults may have a better ability to think about language and use for their learning of an L2 some of the same skills they acquired in learning and mastering their L1. So far, studies have been conducted to determine whether the amount of language use affects the degree of perceived foreign accent by Italian and Canadian-English speakers (Flege et al., 1997), Chinese and English speakers (Flege, 1988) and Spanish and Quichua speakers (Guion et al., 2000). In support of the ECH, it has been shown that the more individuals use their L2 the better able they are at gauging the degree of foreign accent in their target language. Perhaps, as asserted by the ECH, when compared to adults who use their second language rarely, adults who often use their second language establish phonetic categories in their target language that more closely approximate those of a native speaker, making them better able to detect foreign accents in their L2 than individuals who use their L2 less often. The first aim of the present study was to extend these findings to Spanish-English bilingual speakers. In the present study, the question of wh ether the degree of perceived foreign accent in an L2 is related to the listenerÂ’s amount of L1 us e was also investigated. Bohn
28and Flege (1990) found that the perception of English vowels by adult native speakers of German improves with their amount of experi ence with English. Likewise, Flege, Bohn, and Jang (1997) found that as the amount of e xperience with the L2 increased, subjects perceived English vowels more like natives. In these studies, experience was defined as the length of residence in the U.S. Additionally, Best and Bradlow (2003) and Wijngaarden et al. (2002) found that intelligib ility of L2 speech improves if the listeners had shared L1s. Although those studies fo cused on the amount of L2 exposure, the present study investigated th e amount of L1 use. In the present study, the relationship between amount of daily L1 use and perceived degree of foreign accent were explored. It was hypothesized that individuals who used their L1 less would be better able to detect an accent in nonnative speakers of their L2 than would individuals who use their L1 more often. The second purpose of the present st udy was to determine whether the phonological difficulty of the stimuli would a ffect how well the speaker was able to produce them and whether this increased phonological difficulty would result in a more detectable accent. Difficulty was determ ined by how many potentially challenging targets were included in the stimuli. It was predicted that the level of phonological difficulty of the stimuli would affect the de gree of perceived foreign accent, with more difficult sentences eliciting higher accent ratings. The less difficult sentences were predicted to be produced with less of an accent and, theref ore, rated as less accented by all listener groups. The more difficult senten ces were predicted to be produced with more of an accent and be rated as more acc ented by all listener groups. The low L1 use group was predicted to rate sentences as more accented than would the high L1 use
29group. This was predicted because the low L1 use group would have more experience with English and would therefore, more read ily notice deviations from native-language norms in the productions of the speakers.
30 Chapter 2 Method Design The experimental design was a 3 x 3 x 2 mixed-model factorial. Listener group was varied between subjects (native Eng lish, high-L1 use bilingual, low L1-use bilingual), and both speaker category (heavil y accented bilingual, moderately accented bilingual, and slightly accented bilingual) and level of phonological difficulty (less difficult, more difficult) were manipulated wi thin subjects. The dependent variable was ratings of accentedness. Participants Speakers. Three groups were used as speakers: one heavily accented bilingual Spanish-English speaking group, one moderately accented bilingual Spanish-English speaking group, and one slightly accented bilingual Spanish-English speaking group, with two speakers per group. Th ese participants were recruited from the University of South Florida English Language Institute ( ELI) and through advertisements within the Departments of Psychology and Communicati on Sciences and Disorders (CSD) (see Appendix A). Extra credit points were offere d to those subjects enrolled in courses within the Psychology and CSD Departments.
31The inclusion criteria were that speaker s were Spanish/English bilingual women between the ages of 18 and 35 years who did not report a history of hearing, speech, or language disorders. The investigator used he r clinical judgement to discern whether any speakers had any speech defects that would pr eclude them from being participants. Race was not considered as either an inclusion or exclusion criteri on. According to self-report, all participants had Spanish as their L1 a nd English as their L2. Additionally, these participants had not learned an L3 because, according to the equivalence classification hypothesis, they may have experienced interfer ence from their third language that would have affected their first and second languages in ways that were dissimilar from bilingual speakers. As shown in Table 1, the speaker s in each group were of roughly the same age and came from five different Spanish-speaking countries. Table 1. Speaker Characteristics Speaker Group Mean Age in Years Country of Origin Slightly Accented 26.5 U.S., Dominican Republic Moderately Accented 25.5 Dominican Republic, Colombia Heavily Accented 27.0 Venezuela, Mexico Listeners. A power analysis was performed to estimate a sample size that would ensure a power of .82. Using the PearsonHartley charts (Meyers & Well, 1995), it was determined that 12 subjects per group would be needed to obtain th e desired significance level of .05. Twelve native monolingual English speakers from the United States comprised the first group. The listeners in the other groups were 24 native-Spanish speakers, 12 higher use and 12 lower-use, who learned English as a second language
32some time after adolescence and spoke no ot her languages. All lis teners were between the ages of 18 and 45, and the groups include d both males and females. Listeners were recruited from the English Language Instit ute and the Departments of Psychology and Communication Disorders at the University of South Florida and were offered extra credit points for their participation (see Appendix A), paid $5.00 cash, or provided the equivalent in gift certificates to on-cam pus restaurants. They all passed the Speech Listening Test (Griffiths, 1967) before being accepted into the study, indicating that they had normal hearing and speech discrimination abilities. This test demonstrated the listenersÂ’ ability to discriminate speech sounds presented in groups of words that differed in initial or final consonant, such as lake, rake take, bake These individuals also completed a language background questionnaire (see Materials) and, based on their responses, were designated as either higher-use Spanish L1 listeners or lower-use Spanish L1 listeners. Participants who indi cated that they spoke Spanish 50% of the time or more were placed in the high-L1 use group, and those who indicated that they spoke Spanish 49% of th e time or less were place in the low-L1 use group. As shown in Table 2, the low-use list enersÂ’ mean amount of daily L1 use was 23% and the high-use listeners Â’ daily amount of L1 use was 72.1%. Although all three groups of listeners were fairly similar in te rms of gender, there wa s significant variability in terms of their country of origin, with 12 countries being represented among the 24 bilingual listeners.
33Table 2. Listener Characteristics Listener Group Gender Mean Age In Years County of Origin Mean Percent Self-reported Daily L1 Use Native English M 5 F 7 29.4 (8.9) U.S. Low-use Bilingual M 5 F 7 22.2 (3.3) Colombia, Cuba, Nicaragua, Dominican Republic Guatemala, Spain, Venezuela, El Salvador, Puerto Rico 23% (14.86) High-use Bilingual M 4 F 8 27.9 (8.1) Colombia, Mexico, Peru, Cuba, Venezuela, Puerto Rico, Panama 72.1% (8.38) M=Male; F=Female Standard deviations are presented in parentheses. Materials Demographic questionnaire for the speakers. Potential speakers completed a questionnaire prior to being accepted in th e study. The questionnaire was reviewed before the speaking task began to ascertain whether the individual met the inclusion criteria. It consisted of the following: 1. What is your country of origin? 2. At what age did you begin to learn English? 3. Do you speak any other language besides Spanish and English? 4. Have you ever been diagnosed with a speech, language, or hearing disorder? In addition to the questions listed, participants also provided basic biographical information such as age, gender, and educational level. Demographic questionnaire for the listeners. Responses to these questions indicated the listenerÂ’s language preference and use in given social situa tions and, therefore, provided a measure of her overall daily L1 use. The questionnaire was reviewed during
34the listening task, since it was not necessa ry to ascertain in which group the listener would be included prior to this. 1. At what age did you be gin to learn English? 2. Which language do you use most at home? 3. Which language do you use most at work? 4. Which language do you use most at parties? 5. Which language do you us e most with friends? 6. On average, how much do you use Spanish daily? 7. If you learned in school, how old were you when you first started English classes? 8. How many times/hours per week did you have English class at that time? 9. Do you or have you ever had a diagnos ed speech and/or language disorder? 10. Do you speak any other language be sides English or Spanish? In addition, participants provi ded basic biographical inform ation such as age, gender, birthplace, and educational level. Questions two, three, and five were ta ken from a previous study, which also looked at the effects of L1 use on the de gree of foreign accent (Guion et al., 2000). Questions four and five are a modification a nd combination of seve ral of the questions used in that same study (Guion et al., 2000). Those authors had asked their participants about language usage with siblings, with friends at parties, and when meeting friends on the street. In order to participate in the st udy, individuals indicated in question 1 that they learned English after adolescence (operationali zed as 12 years of ag e), and in question 10 they indicated that they spoke no other languages besides English and Spanish.
35Questions two and three were used to determ ine the participantsÂ’ preferred language for family and work situations because thes e areas constitute the bulk of their spoken interactions. Questions four a nd five provided an indication of the participantsÂ’ social language use. Question six was included to obt ain the participantsÂ’ overall estimation of their daily L1 use. Individua ls were assigned to higher-use group if they indicated that Spanish was the main language used at work or home (questions 2 and 3) and indicated that Spanish was the main language used in social settings (questions 4 and 5). Additionally, to be included in the higher-use L1 group, individuals no ted that they used Spanish more than half the time on a daily basis. Questions 7-9 were included for descriptive purposes only. Practice sentences Prior to the speaking and listening tasks, participants had the opportunity to practice the ta sk using ten sentences. These sentences are listed in Appendix B and were constructed to be simple enough for speakers to produce and listeners to comprehend with little difficulty. The speakers for the listening practice tr ials were different from those in the actual listening task and in cluded bilingual speakers judged to have a both strong and slight Spanish accents by doctoral student s and faculty in the Department of Communication Sciences and Disorders trained in phonology. Although the task was fairly simple and five practice sentences woul d probably have trained the listeners to the task, it was determined that using ten senten ces would cause the listeners to focus on the accents of the speakers rather than on the m eaning of the sentences. The use of anchors was therefore not necessary.
36Stimulus recording Because they were intended to elicit a foreign accent from the speakers, the stimuli for this study consiste d of 20 English sent ences, each containing several words with sounds presumed to be difficult for native Span ish speakers learning English. The stimuli were modified from those used in a previous study (Magen, 1998) and included phonemes and phrases in American English that were e xpected to result in the production of a foreign accent when spoken by Spanish speakers. Specifically, they were constructed to include phonemes and phonem e sequences that are present in English but not in Spanish or that differ between English and Spanish (Magen, 1998; see Appendix C). Sentences, rather than words, were used because they more nearly approximate the short utterances typical of convers ational speech. The sentences were grouped as either phonologically more difficult or phonologically less difficult based on the numbe r of Â“targetsÂ” (i.e ., challenging phonemes or phonemic sequences) present in each se ntence. The more phonologically difficult sentences contained between four and eight speech targets. The less phonologically difficult sentences contained at most three speech targets. The number of targets per sentence was found to be significant ly different between the levels, ( t (18)= 2.75, p < .05). It was predicted that the mo re phonologically difficult sentences would result in a greater number of errors from L2 native-speaker norms by the heavily accented group relative to the slightly accented speaking group. It was expected that the differences from L2 native-speaker norms that the speakers made would vary according to their level of accent, with heavily accented individuals produc ing more errors on the targets than the slightly and moderately accented individuals. It should be noted that although the term Â“errorÂ” is used throughout this paper, the authors recognize that non-native speakers are
37really producing differences compared to nati ve speaker norms and not error per se. It was expected that the speakers would insert an epenthetic / / before fricative plus stop clusters in English syllables that are not allo wable in Spanish. It wa s also expected that the speakers would produce the vowels in the un stressed syllables fully, as tense vowels. The Spanish speakers were also expected to dr op the word final /s/ or /z/ to simplify final clusters not allowed in Spanish, and because plurality is marked by redundancy across the verb phrase, the plural /s/ wa s expected to be dropped. The speakers were also expected to substitute their only affricate, /t /, for the English / /. The voicing distinctions between intervocalic /s/ and /z/ in English words were not expected to be produced by speakers. Finally, the speakers were expected to stress the final syllable of English multisyllabic words instead of the penultimate or antepenultimate as appropriate. Speakers were also expected to produce several weak stress syllab les in a row, because unlike English, which has a relatively regular alte ration between syllables of weak stress and those of strong stress, Spanish has fairly long sequences of relatively w eak-stress syllables. Recording of stimulus sentences. The speaking task was administered individually in the Acoustics Laboratory in the Department of Communication Sciences and Disorders at the University of South Flor ida. Upon arrival in the lab, speakers were greeted and told about the nature of the study. They were direct ed to have a seat and then were given the informed consent forms. Th en, they were given the language background questionnaire and instructed to answer the questions to the best of their ability and recollection. After each participant comp leted the language background questionnaire, the investigator quickly determined whether she met the inclusion criteria before proceeding. All participants met the criteria and were provided the list of practice
38sentences (see Appendix B) to rehearse the recording task. Approximately three minutes before the scheduled recording the speaker s were given the stimulus sentences (see Appendix C) so as to allow for familiarizati on with the stimuli and ensure fluent speech during recording. These sentences were provided on a sheet of paper with size 16 font so as to make them large enough to read easily. The speaker s were instructed to read the sentences three times each at a normal conversationa l pace and to leave about three seconds between sentences. To avoid ambiguity, the experimenter demonstrated an acceptable pace by reading a practice, non-stimulus sent ence to the speaker. For complete speaking task instructions, see Appendix D. Once the speakers demonstrated understanding of the task and had familiarized themselves with the stimuli, the speaking task began. The experime nter exited the booth and returned to the recording equipment to monitor the experiment while the speaker read the practice sentences. After reading a ll 20 stimuli sentences three times each, the participant had completed the speaking task and was thanked for her time. The speaking task took approximately 20 minutes to complete. A digital file was constructed by recordi ng all three readings of the 20 sentences by each of the six speakers using a Roland VS -1824 24 bit digital studio workstation with a sample rate of 44.1kHz. After all the speak ers recorded the sentences, speakers were assigned to the heavily accented group, moderately accented group, or the slightly accented group based on accent judgements made by doctoral students and faculty in the Department of Communication Sciences a nd Disorders trained in Spanish-English phonology. These individuals met and jointly li stened to the speakersÂ’ recordings and
39independently rated their accents using a nine-point rating scale. There was complete agreement among the listeners as to what group to assign the speakers. The two speakers who were categorized as sli ghtly accented obtained a mean accent rating of 2.09. The two speakers placed in the moderately accente d group received a mean accent rating of 5.08 and the two speakers in the heavily accented group received a mean accent rating of 8.76. Stimulus preparation. The sentences produced by the speakers were saved in the workstation and to a disk for subsequent di gital editing. They were edited using the digital waveform editing program PraatÂ™ on a De ll computer so that only the second of the three sentences was used for presentati on to the listeners. Th is method allowed the experimenter the option of using the first or third sentence in the series if the second was in some way inaccurately produced (e.g., hesita tion or stuttering on a word or syllable). The interest was in the second sentence because it was thought that after having read the sentence once, the second r eading would be produced more fluently. The selected sentence was saved in a separate file for each of the 20 sentences for all six speakers. Using the program Resample, the sentences we re then resampled to a rate of 48,828 Hz to make it compatible for use with the other pr ograms needed to create the experiment. The 120 files (six speakers x twenty senten ces) containing the sentences were then arranged for order of presentation and the experiment was created and run using the ECoS Version 2Â™: Experiment Generator a nd Controller program. To avoid order effects, four different orders of the stimu li were created in which the sentences were randomized with the constraints that no more than two sentences per speaker would occur in succession. Sentences were presented to the listeners with approximately three
40seconds between trials. To test for reliabilit y, the first 20 sentences were presented again in the same order at the end of the experime nt, but were not used for other statistical analysis. Procedure Participants in the listeni ng task were tested individually or in groups of two. Upon arrival in the lab, they were greeted a nd told about the nature of the study. After giving consent to participate they completed the language background questionnaire. The questionnaire was administered orally by th e experimenter. This was done to better ensure that the participan ts understood the questions a nd that the experimenter understood with the listenerÂ’s language background. After it was determined that each individua l met the criteria to participate, the listening task began. The listening task took place in the Acoustics Laboratory in the Department of Communication Sciences and Di sorders. Participants were directed to have a seat in front of one of the computers. First, they completed the Speech Listening Test (Griffiths, 1967). This test consisted of 25 sets of four words that were minimal pairs. The minimal pairs selected did not include phonemes deemed difficult for Spanish speakers learning English in or der to avoid interference from the listenerÂ’s L1. Thus, they provided a clear picture of participan tsÂ’ speech discrimina tion abilities in the absence of L1 interference. In this test, the listener first had four practice items to familiarize him or her with the task and to allow for appropriate volume adjustment. The listener heard a word spoken via headphone s and saw four words displayed on the computer screen (e.g. rake, take, make, bake). He or she was instructed to click with the mouse on the word that he or she thought wa s spoken. The test was scored after the
41entire listening task had been completed. All participants scor ed 100% on this task, which indicated that they had speech discrimi nation abilities that were within normal limits and were included in the study. After the Speech Listening Test had been completed, the participants were invited to st and and take a short br eak before proceeding to the listening task. After the break, they were instructed to return to their seats in front of the computer. They then heard the experime nter read the listeni ng task instructions. For complete listening task inst ructions, see Appendix E. The listeners were told that they woul d use headphones to hear sentences spoken by non-native English speakers and that they should rate each sentence independently. Listeners were instructed to estimate the degree of Spanish accent in each sentence by using a 9-point rating scale that was displayed on the computer screen. The scale had the labels: Â“slight Spanish accentÂ” at the left side of the scale and Â“strong Spanish accentÂ” at the far right side of the scale, but did not di splay numbers. Listener s were instructed to drag the curser to any one of the nine point s along the scale to i ndicate the degree of Spanish accent present in each of the sentences. Each session began with ten practice tr ials consisting of non-stimulus sentences (see Appendix B) to familiari ze the listeners with the use of the scale and to attempt to shift the listenersÂ’ focus from the meaning of the sentences to the speakersÂ’ accents. Following these practice trials, the listener was given the opportunity to ask questions regarding the task. Feedback during the practice trials was re stricted to the use of the scale, not the rating of the speakersÂ’ accents. For both the practice sentences and the stimulus sentences, the listener dragged th e cursor to a point along the scale that corresponded to a number representing the degree of foreign accent that he or she judged
42the speaker of the sentence to have. After each trial the participan t clicked Â“AcceptÂ” at the top of the screen to advan ce to the next trial. After the listener had rated all sentences, he or she was asked whether any voices s ounded more distinctive than others and whether this recognition influenced their ratings At this point, the listener had completed the listening task and was free to leave. Data Reduction After the experiment was finished, the da ta were saved in an Excel file that contained the responses sorted by listener type speaker type, and stimulus type. A mean rating was calculated for each speaker group a nd stimulus type, which yielded six values per listener. These data were put into SAS for statistical analysis.
43 Chapter 3 Results This study investigated the relationship between amount of daily first language use, the phonological difficulty of the stimuli, and degree of perceived foreign accent. It was hypothesized that individuals who use their L1 less ofte n, in this case the low-use bilinguals, would be more able to gauge th e degree of foreign accent like native speakers compared to those who use their L1 more often when rating the degree of accent in nonnative speakers of their L2. The second purpose of the study was to determine whether the phonological difficulty of the s timuli would affect how well the speakers were able to produce sentences and whether this increased phonologi cal difficulty would result in a more detectable accent. Diffi culty was determined by how many potentially target phonemes were included in the stimuli. It was pr edicted that the level of phonological difficulty of the stimuli would aff ect the degree of perceived foreign accent, with the more difficult sentences el iciting higher accent ratings. The study required that speakers who differe d according to their degree of foreign accent (slight, moderate, and heavy) read E nglish sentences that varied by level of phonological difficulty (more or le ss difficult). The sentences we re then rated for degree of foreign accent by listeners who differed according to amount of daily L1 use but not necessarily in proficiency as th is variable was not controlled.
44 Intra-rater Reliability To determine intra-rater reliability, the fi rst 20 sentences were presented again in the same order at the end of the experime ntal procedure. A Pearson Product-Moment Correlation was run to investigate intra-rate r reliability for ratings of the repeated sentences. The correlation coefficients fo r the native English listener group ranged from .63 to .96 ( M = .80, SD = .098). The correlation coefficien ts for the low-L1 use bilingual group ranged from .29 to .93 ( M = .80, SD = .18). When the score of the outlier in the low-use bilingual group was removed, the mean was .86. A correlation of .30 is considered to be moderate, therefore, the de cision was made to include the data of the outlier. Also, the mean correlation for the lowL1 use group was still higher than that of the high-L1 use group with the outlier include d. The correlation coefficients for the highL1 use bilingual group ranged from .48 to .93 ( M = .75, SD = .15). With the exception of the one outlier, all of the corre lation coefficients for all list ening groups were considered moderate to high. However, there was grea ter variability among th e listeners in the bilingual groups. Generally, it was determined th at participants were able to do the task with an acceptable level of reliability. Degree of Perceived Accent A three-way mixed-model analysis of variance (ANOVA) was run analyzing speaker, listener, and sentence difficulty f actors. Speaker group (strongly accented, moderately accented, and slightly accented native Spanish) was a within subjects factor, listener group (native English, high-L1 use bi lingual, and low-L1 use bilingual) was a between subjects factor, and difficulty of se ntences (more difficult and less difficult) was a within subjects factor. Comparisons were co nducted for all significan t effects related to
45the hypotheses. All effect sizes were calculated using the formulas provided by Rosenthal and Rosnow (Levine & Hullett, 2002). As predicted, there was a main effect of degree of speaker accent on perceptions of accentedness, F (2, 66) = 317.25, p < 0.0001, = .91. As shown in Figure 1, the slightly accented speaker gr oup was rated as least accented ( M = 2.269, SD = 1.06). The moderately accented speaker group received ratings intermediate to the slightly accented and heavily accented speaker groups ( M = 3.54, SD = 1.09). The heavily accented speaker group was rated as most accented by all listener groups ( M = 6.82, SD = .92). Although there were no other main effect s, there were two interactions. A significant two-way interaction was f ound between speaker and listener group, F (4,66) = 3.61, p < .0159, = .22. Tests of Least Significant Differences (LSD) revealed that the high L1-use bilingual listener group rate d the slightly accen ted speaking group as significantly more accented than did the na tive English listener group (see Figure 1). There were no differences among listener groups in how they rated the moderately and heavily accented speaker groups. Thus, in ga uging the degree of foreign accent of the slightly accented speaker group, it could be that amount of L1 use is inversely related to the ability to detect subtle accent or that th e high L1 use group rated the slightly accented speaker group more harshly.
46 Figure 1. Mean Accentedness Ratings by Listener and Speaker Groups A significant three-way interaction wa s found for speaker group, listener group, and phonological difficulty, F (4, 66) = 2.47, p < .0530, =0.15. This interaction was not significant after the Greenhouse-Geisse r Epsilon correction was made, and was significant only at p values of .0721. However, after ch ecking the data for additivity and sphericity, it was concluded that the resu lts were acceptable before the correction, and were therefore significant. As Figure 2 suggests, LSD tests re vealed that the high L1-use 1 2 3 4 5 6 7 8 9Slight AccentModerate Accent Heavy Accent Speaker GroupsRatings of Degree of Accentedness Native English Listeners LowL1 Use Bilingual Listeners High-L1 Use Bilingual Listeners
47listener group rated the slightly accented speaking group as significantly more accented than did the native English listener group for both the more and less phonologically difficult sentences. The high L1-use listene r group also rated the slightly accented speaker group as significantly more accented than did the low L1-use listener group for the more phonologically difficult sentences. There were no differences found between listener groups in how they rated the modera tely and heavily accented speakers in terms of the phonological difficulty of the stimuli. Generally, the listen ing groups performed similarly, with the exception of their ratings of the slightly accented speaking group. Although these results were not expected, th ey are nonetheless compelling and require further attention and analysis.
48 Less Difficult Sentences More Difficult Sentences Figure 2. Mean Ratings of Accent by Both Sp eaker and Listener Groups for Both Less and More Difficult Sentences 1 2 3 4 5 6 7 8 9Slight AccentModerate AccentHeavy AccentSpeaker GroupsRatings of Degree of Accentedness 1 2 3 4 5 6 7 8 9Slight AccentModerate Accent Heavy AccentSpeaker GroupsRatings of Degree of Accentedness Native English Listeners Low-use Bilingual Listeners High-use Bilingual Listeners
49Analysis of Individua l SpeakersÂ’ Ratings In order to determine whether there we re differences between the individuals within the speaker groups that may have attributed to the ef fects found among speaker groups, an analysis of variance (ANOVA) usi ng a hierarchical design was run analyzing effects of speaker group, individual speaker nested in speaker group, listener group, and sentence difficulty factors on ratings of accen tedness. There was a significant two-way interaction between indivi dual speaker number and speaker group. As suggested by Figure 3, there were significant differences in perceived degree of accent between the two speakers in the heavily accented group, F (2, 198) = 13.52, p < 0.0001, = .12 All listeners rated these two speakers as the mo st heavily accented, w ith one speaker (Heavy 1) rated as more accented than the other (H eavy 2). Speaker Heavy 1 had a mean accent rating of 7.54 while speaker Heavy 2 had a mean accent rating of 6.2. Listeners tended to rate speakers in the other two groups similarly. In fact, bot h speakers in the moderately accented speaking group had a mean accent rating of 3.57. Generally, the results echoed those found for the analysis by speaker group.
50 Figure 3. Mean Accentedness Ratings of Each Speaker Another two-way interaction was found between speaker group and listener group, F (4, 198) = 2.636, p < .035, = 05. The high L1-use bili nguals rated the slightly accented group as significantly more accented than did the native English listener group. There was also a significant three-way inte raction between listener group, speaker group, and difficulty level, F (4, 198) = 3.376, p < .011, = .064. Tests of Least Significant Differences revealed that the native English listener group rated the moderately accented speaker group as significantly more accented on the less difficult sentences compared to the more difficult sentences, see Figure 4. Also, the low L1-use bilingual listener group rated the heavily accented speaker group as significantly more accented on the more difficult sentences compared to the less difficu lt sentences. Finally, there was a threeway interaction found between speaker group, in dividual speaker, and difficulty level, F (2, 198) = 4.293, p < .0001, =.173. As Figure 5 suggests, LSD tests revealed that 1 2 3 4 5 6 7 8 9 Heavy 1Heavy 2Mod 3Mod 4Slight 5Slight 6SpeakerTypeRatings of Accentedness
51speaker 2 (heavily accented speaker 2) was ra ted as significantly more accented on the more difficult sentences, speaker 3 (mode rately accented speaker 1) was rated as significantly more accented on the more difficult sentences, and speaker 4 (moderately accented speaker 2) was rated significantly mo re accented on the less difficult sentences. Figure 4. Mean Ratings by Listener and H eavily and Moderately Accented Speaker Group for Less and More Difficult Sentences 1 2 3 4 5 6 7 8 9 Heavy Less Difficult Heavy More Difficult Moderate Less Difficult Moderate More DifficultRatings of Degree of Accentedne s Native English Low-use bilingual High-use bilingual
52 1 2 3 4 5 6 7 8 9 Heavy 1Heavy 2Mod 3Mod 4Slight 5Slight 6 Individual SpeakersRatings of Accentedness Less Difficult More Difficult Figure 5. Mean Accent Ratings for Indivi dual Speakers for Less and More Difficult Sentences Analysis of SpeakersÂ’ Pr oduction of Target Stimuli The analysis of production errors was begun by calculating th e percentage of targets correct in each sentence produced by each speaker. In order to determine whether the speaker had produced the target diffe rently from native speaker norms, the investigator and an independe nt rater listened to the reco rdings over headphones using the digital waveform editing program PraatÂ™, whic h also provides a visual display of the speech samples showing both waveform and spect rogram. The targets were then marked as correct or incorrect producti ons and tallied. An independent rater listened to the entire sample of the more heavily accented speaker (Heavy 1) and 10% of the sentences of the remaining speakers. Agreement between th e two raters was 7 9.76% for the heavily accented speakerÂ’s sentences and 88.10% for the remaining speakersÂ’ samples. In cases of disagreement, the raters conferred until agreement was reached. Total differences in production compared to native norms were obt ained by sentence (see Figure 11). The
53slightly accented speakers produced the fewe st errors. The moderately accented group produced more errors, although they differed onl y somewhat from the slightly accented speaker. The heavily accented speaker group pro duced the most target errors for both the less and more difficult sentences. The analys is by individual speaker yielded similar results as the analysis by speaker group. Figure 6. Mean Percent Target Errors by Individual Speakers and Sentence Type To further investigate the number and type s of differences from native norms produced by each speaker, the percentage of correct pr oductions for each target were calculated for each target for the less and more difficult sentences. To test the assumption that the speake rs produced more errors on the more difficult sentences and that th ese errors included the targ ets, the productions of the targets by the individual speakers were inve stigated. Table 3 and Figure 6 show the distribution and proportion of the targets for the less and more difficult sentences. 0 20 40 60 80 100Slight 1Slight 2Mod 1Mod 2Heavy 1Heavy 2SpeakersMean Percent Targets Produced Different from native Norms Less Difficult More Difficult
54Generally the targets that were most in erro r by all speaker groups we re the tense and lax vowel distinctions, the intervoc alic voicing distinctions betw een /s/ and /z/, and lexical stress differences. However, the heavily accented speaker group produced numerous errors on all the targets except for final cons onant deletion. It should be noted that the prediction that speakers would consistently make these e rrors in all contexts was not supported by the data. Most of the time, speakers tended not to commit the expected errors on the targeted English words. Perhap s this explains why th ere were no significant differences found between the accentedness ratings for the less and more difficult sentences. Target # Targets of This Type % of Total Targets # in Less Difficult Sentences % in Total Less Difficult Sentences # in More Difficult Sentences % in Total More Difficult Sentences EP 12 14.81 4 13.79 8 15.38 ED 9 11.11 4 13.79 5 9.62 RD 5 6.17 1 3.45 4 7.69 TL 15 18.52 4 13.79 11 21.15 CO 8 9.88 4 13.79 4 7.69 SC 13 16.05 4 13.79 9 17.31 SZ 8 9.88 2 6.9 6 11.54 LX 11 13.58 6 20.69 5 9.62 Totals 81 29 52 EP=epenthetic schwa; ED= -ed ending; RD= vowel reduction; TL= tense-lax; CO= final consonant deletion; SC= /-t /; SZ, /s-z/; LX= lexical stress. Table 3. Distribution of Targets For Less and More Difficult Sentences
55 Less Difficult Sentences More Difficult Sentences Figure 7. Mean Percent Target Errors for Less and More Difficult Sentences EP=epenthetic schwa; ED= -ed ending; RD= vowel reduction; TL= tense-lax; CO= final consonant deletion; SC= /-t /; SZ, /s-z/; LX= lexical stress 0 20 40 60 80 100 EPEDRDTLCOSCSZLX Slight Accent Moderate Accent Heavy Accent 0 20 40 60 80 100 EPEDRDTLCOSCSZLX Slight Accent Moderate Accent Heavy Accent
56It was expected that speakers would inse rt an epenthetic schwa before consonant clusters in English. The heavily accented sp eaker group produced this target differently from native norms 75% of the time for the less difficult sentences and 25% of the time for the more difficult sentences. Interestingl y, insertion of the epen thetic schwa seemed only to happen when the cluster was preced ed by a consonant. For example, in the sentence, ItÂ’s not easy to learn Spanish an epenthetic schwa was inserted before the word Spanish by the heavily accented speakers However, in the sentence, She liked the crazy spider the target was not inserted si nce the phoneme /i/ preceded the word spider Perhaps the epenthetic schwa was not insert ed before clusters that were preceded by vowels because the vowel served to facilitate the production of the cluster in much the same way as the schwa would have. The next target investigated was the Â–ed ending (ED), in which the bilingual speakers were expected to produce / d/ in regular past tense word s instead of /d/ or /t/. For example, the word liked was expected to be produced /lik d/ rather than /likt/. The heavily accented speakers produced this target error 25% of the time for the less difficult sentences and 30% of the time for the more difficult sentences. The moderately accented group did not produce the ta rget error at a ll in either the less or more difficult sentences. The slightly accented speaker group did not pr oduce the target error in the less difficult sentences and only 10% of the time for the mo re difficult sentences. It was suspected that the orthography may have elicited inco rrect production in some of the words. Because the speakers were reading the sentences, they may have pronounced / d/ in words that were unfamiliar to them, like stalled. In contrast, they did not produce the target error when reading more familiar words like closed and called
57 It was also expected that the bilingua l speaker groups would produce the vowels in unstressed syllables fully (RD), e.g. mel[o]ns for melns. However this error was only produced by the heavily accented speaker group in the more difficult sentences and was not seen in the slightly and modera tely accented speaker groups. The heavily accented speaker group also made more errors than the other speaker groups on the other vowel target, the distincti on between tense and lax vowel production (TL). It was expected that the bilingual speakers would produce lax English vowels as tense vowels, e.g., ch[i]p for ch[I]p The slightly and moderately accented speaker groups only produced this in the more difficult senten ces, possibly due to the phonetic context. It was further expected that the bilingua l speakers would drop the word final /s/ or /z/, but none of the bilingual speaker groups produced errors on this target. It was also expected that the speakers would substitute /t / for the English / /, but again very few errors were produced on this target. This error only occurred when the target was followed by a tense vowel, e.g. ship It was not seen when the target was followed by a lax vowel, as in shop Interestingly, two of the speakers produced / / for /t / on the words stopwatch and ch osen One of these speakers was from the Dominican Republic and the other was from Mexico City. It is not understood why these speakers made this substitution, especially since their dialec ts differ and the other speaker from the Dominican Republic did not make this error. The consonant target that did yield a mo re significant number of errors was the voicing distinctions between in tervocalic /s/ and /z/, e.g., free[s]er for free[z]er (SZ). The moderately and heavily accented speaker groups were not able to produce these targets correctly at all in the less difficult sent ences. They were more successful in their
58productions in the more difficult sentences. Th e slightly accented speaker group also saw a higher success rate for this target in the mo re difficult sentences versus the less difficult, possibly due to the phonetic c ontext. Perhaps this help s explain why there were no predicted significant differences found between the less and more difficult sentences; at times, the speakers made more errors on the less difficult sentences. Hence, instead of being influenced by the number of targets present, accentedness may be more wordor context-specific. Finally, it was expected that the sp eakers would produce the multi-syllabic English words in the stimuli with different stress patterns (LX) than would native speakers. All speaker groups made errors on th is target in both th e less and more difficult sentences. A descriptive exploratory analysis wa s conducted to determine whether the listeners based their ratings of accentedness on the number of target errors per speakers. To determine this, the total number of target errors per sentence was related to the mean ratings assigned to that sentence by the lis tener groups. Figure 8 shows the errors per sentence. Where no bar line appears, no erro rs were made by the speaker for that sentence. The heavily accented speakers made most of the errors in both the less and more difficult sentences. An examination of the ratings assigned to each sentence by the listener groups indicated that the listeners were not rating the accents of the speakers based on the number of target errors produced in each sentence. To illustrate this point, consider the first heavily accented speaker w ho produced none of the targets correctly in sentence number seven of the less difficult sentences. However, she received ratings comparable to those given to sentence one of the less difficult sentences in which she
59produced few errors (see Figure 9). The arbitr ary nature of the ratings suggests that the listeners rated the sentences based on some thing other than the correct or incorrect production of the targets. Perhaps, the lis teners rated the sentences based on the speakersÂ’ productions that diffe red from native speaker norms but were not target items not included in the analysis.
60 Less Difficult Sentences More Difficult Sentences Figure 8. Mean Percent Erro rs by Individual Speakers for Less and More Difficult Sentences 0 20 40 60 80 100 12345678910Sentence NumbersPercent Differences on Targets 0 20 40 60 80 100 12345678910 Sentence NumbersPercent Differences on Targets Heavy 1 Heavy 2 Mod 1 Mod 2 Slight 1 Slight 2
61 Figure 9. Frequency Distribution of Ratings Given to Less Di fficult Sentences 1 and 7 of Heavily Accented Speaker 1. Summary Generally, listeners tended to rate the sp eakersÂ’ degree of foreign accent similarly. Amount of daily L1 use was only a significant variable in the rati ngs of the slightly accented group; here the high L1-use bilingu al group gave significantly higher accent ratings than did the native English gr oup, regardless of the level of phonological difficulty. As was evident in the data, list eners were able to differentiate slight, moderate, and heavy accents. Listeners also rated one of the speakers of the heavily accented speaker group as significantly more accented than the other, although this speaker group made roughly the same number of target errors. The level of phonological difficulty in each sentence was not a vari able that was generally responsible for differences in accent ratings. The speakers did not make the predicted errors on the 0 3 6 9 12 15 18 123456789Accentedness RatingsNumber of Raters Sentence 1 Sentence 7
62targets on most occasions. In fact, the moderately and slightly accented speaker groups made very few of the predicted errors. Consequently, listeners may have rated the accentedness in the sentences based on somethi ng other than the number of target errors contained in each. Generall y, all listening groups rate d the speaking groups as significantly different from each other, with the least accented sp eakers receiving the lowest accent ratings and the most h eavily accented group receiving the highest.
63 Chapter 4 Discussion This study was undertaken to determine wh ether the degree of perceived foreign accent in an L2 is related to the listenerÂ’s amount of L1 use and the phonological difficulty of the stimuli. It was hypothesized that individuals w ho used their L1 less would be better able to gauge the degree of foreign accent in nonnative speakers of their L2 than would individuals who use their L1 more often. This hypothesis was only partially supported. Originally, it was predicte d that the high-L1 use listeners would rate the speakers as less accented compared to the ratings given by the low-L1 use listeners. Although the high-L1 use listeners in this study rated the slightly accented speakers as more accented than did the native English listener s, the results can stil l be interpreted to indicate that the high-L1 us e listeners did not possess e nough information about how the phonetic segments in English should sound, wh ich was reflected in the differences in their rating patterns compared to the native E nglish listeners. This finding agrees with previous studies showing that as individuals gain experience with the L2, in this study defined as amount of L1 use, they are better able to gauge the degr ee of foreign accent in the L2 (Flege et al., 1997).
64Additionally, it was predicte d that the level of phonol ogical difficulty of the stimuli would affect the speakersÂ’ producti on accuracy and would also influence the degree of perceived foreign accent, with mo re difficult sentences eliciting more production errors and higher accent ratings Difficulty was determined by how many targets were included in the stimuli. The le ss difficult sentences were predicted to be perceived as less accented by both bilingual lis tener groups, with the low-L1 use listener group perceiving them as more accented than the high-L1 use listener group. Finally, it was predicted that the more difficult sentence s would be perceived as more accented by all listener groups, specifical ly, the low L1 use listener group was predicted to give higher accent ratings than the high L1 use group. The hypothesis that the more difficult sentences would receive higher a ccent ratings was not supported. Differences Between Listener Groups The results of this study support FlegeÂ’ s Equivalence Classification hypothesis (Flege et al., 1997), which contends that t hose who rarely use th eir first language are better able to gauge non-native accents in th eir target language than are those who use their first language often. The non-native listeners who sp oke their first language less often than their second language became more experienced in English, and they gained more information concerning how the phonetic segments in English should sound. When the accents did not differ significantly from native productions, it became more difficult for the high L1-use listeners to accurately gauge the degree of foreign accent. Flege (1988) hypothesized that nativ e speakers develop detailed phonetic prototypes against which to judge goodness of phones produced by non-native speakers. For non-native speakers, the ability to gauge a foreign accent in English senten ces is a skill that develops
65slowly with English-language experience. With this experience, adult L2 learners become better able to detect a foreign accen t and gauge its strength by establishing these phonetic prototypes. In this study, listener groups differed signi ficantly in how they rated the slightly accented speaker group. The high-L1-use bi lingual group was able to distinguish between all listener groups in the same way as the other liste ner groups; that is, they gave the slightly accented speaker group the lowest ratings and the heavily accented group the highest ratings. However, the high L1-use bilingual group rated the slightly accented speaker group as significantly more accented th an did the native English and low L1-use listener groups. An explanation consistent with both the hypothesis and previous literature suggests that as L2 learners gain experience with the targ et language, they are better able to gauge accents in that la nguage (Flege, 1984). Although the high L1-use listeners gave higher accent ratings to the slightly accented speaker group, this did not necessarily mean that they were more sens itive to the phonetic and prosodic deviations from the target norms. A possible alternative is that when the speakersÂ’ accents differed greatly from the native norms, as was the case for the heavily and moderately accented speaker groups, the high L1-use bilingual gr oup could more easily gauge their degree of foreign accent and thus, performed similarl y to the native English and low L1-use bilingual listener groups. However, when the accents differed very little from the native norms, as was the case for the slightly accented speaker group, the high L1-use bilingual listener group was not able to detect the de partures from the target language phonetics, and therefore rated the slightly accented sp eaker group differently from how the native English and low L1-use list ener groups rated them.
66This makes sense when one recalls th e Equivalence Classification hypothesis (Bohn & Flege, 1990; Flege & Eefting, 1987). This hypothesis states that for both perception and production, sounds that are identi cal in two languages are unlikely to cause a problem, but similar sounds might of fer persistent although subtle problems for the second language learner. Similar sounds should be most difficult because they will probably be substituted by a similar sound fr om the first language, even after the L2 learner has gained considerable experience wi th the target language. Sounds that are similar between languages are typically vowel s, but may also include consonants. Sounds that are completely new, in the sense th at they are not equivalent or even similar to sounds from the individualÂ’s L1, will be esta blished into a new category as a result of phonetic learning that is not ha mpered by equivalence classification. As the amount of experience with the L2 increases, indivi duals will produce second language vowels more like natives. According to the single system hypothes is, bilinguals have a single phonological system in which the phonetic inventories of both languages reside and they cannot fully isolate either phonetic system (Guion et al., 2000). The hypothesis also asserts that the less L1 there is, the smaller will be its influe nce on the L2 (Flege et al., 1997). According to Grosjean (1992), the L1 phonetic system infl uences that of the L 2, and the nature of this influence depends on several variables, including the amount and type of use of each language. Generally, the more individuals sp eak their native language, the stronger will be their accent in their se cond language (Flege et al ., 1997; Guion et al., 2000). Furthermore, this relationship seems to be as ymmetrical; although L1 use has an effect on accent in the L2, the L2 typically has a much smaller effect on L1 production. The single
67system hypothesis also states that the loss of L1, through high use of the L2, may reduce the degree of perceived foreign accent in the L2. In terms of the relation between spee ch perception and speech production, the ability to gauge an accent and discern the phonetic differences between L1 and L2 sounds is determined by the individualÂ’s age of learning and the perceived amount of dissimilarity of L2 sounds from the closes t L1 sounds (Flege, 1995). The production of an L2 sound will correspond to the mental phonet ic category representation that the L2 learner has developed after exposure to the target language. That is why accurate perception of L2 sounds tends to precede th eir accurate production. However, the only way to really determine whether perception l eads production is to develop measures to assess L2 learnersÂ’ percep tion and production abilities. It would therefore be interesting to obta in accent ratings of the bilingual listeners in this study to determine the correlation betw een their ability to ga uge a foreign accent in their L2 and their ability to produce it. A study that tested this idea (Flege, 1988) used bilingual listeners who differed according to th e number of years they lived in the United States. The listeners rated the degree of fo reign accent in sentences spoken by individuals with varying degrees of accent. The ratings of the bilingual listeners correlated strongly with those given by native English listeners, with the highest corre lations between the more experienced listeners and the native En glish listeners. Following the perceptual tasks, the L2 production of the bilingual listen ers was then investigated. It was found that the more and less experienced bilingual list eners had accents that were judged to be equally strong. Thus, the more experienced bi lingual listeners were more perceptually sensitive to the phonetic features of English than the less experienced listeners. This may
68explain why the listeners in this experiment tended to perform simila rly to each other and to the native English listener s in the ability to gauge a foreign accent in speakers with moderate to heavy accents. It was when the speakersÂ’ production of the English phonetic features differed only slightly from native norm s, as was the case in the slightly accented speaker group, that the less experienced li steners had difficulty gauging the degree of foreign accent. This seems to be true even though the speakers did not make the predicted errors on the targeted features. Clear ly, all listeners were sensitive to features of the speakersÂ’ productions that signaled a foreign accent. An alternative explanation considers the reasons that the listeners in this study spoke either Spanish or English most of th e time. Perhaps the high L1-use group spoke Spanish most of the time because of their id entification with the culture of the L1. Likewise, the low L1-use listeners may want to assimilate more into the culture of the L2, prompting them to use their second language mo st of the time. The identification with the L1 culture could have resulted in the high L1-use listener group having recognized the slightly accented speakers as Latin, and rated them as accented, more accented than did the other listener groups. In other words, the detection of an accent, albeit slight, was enough to prompt the high L1-use listener s to rate the slight ly accented speakers significantly higher than the othe r listener groups. However, they maintained the relative differences in ratings between slight, m oderate, and heavily-accented speakers. In a tangentially related st udy, Bresnahan et al. (2002) f ound that participants with strong ethnic identity deemed American English, which refl ected their ethnicity, to be more pleasing and have higher status than the unintelligible foreign accent, whereas those with weak ethnic identity f ound unintelligible foreign accent more pleasing and attributed
69higher status to it than American English. Wh at is relevant to th e present study are the authorsÂ’ conclusions that peopl e with strong ethnic identity ma y be more attached to their ethnic group and, therefore, more likely to recognize a foreign accent as representing someone in an out-group. Perhaps that is what affected the ratings of the slightly accented speakers given by the high L1-use group in the present study. High L1-use listeners may have spoken Spanish most of the time because of their strong ethnic identity and, therefore, r ecognized the slightly accented speakers as members of a different ethnic group, as revealed by th eir dialects, and rated them accordingly. Remember, the speakers and listeners came fr om various countries of origin with different regional dialects. Perhaps the high L1-use listeners based their ratings on their recognition of the dialects of the speakers, so mething that would have been difficult for the native English listeners to do. Another possible explanation is that the ratings of the slightly accented speaker group given by the high-L1 use listener group ma y have reflected their bias. Remember, listeners were told that all the speakers we re non-native speakers of English. Also, the rating scale was labeled Â“slight Spanish acc entÂ” instead of Â“no Spanish accentÂ”. The high-L1 use listener group could have been thought that the slightly accented speakers sounded like native English speakers, but because they were told that all speakers were non-native, they rated them higher than they w ould otherwise. It would be interesting to see how the listeners would have rated the accents if the backgr ound of the speakers was more ambiguous. Likewise, it would have been useful to incl ude a native English speaker group and to modify the rating scale to have the label Â“no Spanish accent.Â” That
70would reveal whether the highL1 use listeners could disc ern between native English speakers and those with slight Spanish accents. Characteristics of the Stimuli The analysis then turned to the stimuli to determine whether aspects of the speakersÂ’ productions affected the listenersÂ’ ratings. It wa s predicted that the more phonologically difficult sentence s would result in more misp ronunciations, and that the errors made would vary according to the speak ersÂ’ level of accent, with strongly accented individuals producing more errors on the targets than the slig htly accented individuals. However, the results of the analysis reveal ed that the level of phonological difficulty in each sentence was not generally responsible for differences in accent ratings. The slightly accented speakers produced the highest percentage of correct targets for both the less and more difficult sentences. The moderately accented group had the next highest percentage correct targets, although they differed only so mewhat from the slightly accented speaker group. The heavily accented speaker group produ ced the most target errors for both the less and more difficult sentences. Although the speaker groups performed as expected in terms of which groups committed the most ta rget production errors, the speakers only made the errors some of the time. In fact, the moderately and slightly accented speaker groups made only a small percentage of th e predicted errors, and even the heavily accented speaker group made far fewer target errors than expected. Perhaps speakers were being overly careful in their productions because they were read ing sentences rather than speaking naturally in conve rsation. It is also possible that speakers of dialects not used in this study would have produced more ta rget errors than the speakers used here. However, the speakers within each speaker group were from different countries of origin,
71and there were no significant within group differences in te rms of number of target errors. Still, listeners were able to gauge th e degree of foreign a ccent in each of the sentences. Consequently, it a ppears that listeners tended to rate the accentedness in the sentences based on something other than the num ber of target errors contained in each. Flege (1984) argued that listenersÂ’ detecti on of foreign accent may be based on suprasegmental differences in prosodic features such as timing, stress, and intonation. Perhaps that is what listeners did here as we ll. Certainly, they did not rate the degree of foreign accent based on phonemic features targeted alone. For this to have been the case, one would expect the ratings to reflect the sp eakersÂ’ correct productions of the targets, as well as their incorrect productions. However, there seemed to be little relationship between the production accuracy in one sent ence and the ratings assigned to it by the groups of listeners. Alternativ ely, the listeners c ould have rated the degree of foreign accent based on features and items that were not measured in this study. Perhaps it would have been more useful to classify th e sentences as more or less difficult after a transcription was performed. A further explanat ion is that because the sentence structure varied a lot, some sentences may have been more difficult to process by listeners, so the errors may have been more or less distracting. This finding supports previous literatu re that attempted to discern what contributes to the perception of forei gn accent (Magen, 1998). In her study on the perception of foreign accented speech, Magen f ound that listeners rated degree of foreign accent based on factors other than just the targeted phonetic sounds or sequences of sounds. She assessed the contribution of vari ous factors to the pe rception of foreign
72accent by having listeners rate speakersÂ’ de gree of accent in phrases as originally produced and edited acoustically. The edited versions were intended to more closely resemble the productions of native American English speakers. The edited versions received significantly lower accent ratings fo r epenthetic schwa, -ed ending, tense-tax distinction, final /s/, /t /, and lexical stress for one of the speakers and epenthetic schwa, final /s/, and /t / distinction for the other speaker. Listeners were insensitive to voicing differences between the edited and uned ited samples, suggesting that listeners are not as likely to attend to voi cing distinctions when they ar e in the context of larger phonological distinctions. This may help expl ain the why ratings in this study did not reflect the number of target errors in each sentence. For example, the /s-z/ voicing distinction was the targ et most often in error for all sp eakers, yet its mispronunciation did not affect higher accent ratings. It seems po ssible for this study, as Magen contended for hers, that suprasegmental factors may have contributed to listen ersÂ’ perceptions of accentedness. In other words, listeners may have been attending to the prosodic features rather than the individual sounds in error. However, as has been noted, the listeners could have been attending to feat ures not measured in this study. Although when listeners were asked on what they based their ratings, nearly all responded that some of the words sounded Â“wrongÂ” while others sounded Â“right.Â” According to the listeners, sentences with wo rds that sounded Â“wrongÂ” were given higher accent ratings than those with words that sound ed Â“right.Â” However, an investigation into the ratings revealed that they did not always reflect the speakersÂ’ production accuracy. Indeed, sentences that had the most sounds in error were not necessarily rated as more accented than sentences that had the fewest target errors. For some targets,
73speakers made fewer errors in the more diffi cult sentences and more errors in the less difficult sentences. Perhaps this helps expl ain why there were no predicted significant errors found between the less and more difficult sentences. Reasons for this may be the phonetic context of the target wi thin the less versus more difficult sentences. It may be that certain phonetic features are more critical or weighted more heavily. It may also be the case that certain phonetic contexts facil itate correct production of the target and certain phonetic contexts make the target error more likely. For example, speakers tended only to insert the initial epenthetic scwha when the cluster was preceded by a consonant. Certainly, it remain s unclear as to what features the listeners attended that influenced their ratings. The listener groups seemed not to base their ratings on the presence of errors on the targeted features. Perhaps they rated degree of accent on phonetic features that were not controlled for, like VOT or vowel duration. An investigation of the phonetic c ontexts should, therefore, be conducted to determine which hinder or facilitate production. Future Directions In this study, there was great variability w ithin the listener groups in terms of how long they had been speaking English and their length of residence. For example, some listeners in the high L1-use listener group (t hose who spoke Spanish most of the time) reported having learned English as much as 15 years ago. Conversely, some listeners in the low L1-use bilingual group (those who spoke English most of the time) indicated that they had learned English as little as three years ago. This experi ment only controlled for the amount of first language use at the time of the experiment, rather than earlier L1 and L2 use or the aggregate amount of L1 and L2 us e over the listenersÂ’ lifetimes. Therefore,
74it is unclear whether the categ orization of listeners into high and low L1-use groups is fully accurate. Additionally, the listener groups were formed based on the self-reported amount of L1 use. Although th is is consistent with previ ous literature, self-report may suffer from respondentsÂ’ miscalculation of th eir language use and cause listeners to be incorrectly categorized. Considering that th is was a perceptual task, it may have been helpful to consider the amount of time listeners spent listening to the L2 also and the type of L2 input they received. Perhaps futu re studies should include questions on the language background questionnaire regarding amount of time listening to and speaking the L2 since the participantÂ’s first exposur e. Most ideally, a l ongitudinal study would elucidate how the nature of L2 input and th e amount of L2 listening and speaking affect the ability to gauge foreign accent in the L2. Considering the results, it may have been helpful to include a native English speaking group. The high L1-use listeners rated the slightly accented speakers significantly higher than did the native English listeners. It would, therefore, have been interesting to compare accent ratings given by the high L1-use listeners to native English speakers versus those given the slightly accente d speakers. Such a contrast would reveal whether the reason high L1-use listeners ha d difficulty gauging the degree of foreign accent in the slightly accented group was that the speakersÂ’ productions were so close to native norms. Finally, it may be helpful to perform ac oustic analyses on the speech samples in this study to quantifiably determine where the departures from the native norms occurred. It has been shown that even when listeners do not consciously detect phonetic differences between native and non-native listeners at the se gmental level, these differences can still
75lead to the detection of accen t (Bohn & Flege, 1990). This su ggests that listeners assess everything they hear in the speech stimuli, even acoustic aspects that are not overtly evident; this explains the listenersÂ’ ability to gauge the degree of foreign accent between speaker groups despite their lack of ta rget error productions. The list of phonemes targeted in this study was not exhaustiv e; perhaps many other potentially difficult phonemes existed in the sentences. Conclusions In summary, the results of this study suppor t the view that as adults use their L2 more often, the better able they will be to perceive and gauge degree of accentedness in the L2. This study was concerned with two major issues: whether ability to gauge the degree of foreign accent in an L2 improves as one uses their L2 more often compared to their L1, and whether the de gree of phonological difficulty would affect the production and perception of the stimuli. Effects of L1 use were found for thos e listeners who spoke Spanish most of the time and, therefore, had the least amount of expe rience with the L2 in terms of their daily language use. In general, the high L1-use bilingual group rated the slightly accented speakers as more accented than did the native English and the low L1use bilingual groups. Additionally, the degr ee of phonological difficulty did not affect the ratings of the sentences. Speakers di d not produce the errors as predicted and, subsequently, the listeners did not rate th e accentedness in the sentences based on the presence or absence of these errors. The imp lication is that listener s may be attending to something other than the segmental aspects of the stimuli. Perhaps, as has been suggested in previous literature listeners do not attend to segm ental features, but rather to suprasegmental features (Flege, 1984; Magan, 1998).
76 The results summarized provide sup port for the equivalence classification hypothesis. All of the bilingual listeners learned English afte r the age of twelve. Yet, they performed differently in their ability to gauge the degree of foreign accent in nonnative speakers of their L2. Th erefore, the notion that a critical period exists for the learning of an L2 can not be accepted in light of the present findings. Many factors in addition to age of learning differentiated th e listeners in this study. Those who spoke English most of the time performed more simila rly to the native Englis h listeners than did those who spoke Spanish most of the time. The finding suggests that amount of L2 usage has an effect on the ability to perceive L2 phones accurately. This is due perhaps to the refinement of intern al phonetic category representations, as described in the equivalence classification hypothe sis. However, the amount of daily L1 use was the only experience variable used to categorize the list eners. Certainly, it is possible that prior amounts and types of L2 use, prior and curre nt amounts of L2 listening, nature of L2 input, and social factors, like cu ltural identificat ion, affected the ability of the listeners to gauge foreign accent. The relative contributions of these factors s hould be explored and disentangled in future studies to provide a clearer understandi ng of the processes involved in L2 speech learning.
77References Anderson, J., & Koehler, K. (1988). The eff ect of foreign accent and speaking rate on native speaker comprehension. Language Learning, 38, 561-599. August, D. & Hakuta, K. (1998). Bili ngualism and second-language learning. In D. August & K. Hakuta (Eds.), Educating language-minority children (pp. 1119). Washington, D.C.: National University Press. Bent, T. & Bradlow, A. (2003). The interl anguage speech intel ligibility benefit. Journal of the Acoustical Society of America, 114, 1600-1610. Best, C., (1995). A direct realist view of cross-language speech perception. In W. Strange (Ed.), Speech perception and linguistic experience: Issues in crosslanguage research (pp. 171-204). Baltimore, MD: York Press, Inc. Birdsong, D. (1992). Ultimate attainment in second language acquisition Language, 68 706-755. Bohn, O., & Flege, J. (1990). In terlingual identification and th e role of foreign language experience in L2 vowel perception. Applied Psycholinguistics, 11, 303-328. Bresnahan, M., Ohashi, R., Nebashi, R., Li u, W., & Shearman, S. (2002). Attitudinal and affective response toward accented English. Language and Communication, 22, 171-185. Cummins, J. (2000). Language, power, and pedagogy: Bilingual children in the crossfire Buffalo, NY: Multilingual Matters. Flege, J. (1984). The detection of French accent by American listeners. Journal of the Acoustical Society of America, 76, 692-707.
78Flege, J. (1988). Factors affects degree of perceived accent in English sentences. Journal of the Acoustical Society of America 84, 70-79. Flege, J., Bohn, O., & Jang, S. (1997). Effect s of experience on non-native speakersÂ’ production and perception of English vowels. Journal of Phonetics, 25, 437-470. Flege, J., & Eefting, W. (1987). Production and perception of Eng lish stops by native Spanish speakers. Journal of Phonetics, 15 67-83. Flege, J., & Fletcher, K. (1992). Talker and list ener effects on degree of perceived foreign accent. Journal of the Acoustical Society of America, 91 370-389. Flege, J., Frieda, E., & Nozawa, T. (1997). Am ount of native-language (L1) use affects the pronunciation of an L2. Journal of Phonetics, 25 169-186. Flege, J., MacKay, I., & Meador, D. (1999a). Native Italian speak ersÂ’ perception and production of English vowels. Journal of the Acoustical Society of America, 106 2973-2987. Flege, J., Yeni-Komshian, G., & Liu, S. (1999b). Age constraints on second-language acquisition. Journal of Memory and Language, 41, 78-104. Gass, S., & Varonis, E. (1984). The effect of familiarity on the comprehensibility of nonnative speech. Language Learning, 34, 65-89. Gottfried, T., & Beddor, P. (1988). Perception of temporal and spectral information in French vowels Language and Speech, 31 57-75. Griffiths, J. (1967). Rhyming minimal contras t: A simplified diagnostic articulation test. The Journal of the Acoustical Society of America, 42 236-241. Grosjean, F. (1992). Another view of bilingualism. In R. Harris (Ed.), Cognitive processing in bilinguals (pp. 57-63). Amsterdam: North-Holland.
79Guion, S., Flege, J., & Loftin, J. (2000) The effect of L1 use on pronunciation in Quichua-Spanishbilinguals. Journal of Phonetics, 28 27-42. Hill, J. H. (1970). Foreign accents, language acquisition, and cerebral dominance revisited. Language Learning 20, 237-248. Iglesias, A., & Anderson, N. (1993). Dialectic al variations. In J. Bernthal and N. Bankson (Eds.), Articulation and phonological disorders (pp. 147-161) Englewood Cliffs, NJ: Prentice Hall. Johnson, J. & Newport, E. (1989) Critical period effects in second language learning: The influence of maturait onal state on the acquisition of English as a second Language. Cognitive Psychology, 21 60-99. Johnson, J. & Newpoot, E. (1991). Critical pe riod effects on universal properties of language: The status of subjacency in the acquisition of a second language. Cognition, 39 215-258. Jones, R. (1997). Beyond Â“Listen and repeat Â”: Pronunciation teaching materials and theories of second language acquisition. System, 25, 103-112. Krashen, S. D. (1973). Lateralization, language learning and the cri tical period: Some new evidence. Language Learning, 23 63-74. Lenneberg, E.H. (1967). Biological foundations of language. New York: Wiley. Levine, Hullutt, (2002). Long, M. (1990). Maturational constr aints on language development. Studies in Second Language Acquisition, 12 251-285. Magan, H. (1998). The percepti on of foreign-accented speech. Journal of Phonetics, 26 381-400.
80Matsura, H., Chiba, R., & Fujieda, M. (1999). Intelligibility and comprehensibility of American and Irish Englishes in Japan. World Englishes, 18 49-62. Myers, J., & Well, A. (1995). Research design and statistical analysis Hillsdale, NJ: Lawrence Erlbaum Associated, Inc. Munro, M., & Derwing, T. (1998). The effects of speaking rate on listener evaluations of native and foreign-accented speech. Language Learning, 48, 159-182. Munro, M., & Derwing, T. (2001). Modeli ng perceptions of the accentedness and comprehensibility of L2 speech. Studies in Second Language Acquisition, 23, 451-468. Munro, M.J., Flege, J.E., & MacKay, I.A. ( 1996). The effects of age of second language learning on the production of English vowels. Applied Psycholinguistics, 17 313334. Neufeld, G. (1979). Towards a theo ry of language learning ability. Language Learning, 29, 227-241. Patkowski, M. (1980). The sensitive period fo r the acquisition of syntax in a second language Language Learning, 30 449-472. Piske, T., MacKay, I., & Flege, J. (2001). Fact ors affecting degree of foreign accent in an L2: A review. Journal of Phonetics, 29 191-215. Scovel, T. (1969). Foreign accents, language acquisition, and cerebral dominance. Language Learning, 19 245-254. Snow, C., & Hoefnagle-Hohle, M. (1978). The critical age for second-language acquisition: Evidence from second language learning. Child Development, 49 1114-1128.
81Southwood, H., & Flege, J. (1999). Scaling foreign accent: direct magnitude estimation versus interval scaling. Clinical Linguistics and Phonetics, 13 335-349. Stockwell, R., & Bowen, J. (1965). The sounds of English and Spanish Chicago: The University of Chicago Press. Tahta, S., Wood, M., & Lowenthal, K. (1981) Foreign accents: Factors relating to transfer of accent from the first language to the second language. Language and Speech, 24, 265-272. Wijngaarder, S., Steeneken, H., & Houtgast, T. (2002). Quantifying the intelligibility of speech in noise for non-native listeners. Journal of Acoustical Society of America, 111, 1906-1916. Zecker, L. (2004). Learning to read and wr ite in two languages: The development of early biliteracy abilities. In C.A. St one, E.R. Silliman, B. Ehren, & K. Apel, (Eds.), Handbook of language and literacy : Development and disorders (pp. 248266). New York: Guilford Press.
Appendix A: Study Advertisement Earn Extra Credit Points !! Participate in a study on Foreign Accents Help us to understand the factors that influence foreign accent. We need monolingual English speakers and bilingual speakers whose first language is Spanish and second language is English. If youÂ’re interested, please contact Astrid Doty at email@example.com
84 Appendix B: Practice Sentences 1. The dog bit the man. 2. The boy hits the ball. 3. It is a hot day. 4. The bike is red. 5. I like candy. 6. The cat is playing. 7. I feel sick today. 8. Sometimes we go to the park. 9. My blue car is fast. 10. Please donÂ’t eat the apples
85Appendix C: Stimulus Sentences Less phonologicalally difficult sentences LX EP CO The operator stands by. (7) LX RD CO The doctor had questions for me. (8) TL/SC LX He wished for conversation. (8) CO SZ EP ItÂ’s not easy to learn Spanish. (8) SC TL CO He showed it to ElizabethÂ’s mom and dad. (10) SC EP ED The shirt is stained with mud. (7) ED SZ He stopped freezing it for me. (9) LX EP ED Two automobiles stalled. (7) SC/TL LX The ship-builder and the sailor work. (9) LX ED TL The hospital was called by him. (8) More phonologicalally difficult sentences SC LX CO EP TL/SZ ED The shop-keeperÂ’s store is closed. (7)
86 SC ED LX/SC/TL She earned one fellowship that year. (8) TL/SZ/TL/RD/EP ED Opposition stalled the plan. (7) EP SC LX CO TL Stop and Shop supermarkets are big. (9) EP ED LX TL SC Steve liked good television shows. (8) SC ED EP SZ She passed the stopwatch to the boys. (8) SC ED SZ EP She liked the crazy spider. (7) SZ TL/RD/CO TL SC Frozen melons are in the shop. (9) SC RD/TL/LX/TL EP/TL/CO She was confident in her skills. (8) SC RD RD SZ EP She heard the professor chosen to speak. (10) EP=epenthetic schwa; ED= -ed ending; RD= vowel reduction; TL= tense-lax; CO= final consonant deletion; SC= / -t /; SZ, /s-z/; LX= lexical stress. Number in parentheses indicates nu mber of syllables per sentence.
87Appendix D: Instructions To Speakers Explanation of study. We interested in foreign accents and what makes them more or less noticeable. Language background questionnaire. LetÂ’s complete this questionnaire together that tells us a little about yourself and your Spanish and English use. Once I have looked it over, we can continue with the rest of the study. Recording of sentences. Have a seat in the booth. I am going to read you some instructions. Here is a list of ten practic e sentences and the 20 sentences we will use for the study. You are to put on these headphones an d then read the practice sentences three times each. After you finish reading the sentence once, take a breath or pause for just a moment and then go on to the next sentence. You will do the same for the 20 sentences that are part of the study. Le tÂ’s practice with the first five sentences. Please put on your headphones and make sure they are comfortable. Now read each sentence three times in a normal conversational rate and volume. (Experimenter may demonstrate with the first practice sentence if necessary. After the parti cipant completes the practice trials, ask if she has any questions before proceeding). Now letÂ’s go on to the next set of 20 questions and read them in the same way you did the five practice sentences. Remember, after you finish reading one sentence, please pause and take a breath before reading the next sentence. When you are finished reading all 20 sentences three times each you may leave.
88Appendix E: Instruct ions to listeners Explanation of study. We are interested in foreign accen ts and what makes them more or less noticeable. Language background questionnaire. LetÂ’s complete this interview that tells us a little about yourself and your Spanish and English us e. Once I have looked it over, we can continue with the rest of the study. Griffith Listening Test. For the first part of the study, you will see four words displayed on the computer screen. You will hear one of these words spoken. Your job is to select the word you thought you heard by clicking on the word with the mouse. You will only hear the word once. You will do some prac tice items first. After you completed the practice, we will go on to the real items. When you have finished this part, feel free to take a short break and then return to the computer. Rating of sentences. You will hear sentences spoken by people who are native Spanish speakers who have learned English as a sec ond language. Your job is to rate each sentence according to how much of a Spanis h accent you think each person has. There are no right or wrong answers, just your opinion about how much of an accent you think each person has. You will use this scale on th e computer and click with the mouse to the very far left of the screen if you think the person speaks English with very little Spanish accent. You will click to the very far right of the screen if you think the person speaks English with a very heavy or strong Spanish accent. Clic k anywhere along the scale depending on how strong you think the speakerÂ’s accent is. Click Â“Acce ptÂ” at the top of the screen after you have rated the sentence. LetÂ’s practice first. (Present the ten practice trials.) You are to rate each one individually and not base your judgement on the one
89before it. Now let us get started on the real sentences. Remember, rate each sentence independently. You may drag the cursor anyw here along the scale to show how much of an accent you think the speaker has. You may advance to the next sentence by clicking on Â“AcceptÂ” at the top of the screen, the computer will not let you advance until it has played the entire sentence.