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 2200409Ka 4500
controlfield tag 001 002317557
007 cr cnu|||uuuuu
008 100902s2009 flua ob 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0003051
Doty, Astrid Zerla.
Spoken word recognition in quiet and in noise by native and non-native listeners
h [electronic resource] :
b effects of age of immersion and vocabulary size /
by Astrid Zerla Doty.
[Tampa, Fla.] :
University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 96 pages.
Dissertation (Ph.D.)--University of South Florida, 2009.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
ABSTRACT: In spoken word recognition, high-frequency words with few neighbors and less frequently occurring minimal pair neighbors (lexically easy words) are recognized more accurately than low-frequency words with many and more frequently occurring neighbors (lexically hard words). Bradlow and Pisoni (1999) found a larger easy hard word effect for non-native than native speakers of English. The present study extends this work by specifically comparing word recognition by non-native listeners with either earlier or later ages of immersion in an English-speaking environment to that of native English speakers. Listeners heard six lists of 24 words, each composed of 12 lexically easy and 12 lexically hard words in an open-set word identification task. Word lists were presented in quiet and in moderate noise. A substantially larger easy-hard word effect was obtained only for the later learners, but a measure of oral vocabulary size was significantly correlated with performance for the non-native listener groups only. Thus, the increased easy-hard word effect for non-native listeners appears to be explained as an effect of phonetic proficiency and/or vocabulary size on the structure of the lexical neighborhoods.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Co-Advisor: Catherine Rogers, Ph.D.
Co-Advisor: Judith Bryant, Ph.D.
t USF Electronic Theses and Dissertations.
Spoken word recognition in quiet and in noise by native and no n-native listeners: Effects of age of immersion and vocabulary size by Astrid Zerla Doty A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Psychology College of Arts and Sciences University of South Florida Co-Major Professor: Catherine Rogers, Ph.D. Co-Major Professor: Judith Bryant, Ph.D. Committee Member: Cynthia Cimino, Ph.D. Committee Member: Stefan Frisch, Ph.D. Committee Member: Douglas Nelson, Ph.D. Date of Approval: June 30, 2009 Keywords: phonological, neighborhood, proficiency, noi se, speech, perception Copyright 2009, Astrid Zerla Doty
Dedication I dedicate this dissertation to my family, especially . to Jay, for being my most adamant supporter and sharing in al l the uncertainties, sacrifices, and joy; to August and Ava, whose innocence and happiness revive me daily ; to Mom and Dad, your desire for my success and happiness made me strive for both; to Kurt, Heidi, Christine, Michele, and Eric for your e ncouragement and love. to you all, I give my deepest expression of love and appreci ation. ~A dream you dream alone is only a dream. A dream you dream together is reality. John Lennon (1940-1980)
Acknowledgements This dissertation was completed with the help of many a nd I want to express my sincerest gratitude for their assistance and support. It w ould not have been possible without the guidance of Dr. Catherine Rogers and Dr. Judit h Bryant. Dr. Rogers, my advisor and friend, for having faith and confidence in me. S he managed a fine balance between giving me the freedom to pursue my ideas and reining in my imagination when it got the better of me. Dr. Bryant was always avail able for me and her comments were extremely perceptive and helpful. Additionally, both wom en are two of the nicest people I have ever known. The members of my committee, Drs. Doug Nelson, Stefan Frisch, and Cynthia Cimino generously gave their time and expertise to be tter my work. I thank them for their contribution and their good-natured support. I mo st want to thank my family. I am indebted to my parents for instilling in me the dedic ation and discipline to finish whatever I start and to do it well. For my motherÂ’s m any home-cooked meals, for the care she gave my children during my long hours at the lab, for her tough but tender love and her tireless efforts, I am indeed thankful. My fa ther listened to me rave about my dissertation and shared similar stories of his dissert ation experience, always ending the conversations with some piece of sage advice or excrucia tingly dry humor. I want to thank my siblings, Kurt, Heidi, Christine, Michele, an d Eric for their support. I want to thank my husband and friend, Jay, for his love, sacrifice, and encouragement. I thank him for boosting my confidence, loving my eccentricities, and bringing joy to my life in so many ways. I also want to thank my children, August and Av a, who amaze and inspire me every day and have helped me keep perspective on what is important in life. I cannot thank them enough for showing me what it is to be a free spirit.
i Table of Contents List of Tables iv List of Figures v Abstract vi Chapter 1 Introduction 1 Cross-language studies of speech perception 2 Effects of phonological characteristic o n word recognition and word recall 7 The effects of cognitive demand and noise on word recognit ion and word recall 21 Semantic Characteristics 26 Chapter 2 Methods 31 Design 31 Participants 31 Stimuli 36 Speakers 36 Word lists 36 Recording procedures 37 Noise Mixing 39 Materials 40 Procedure 41
ii Chapter 3 Results 44 Word Recognition 45 Vocabulary, Language Scores, and Familiarity Ratings 50 Correlational Analyses 52 Regression Analyses 55 Digit Recall 57 Semantic Features 58 Chapter 4 Discussion 59 Phonological Neighborhood 60 Proficiency and Vocabulary Level 64 Noise 68 Digit Recall 68 Semantic Characteristics 71 Future Directions 71 References 76 Appendices 85 Appendix A: Monolingual Language Background Questionnaire 86 Appendix B: Bilingual Language Background Questionnaire 88 Appendix C: Easy and Hard Word Lists 90 Appendix D: Target Words Phonological Neighborhood and Semantic Features Data 91
iii Appendix E: Practice Words 94 Appendix F: Distracter Words 95 Endnote 96 About the Author End Page
iv List of Tables Table 1 Demographic information of individual earlier-lear ner 33 bilingual participants Table 2 Demographic information of individual later-learner bilingual participants 34 Table 3 Demographic information for the EL and LL bilingual groups 35 Table 4 Means of Word Recognition for Easy and Hard Words by Listener Group 46 Table 5 Means of Digit Recall by Listener Group 46 Table 6 Mean percentage of times groups chose a neighbor when in error during word recognition task 49 Table 7 Vocabulary and Language Tests, Familiarity Ratings and Demographic Variables 52 Table 8 Correlations among overall word recognition, voca bulary level (PPVT), receptive language scores (OWLS), and demographic variables (AOI and LOR) for all listeners 53 Table 9 Correlations between spoken word recognition, vocabulary, listening comprehension scores, word familiarity, and demographic variables for the EL group 54 Table 10 Correlations between spoken word recognition, voca bulary, listening comprehension scores, word familiarity, and demographic variables for the LL group 54
v List of Figures Figure 1 Percent correct word recognition for easy and hard words for all listener groups in quiet 47 Figure 2 Percent correct word recognition for easy and hard words for all listener groups in noise 47 Figure 3 PPVT and OWLS standardized scores for all listene r groups 51
vi Spoken Word Recognition in Quiet and Noise by Native and N on-native Listeners: Effects of Age of Immersion and Vocabulary Size Astrid Zerla Doty ABSTRACT In spoken word recognition, high-frequency words with few ne ighbors and less frequently occurring minimal pair neighbors (lexically eas y words) are recognized more accurately than low-frequency words with many and more fr equently occurring neighbors (lexically hard words). Bradlow and Pisoni (1999) found a large r easy hard word effect for non-native than native speakers of English. The pres ent study extends this work by specifically comparing word recognition by non-native liste ners with either earlier or later ages of immersion in an English-speaking environment to that of native English speakers. Listeners heard six lists of 24 words, each com posed of 12 lexically easy and 12 lexically hard words in an open-set word identification task. Word lists were presented in quiet and in moderate noise. A substantially larger e asy-hard word effect was obtained only for the later learners, but a measure of oral voc abulary size was significantly correlated with performance for the non-native listen er groups only. Thus, the increased easy-hard word effect for non-native listeners appears to be explained as an effect of phonetic proficiency and/or vocabulary size on the str ucture of the lexical neighborhoods.
1 Chapter One Introduction Non-native speakers of English constitute a rapidly grow ing minority in the United States. Many of them experience significant diff iculty understanding English, especially in less than ideal listening conditions, such as the classroom or work environment. One potential source is differences in word r ecognition due to 1) differences in neighborhood structure; 2) greater diffic ulty in phonetic discrimination; and 3) language competition. Researchers have found that th e effects of phonological word neighborhood on the word recognition abilities of bil inguals are greater than those found for monolinguals (Bradlow & Pisoni, 1999; Imai, Walley & Flege, 2005). A neighborhood is a collection of words that are phonetic ally similar to a given target word (i.e., they sound similar) and is composed of two parts: 1) the number and degree of confusability of words in the neighborhood, referred to a s Â‘neighborhood densityÂ’, and 2) the frequencies of the neighbors in language use, called Â‘ neighborhood frequencyÂ’ (Luce & Pisoni, 1986). The extent to which speech recognition is influenced by bot h phonological word neighborhood and increasing cognitive demand has yet to be investigated in either monolingual or bilingual populations. The present study compares the word recognition performance of monolinguals and bilinguals under conditions of increasing cognitive load, using words that vary in phonological neighborhood c haracteristics. Note that the term Â“recognitionÂ” refers to its traditional use in the s peech perception literature, which is typically considered Â“perceptual identificationÂ” in memor y literature.
2 Word recognition depends in part on the intelligibility of the stimuli, which is the degree to which something is capable of being understood. It is recognized that intelligibility of the stimuli and adultsÂ’ word recognit ion in particular have been shown to be influenced by many variables, such as the listenersÂ’ fam iliarity with various aspects of the signal (e.g., the speaker, the accent of the speaker the topic), the selection of stimuli, the task, the context, the rate of presentation of th e stimuli, the listening conditions including the presence of noise and the level of variability in the stimuli, and the degree of cognitive demand required in the listening task and/or simu ltaneous tasks. The variables of interest in this study include phonological wo rd neighborhood, age of immersion (AOI) of the bilingual listeners, presence of noise, and cognitive demand (in this case, a digit-recall task). In order to gain an unde rstanding of how these specific variables may influence the recognition of speech, the di scussion begins with an overview of nonnative speech perception. Following thi s, the influences of first (L1) and second languages (L2) on each other in cross-language spe ech studies are considered. Next, the effects of stimulus and lexical characterist ics on word recognition and recall are discussed. Finally, the effects of cognitive demand manipul ations on recognition and memory for synthetic speech are explained, and parallel s are drawn between findings from those studies and findings for native and non-native speech perception in noise. Cross-language studies of speech perception Cross-language studies of speech perception have shown th at adults are languagespecific perceivers. That is, although they are able to differentiate easily the phonetic categories of their native language, perception of non-nati ve phonetic contrasts is, in general, more difficult. Phonetic contrasts are pairs of sounds in a language that differ
3 along a given dimension, such as voicing. This language-spec ific pattern of performance is not due to loss in auditory sensitivity to the acoust ic features that differentiate nonnative contrasts (Best, 1995). Rather, it reflects the attunement of selective perceptual processes to the acoustic-phonetic information that is li nguistically relevant in the native language (Strange, 1999). However, a growing body of evidence shows that this result does not hold for all listeners, for all phonetic disti nctions, or all task conditions (Flege & Hillenbrand, 1984; Rochet, 1995; Strange, 1992; Strange, Bohn, Tren t, & Nishi, 2004). Clearly, an understanding of the variables used in speech perception experiments is crucial to understanding these differences and designing future studies. Also, given that selection of the languages in a cross-language investigation is based on phonetic and phonological characteristics, an understanding is necessa ry of the way that first and second languages influence and compete with one another. F or example, if testing the discrimination of a non-native contrast for Spanish tha n for Italian speakers, the targeted L2 sounds may be assimilated into one phonological categor y for the Spanish speakers and into two phonological categories for the Italian spea kers. This would presumably happen because of the closer similarity of the L2 sounds to L1 sounds for the Spanish compared to the Italian speakers. These different assi milation patterns would be predicted to result in better discrimination by the Itali an speakers (Best, 1995). Moreover, researchers may test non-native speakers who are unfamiliar with the target distinction or, alternatively, who are learners wit h varying degrees of proficiency in the language from which the distinction is drawn and thus have varying experience with the target distinction.
4 According to FlegeÂ’s Speech Learning Model (SLM) (1995), the pr ocess of equivalence classification prevents category formation f or L2 sounds that are perceived as similar to L1 sounds. The SLM has four basic tenets relevant to this discussion: 1) the processes used in learning the L1 sound system, including categor y formation, do not atrophy at adolescence as asserted by the critical peri od hypothesis; rather, they remain functional throughout oneÂ’s lifetime, 2) phonetic catego ries are the long-term languagespecific memory representations of speech sounds, 3) the phonetic categories acquired for L1 sounds will eventually come to reflect properties of b oth L1 and L2 sounds that are realizations of each category, and 4) bilinguals must work to maintain contrast between those L1 and L2 phonetic categories just as monolinguals m ust maintain distinctness among all L1 sounds. Moreover, formation of a phonetic category implies the perceptual ability to identify a wide range of phones as being the s ame despite auditorily detectable differences among them along dimensions that are not pho netically relevant, as well as the ability to distinguish the multiple exemplars of a category from realizations of other categories, even in the face of non-critical common alities. As the perceptual dissimilarity between an L2 sound and the closest L1 counterpart increas es, the probability of new category formation also increases. Further, the SLM asserts that the earlier in life L2 learning commences, the smaller the perceptual distance th at is needed between the L1 and L2 sound for L2 category formation to occur (Flege, 1995). Even if a new category is formed for an L2 sound, however, there is no guarante e that the category structure or weighting of acoustic cues will be the same as for mono lingual speakers of the target language, according to the SLM (Flege, 1995). Thus, a mismatc h may exist between a bilingual listenerÂ’s perceptual expectations for a target L2 s ound and the sound that is
5 actually produced by native speakers of the target language (I mai, Walley & Flege, 2005). This mismatch might explain reductions in L2 learne rsÂ’ spoken word recognition accuracy as compared to monolinguals, especially in conditi ons of noise or other degradation to the speech signal (Imai et al., 2005). Another claim of the model is that when a category i s not formed for an L2 sound because it is too similar to an L1 counterpart, the L1 and the L2 categories will assimilate, leading to a merged category. The sounds in this merged category should eventually come to resemble each other in production. A lternatively, when a new category for an L2 sound is established, it may dissimila te from neighboring L2 (or L1) sounds to preserve phonetic contrast of these sounds, whi ch exist in a common phonological space. Support for these hypotheses comes fr om Flege, Schirru, and MacKay (2003), which examined the production of the English /e I / vowel by native Italians who differed according to age of learning. Early bilinguals were found to produce English /e I / with significantly more formant movement than native English speakers. The exaggerated movement of /e I / produced by the early group was attributed to the dissimilation of a new phonetic category they ha d formed for English /e I / from Italian /e/. Many of these speakers may have establishe d a new category for English /e I / (which is produced with less formant change) and produced it w ith more movement than is typical for English in order to make it distinct from their Italian /e/. On the other hand, the undershoot of movement observed for the late bilinguals in the study was attributed to their failure to establish a new category for English /e I /, which led to the merger of the phonetic properties of English /e I / and Italian /e/ through the mechanism of category assimilation.
6 The Speech Learning Model also states that a bilingualÂ’s ph onetic representation of a target speech sound may be based on different featur es or feature weights than those of a monolingual speaker of the L2. Support for this hypoth esis comes from Flege, Bohn, and Jang (1997). Two sets of synthetic continua (11 spectral s teps created by changing the first and second formants plus 3 temporal steps) were used. Subjects had to identify the vowel in one set of continua as either Â“beatÂ” or Â“bitÂ” and as either Â“betÂ” or Â“batÂ” in the other set of continua. Vowel duration influenced the native English subjectsÂ’ identifications primarily for vowels at the middle of t he continua where spectral cues were insufficient to define a vowelÂ’s identity unambiguously When identifying members of the Â“bet-batÂ” continuum, the experienced nonnative speakers (who came from various L1s) made more use of spectral cues than did the inexperienced non-native speakers. They also made less use of temporal cues. In this case, the experienced nonnative speakers (also from various L1s), but not the ine xperienced non-native speakers, resembled the native English speakers because they were us ing the spectral and temporal cues in a similar way as the native English listeners Despite these broad categories of Â“experiencedÂ” versus Â“inexperiencedÂ”, it should be noted that the age of L2 acquisition, the degree of e xposure to the language, and experience with the L2 seem to be factors that may hea vily determine the relationship between perception of the first and second languages and may contribute to changes in these perceptual abilities over time. The perception of oneÂ’s second language might also differ according to the class of sounds, the acoustic a nd perceptual correlates of these classes, and to contextual effects. Although all of these variables did not need to be controlled in the present experiment, careful considerat ion was given to the listenersÂ’
7 proficiency level with English, and particularly age of immersion, because experience has been shown to affect perceptual abilities in the L2. Effects of phonological characteristics on recognition and recall One issue faced when investigating spoken word recognition is the structural relations among the phonological patterns of words in the mental lexicon. In addition to the contextual and stimulus factors that affect a wordÂ’s intelligibility, there are lexical factors that may increase or decrease the probability o r speed with which a listener will correctly identify a spoken word. In fact, it has been argued that the process of word recognition relies on accurate discrimination among competi ng lexical items (Luce & Pisoni, 1998). Thus, understanding the structural organization of words in memory and how these relations influence word recognition and lexical access is crucial to understanding how these factors may influence perception by L2 learners (Luce & Pisoni, 1998). According to the Neighborhood Activation Model (Luce & Pi soni, 1998), the number of similar competitors that a word has and their relative frequency in the language can have both inhibitory and excitatory effects o n lexical access. The claim is that spoken words are recognized in the context of phonolog ically similar words activated in memory; a spoken word activates a set or Â“neighborho odÂ” of similar sounding words in memory, which then compete for recognition. A simil arity neighborhood is defined as a collection of words that are phonetically similar t o a given target word. A similarity neighborhood is composed of two parts: 1) the number and degr ee of confusability of words in the neighborhood, referred to as Â‘neighborhood den sity,Â’ and 2) the frequencies of those neighbors, called Â‘neighborhood frequency.Â’ A neighbor of a given target word
8 is one that differs from the target word by a one phoneme addition, substitution, or omission. For example, some neighbors for the word Â“ satÂ” would be Â“stat, rat, sit, sap,Â” and Â“at.Â” The model proposes that the frequency of a gi ven word, the size of the wordÂ’s neighborhood, and the frequency of the words within that neighborhood will determine the probability of that word being selected over its cl osest phonological neighbors. The effects of phonological neighborhood on word recogni tion and word recall are particularly interesting. In spoken word recognitio n tasks, numerous studies have supported the predictions of the Neighborhood Activation Model (Goldinger, Luce, & Pisoni, 1989; Luce, 1986; Luce & Pisoni, 1998; Luce, Pisoni, & Goldi nger, 1990; Pisoni, Nusbaum, Luce, & Slowiaczek, 1985). For example, in a perc eptual identification task, words with low-density neighborhoods were found to be ide ntified in noise with greater accuracy than those from high-density neighborhoods. Acco rding to the Neighborhood Activation Model, the poorer identification of words from high-density neighborhoods is a consequence of their having more competitors so activatio n of the target much reach a higher level to overcome competition. Sommers (1995) provided further support for the Neighborhood Act ivation Model. He found that identification accuracy of easy wo rds was similar for young and older adults, but identification of hard words was signifi cantly worse for older adults whose identification accuracy was 15% lower for the har d words than for the easy words, compared to younger adults, for whom the difference was o nly 7%. Sommers argued that the older adults may have more than just overall reduced a uditory abilities; they may also have less ability to discriminate the sound patterns in t he speech signals, especially from among phonetically similar neighbors. The easy-hard wo rd effect was also
9 disproportionately greater for older listeners when the task demands were increased by switching from single to multiple talkers, which may sugges t an influence of greater processing demand in addition to an effect of auditory abil ities in the differences obtained between older and younger listeners. Unlike the findings from recognition studies, the effects o f neighborhood characteristics on word recall seem to differ from st udy to study and task to task. Generally, studies agree that there is better recall f or high-frequency words (Allen & Hulme, 2006; Goldinger et al., 1991; Roodenrys et al., 2003). Howe ver, some studies have found better recall for words with low-frequency ne ighborhoods (Goh & Pisoni, 2003; Goldinger et al., 1991), whereas others have found better recall for words with high-frequency neighborhoods (Roodenrys et al., 2003). Simila rly, some have found better recall for words from small neighborhoods (Goh & Pisoni, 2003; Goldinger et al., 1991) and others have found the opposite (Allen & Hulme, 2006; Ro odenrys et al., 2003). The differences seem to stem from the fact that some s tudies considered all three variables together (word frequency, neighborhood frequenc y, and neighborhood density) while others considered each variable separately. Also, authors of these studies used different cut-offs for determining Â“highÂ” versus Â“lowÂ” fr equency of target word and the frequency of its neighbors, as well as the density of the neighborhoods. For example, Goh and Pisoni (2003) used word sets that diffe red on neighborhood density and frequency, but were equated for word frequency. They found that recall was better for words from small, low-frequency neighborhoods t han words from large, highfrequency neighborhoods. The researchers argued that ther e is less lexical competition
10 among similar sounding traces for words from small neighb orhoods, which leads to less confusion among the candidates for reconstruction. Using words that differed systematically in word frequency a nd neighborhood size, Allen and Hulme (2006) found better recall for high-fr equency words and those from large neighborhoods compared to low-frequency words an d those from small neighborhoods. In fact, the words from large neighborho ods were recalled more accurately even though they were perceived less accurately. Thus, it appears that the recall differences between words from large and small neighborhoods do not depend upon differences in how well these words are perceived. However, the recall advantage for high-frequency over low-frequency words may depend in part on the greater ease of perceiving high-frequency words. The authors suggest that i t is the semantic representations that account for the differences in re call between words from large and small neighborhoods. Low-frequency words in high-density, high-frequency phonologi cal neighborhoods (i.e., words that occur relatively infreque ntly in the language and have many similar sounding neighbors that occur relatively oft en in the language) are predicted to be recognized less quickly and accurately tha n high-frequency words from low-density, low-frequency neighborhoods. Thus, the forme r are termed Â“hard,Â” whereas the latter are deemed Â“easyÂ” (Luce & Pisoni, 1986). In sum mary, Â‘easyÂ’ words are those that occur frequently in the language and have relatively f ew phonetically similar neighbors that are relatively low frequency. The Â‘hardÂ’ words, on the other hand, occur less frequently in the language and have many phonetically similar neighbors that are relatively high in frequency.
11 Goldinger et al. (1991) selected word sets that were easy to identify (highfrequency words from sparse, low-frequency neighborhoods) and hard to identify (lowfrequency words from dense, high-frequency neighborhoods) an d used them in a serial recall task. Results showed better recall performance for the easy-to-identify words than the hard-to-identify words. Unfortunately, it is not po ssible to draw any conclusions from this study about the influence of neighborhood charact eristics on verbal short-term memory performance because neighborhood characteristic s were confounded with word frequency. Further, the researchers assert that the degree of confusability for given words only conveys information about the listenerÂ’s internal lexicon and the relative accessibility of its component words. Other studies provide supporting evidence for a link between s peech perception processes and recall of for spoken words (Luce et al., 1983; Paris et al., 2000). These studies suggest that, when the encoding of words becomes dif ficult, memory performance for these words declines. For example, Roodenrys et al (2002) assessed immediate memory for word sets differing in frequency, neighborhood size, and average wordneighborhood frequency. When they considered just word f requency and neighborhood size, they found recall better for high frequency words a nd for words from large neighborhoods. When word frequency and neighborhood frequency were manipulated, they found that recall was better for high-frequency wor ds and words from highfrequency neighborhoods. Finally, neighborhood size and nei ghborhood frequency manipulations revealed better recall for words from highfrequency neighborhoods and for words from large neighborhoods compared to small. The effects were explained in terms of word frequency (the easy to perceive [high frequenc y] words were recalled more
12 accurately than the hard-to-perceive [low-frequency] wor ds. The researchers argued that memory was better for the high-frequency words because the ir representations in longterm memory are more accessible or better specified th an those of low-frequency words. Interestingly, words were more likely to be intruded upon by a neighbor if they were low frequency, had many neighbors, and if the average frequency of the neighbors was high (Roodenrys et al., 2002). Typically, when a neighborhood intrusion occurred the intruding neighbor was higher in frequency than the presen ted word, which seems to support the predictions of Neighborhood Activation Model. Roodenrys et al. (2002) argue that phonological information in LTM plays an ac tive role in recall in STM tasks, which helps explain the recall advantage for high-frequen cy words compared to lowfrequency words. In order to explain why words from large neighborhoods we re recalled better than words from small neighborhoods, Roodenrys et al. (2002) suggest that the finding reflected the role of speech-production processes (e.g., r etrieval of the speech motor programs for words that have to be articulated) in imme diate memory tasks, but not speech-perception processes. These experiments therefor e appear to provide evidence counter to the idea that word recall depends on a reintegr ation process which involves speech-perception mechanisms. The deleterious effects of a large neighborhood on word recognition seems to happen because the listener is require d to select a word from among a large number of competitors. On the other hand, the f acilitative effects of a large neighborhood on recall seem to happen because the neig hbors provide support by keeping the word in active rehearsal longer than a word with few neighbors. From the existing data it appears that the effects of word frequenc y and neighborhood size on recall
13 are robust, but the effect of neighborhood frequency is small and inconsistent across experiments. Although the studies discussed above have been limited to m onolingual participants, a few studies have explored the contributi ons of lexical characteristics on spoken word recognition for non-native listeners. Bradlow and Pisoni (1999) investigated the combined effects of talker-, listener-, an d item-related factors on isolated word recognition. The researchers had ten monolingual English-speaking tal kers record both Â“easyÂ” and Â“hardÂ” lists of words at three differen t rates (slow, medium, and fast). The authors selected these words so as to differ accordin g to three lexical characteristics. The easy words occurred more frequently in the language, th eir mean neighborhood density (the number of phonetic neighbors) was lower than those of the hard words, and the mean neighborhood frequency (the mean frequency of th e neighbors) of the easy list was lower than that for the hard list. Further, the frequency counts from the Brown Corpus of prin ted text (Kucera & Frances, 1967) were used to examine the words to determine tha t the easy list words had a significantly higher mean frequency of usage in the lan guage than did the hard list words (185.24 with a range of 36-895 versus 4.21 per million with a range of 1-35, respectively). Second, the words on the easy list were selected so that their mean neighborhood density (the number of phonetic Â“neighborsÂ” ) was lower than that of the words on the hard list (13.34 neighbors with a range of 3 -19 versus 26.96 neighbors with a range of 21-39, respectively). Bradlow and Pisoni (1999) used t he definition by Greenberg and Jenkins (1967) of a neighbor of a given target word as one that differed from the target word by a one phoneme addition, substitutio n, or omission. Finally, the
14 words on the easy list were selected so that their me an neighborhood frequency (i.e., the mean frequency of usage of the neighbors of the target wor d was lower than that of the words on the hard list (37.50 per million with a range of 2. 33-79.67 versus 282.2 per million with a range of 87.22-1066.59, respectively). Further, the familiarity of each of the words was assessed and all were judged to be highly famili ar to native-Englishspeaking adults. That is, they all received a rating of a t least 6.25 on a 7-point scale, with 1 being lowest familiarity and 7 being highly familiar (Nusbau m, Pisoni, & Davis, 1984). In summary, the easy words are those that occur frequen tly in the language and have few phonetically similar neighbors that are mostly low-fr equency. The hard words, on the other hand, occur less frequently in the language and have ma ny similar neighbors that are mostly high in frequency. Each of the listeners in the study by Bradlow and Pisoni (1999) heard the full set of 150 words spoken by a single talker at a single rate: they heard lists of words and were required to type the words they thought they heard on a com puter keyboard. Note that in order for a participant to recognize a hard word, he or sh e had to discriminate among a large set of alternatives and, necessarily, needed to be a ble to make finer phonetic distinctions among words at the segmental level because the hard words had more similar sounding neighbors that were also more frequent in the l anguage, relative to the easy words. Overall, recognition scores were significantly hi gher for the easy words. The authors argued that this effect of lexical discriminabili ty resulted from the listenersÂ’ knowledge of the sound-based structure of the lexicon. In ge neral, the results for monolinguals from this experiment replicate those of previo us studies (Luce, 1986; Luce & Pisoni, 1998; Luce et al., 1990; Pisoni et al., 1985) and support the assumptions of the
15 neighborhood activation model of spoken word recognition; wo rd recognition takes place within the context of the mental lexicon and, therefor e, is influenced by other phonetically similar words (Luce & Pisoni, 1998). In a second experiment, Bradlow and Pisoni (1999) found that the easy/hard word effect was greater for non-native listeners than for native listeners. In this experiment, the listeners again heard a word over headphones and typed what they heard into a computer keyboard. Two separate lists were used, one produce d by a single talker and the other produced by multiple talkers. Within each list, half of the words were easy and half were hard. For both the native and the non-nativ e listeners, the overall percent correct was higher for the single-talker condition and for the easy words. Native listeners recognized words with greater accuracy than non-natives. However, the difference in percent-correct word recognition between the easy and har d words (i.e., the easy-hard word effect) was several times greater for the non-n ative than for native listeners. Perhaps, as is consistent with theories of non-native speech perception (Best, 1995; Flege, 1995), non-native listeners have greater difficulty recognizing words that require perception of fine phonetic detail for discrimination bec ause they may not have acquired all the native cues or do not have the same cue weightin g as native speakers. Moreover, Bradlow and Pisoni (1999) asserted that, because non-native listeners have more difficulty with hard words than easy words, just as do na tive listeners, their results support the idea that the non-native speakers develop lexicon s of their second language by employing the same sound-based organizational principles a s native listeners. Additionally, the authors administered a measure of word f amiliarity in order to assess the familiarity of the non-native listeners with the target words. For this task,
16 participants used a 7-point scale to rate their familiari ty with a list of English words presented on a computer screen. Because the hard words oc cur less frequently in the language, one possibility the authors considered was that t he non-natives were simply less familiar with the hard words and therefore were un able to recognize them accurately. The pattern of familiarity ratings given by the non-nat ives paralleled those of the native listeners: higher familiarity ratings were assigned t o the easy words and lower familiarity ratings were given to the hard words. Generally, compar ed to the native listeners, the non-native listeners rated themselves as much less fa miliar with the hard words, and this was reflected in their recognition scores as well. Wh en familiarity was controlled, however, by using only words rated as highly familiar to both native and non-native listeners in the analysis, a stronger easy/hard word effect for the non-native listeners than for the native listeners was still observed. Thus, al though part of the non-nativesÂ’ difficulty in recognizing hard words might have stemmed from their lack of familiarity with the words, familiarity alone does not fully accoun t for the effect, suggesting that decreased discrimination of fine phonetic detail or other factors may also play a role. Furthermore, a test of subjective familiarity of the target words may not fully reflect non-native listenersÂ’ lexicon. First, the no n-natives might have recognized those words in spoken form but have more experience and familiari ty with the words in print form than in spoken form. Garlock, Walley, and Metsala (2001) describe familiarity as encompassing two constructs: experienced frequency and age-of-ac quisition. The authors give an example using the word Â“cartoon.Â” This wor d, they argue is acquired early by most children, but it may not be encountered all that frequently by either children or adults. On the other hand, the word Â“cartilage ,Â” they maintain, is encountered
17 later in life, but may be used frequently by individuals in certain professions, such as doctors. The authors argue that high-frequency words over lap with other words on a segmental basis more often than do lower-frequency words and these neighbors tend to be high-frequency as well. This means that the neighborhoo d density and frequency characteristics contribute substantially to perceived wor d familiarity. Furthermore, a rating of an Â“8Â” given by a native English speaker may be very different than a rating of Â“8Â” given by a bilingual who learned English late in life. Also, it is not only the familiarity with the target words that is of concern. Rather, the listenersÂ’ knowledge of the words in the neighborhood of the targets is also of interest because the number of neighbors of a target word known by the subject could drama tically alter the structure of that neighborhood. For example, listeners with small er vocabularies may know fewer of the low-frequency neighbors of easy words but more of th e high-frequency neighbors of the hard words, thereby increasing the easy-hard word effec t. Spoken words are recognized by native listeners in the cont ext of other words in the mental lexicon, and words requiring fine phonetic discr imination (i.e., hard words) are more difficult to recognize than words that do not require a high level of phonetic discrimination (i.e., easy words) (Luce & Pisoni, 1998). T his can be especially true for non-native speakers (Best, 1995; Flege, 1995). However, in wor d recognition tasks, listeners are doing more than just discriminating among pho nemes. They must discriminate among lexical items. Thus, it seems unlikely that spoken word recognition is accomplished solely by phonetic discrimination, rath er, the stimulus input may activate a number of similar acoustic-phonetic representations and recognition must necessarily involve discrimination among lexical items (Luce & Pisoni, 1998). Therefore, it is
18 reasonable to assume that non-native listeners would hav e greater difficulty discriminating among hard words than among easy words because the hard words may be less familiar to them. Further, hard words would have mor e high frequency neighbors that might also be more familiar to the learner. T hus, the relative structure of the neighborhood for a bilingual might differ substantially and systematically from that of a native speaker. Such differences in neighborhood structur e, based on vocabulary size, may explain how the size of the easy-hard word effec t might increase for non-native listeners relative to native listeners. That is, ea sy-word neighborhoods may be effectively smaller for some non-natives because the target words themselves are likely to be known to them and fewer of the neighbors are likely to be know n than for monolinguals. Hardword neighborhoods on the other hand may be of similar s ize for both native and nonnative listeners, but the words themselves less may be less familiar to the non-natives. Thus, the relative difference in neighborhood size could be increased for non-natives with substantially smaller vocabularies. No studies to date have investigated the extent to which voc abulary size may contribute to the processing of speech by bilinguals in cond itions of increased cognitive demand. As suggested by Goldinger et al. (1991), the effects of neighborhood characteristics on word recognition convey information about listenersÂ’ internal lexicon and the relative accessibility of a given word and its neighbors. Therefore, it seems reasonable to gather information from both monolingual a nd bilingual participants that will give insight into their internal lexicons, such as measures of receptive vocabulary size and listening comprehension.
19 Oral receptive vocabulary size is a measure of interes t for several reasons. In the literature on child language acquisition, receptive vocabul ary size has been shown to be a strong predictor of performance on both phonetic discrimi nation and phonological (nonword repetition) tasks. Authors of these studies specu late that a larger vocabulary size requires the child to pay greater attention to fine phon etic detail, resulting in more adultlike category formation (Majerus, Poncelet, Greffe, & Van der Linden, 2006; Walley, 1993). As described by Walley (1993), some believe that childrenÂ’ s lexical processing is more holistic than segmental at the outset. As the vocabulary grows, it is argued, children begin a segmental restructuring of their lexical representations which allows for more phonetically detailed and efficient storage. Moreove r, studies have shown that childrenÂ’s short-term memory performance, as measured by d igit-span and non-word repetition tasks, has a strong positive correlation wit h vocabulary development (Majerus, Poncelet, Greffe, & Van der Linden, 2006). In a related study, although with collegeaged subjects, Lewellen et al. (1993) used three measures to se parate participant groups in their investigation of how differences in subjectsÂ’ lexical familiarity influenced their word recognition and lexical access. They gathered data o n word familiarity, vocabulary level, and language experience. Lexical familiarity was assessed by having participants rate on a 7-point scale the familiarity of 450 words that were selected from WebsterÂ’s Pocket Dictionary and had familiarity ratings from a pr evious study (Nusbaum, Pisoni, & Davis, 1984). Based on their results, the researchers ar gued that IQ measures did not provide insight into the underlying cognitive processes involved in lexical access; rather they found that participants who differed in rated familia rity of the target words also differed in processing efficiency. They assumed that par ticipants with higher scores on
20 the familiarity ratings, vocabulary test, and language e xperience questionnaire had larger lexicons and, therefore, could activate more candidates f or recognition than could individuals with smaller lexicons Based on these issues, the first goal of this study is to compare the size of the easy-hard word effect in the recognition of spoken words by three listener groups: monolingual, earlier-learning non-native, and later-learnin g non-native. As yet, however, the AOI of non-native listeners hav e not been considered in studies that have investigated the easy/ hard word effect or cognitive demand using the pre-load technique. Spoken word recognition by non-native spea kers depends on vocabulary development in the target language, yet I am unaware of any study that has investigated the easy/hard word effect for non-native speaker s that has also measured the participantsÂ’ target language vocabulary level. Imai et al. (2005) defined proficiency as the degree of accentedness of the non-native speakers a s measured by native listeners. They later correlated such factors as number of years o f English-language study with degree of accent. Likewise, Bradlow and Pisoni (1999) perf ormed only correlational analysis of factors such as age of English study onset, number of years of English study, and number of years in an English environment, but neither study directly measured the vocabulary level of their non-native participants. Therefore, I obtained a measure of target-language vocabulary development of the non-native speakers in order to investigate the relation between vocabulary level and wo rd recognition. Theoretically, it was not necessary to control for first language since t he effects of vocabulary size should be present in the pairing of any L1 and L2, and I was not fo cusing on specific phonemes. The effects of cognitive demand and noise on word recognition and word recal l
21 One potential explanation for the difficulty bilinguals e xperience listening to their L2 speech is that speech processing demands greater attenti onal resources even for proficient bilinguals than for monolinguals (Rogers et al., 2006). These differences may not be seen in quiet or undemanding conditions when atte ntional resources are plentiful. Investigation of the effects of bilingualism on the percept ion of speech presented under a range of listening conditions is important because the eff ects of bilingualism on listenersÂ’ perception may combine with the effects of adverse enviro nmental listening factors in ways different from those for monolinguals. Research on synthetic speech intelligibility may offer insights to the experience of bilinguals because synthesized speech, like native speech for second-language learners or non-native speech for native listeners (Imai, Walley, & Flege, 2005), may not match a listenerÂ’s expect ancies for all cues. Thus, similarity effects may be in play during word recognition l eading to similar perceptual effects. Below, a brief discussion of studies of per ception of synthetic speech is provided in order to consider the potential processing parallels fo r native speakers listening to synthetic speech versus non-native persons listening to speec h in their second language. Pisoni and Koen (1981) found that monolingual listenersÂ’ word r ecognition performance on the Modified Rhyme Test (MRT) decreased mor e in noise for synthetic speech than for natural speech, even though performance in quiet was similar for both synthetic and natural speech. Likewise, Koul and Allen (1993) looked at the effects of noise on the intelligibility of synthetic and natural s peech and found that decreasing signal-to-noise ratios had more deleterious effects on synthetic speech, although the patterns of errors were similar for both.
22 Paris, Thomas, Gilson, and Kincaid (2000) found that when li nguistic cues (e.g., prosody, syntax, and semantic cues) were manipulated or eli minated in sentences, immediate recall of both synthetic and natural speech declined. Using the phoneticallybalanced Harvard sentences (sentences that avoid high-pr edictability, too frequent use of one word, and for which phoneme frequency matches that of English), participants heard four different kinds of utterances: normal (with prosodic and contextual cues), no prosody (normal sentences with no prosody), no context (semantically anomalous sentences with prosody), and unstructured (unrelated words wi th no prosody). The semantically analogous sentences were created by re arranging the words in the sentence. Additionally, the sentences with sema ntic context were not highly predictable and, in all speech modes, any within-word or le xical prosody remained. To create the Â“no prosodyÂ” and Â“unstructuredÂ” stimuli in th e natural speech condition, individually recorded words were concatenated into strings. Participants were then required to immediately repeat what they heard. Overall intelligibility and recall were better for natural speech than synthetic speech and for natural sentences with prosodic cues than those without. Interestingly, removing the pr osody from the synthetic speech did not cause a further decrement in immediate recall t han did the synthesis itself. Paris et al. (2000) suggested that the prosodic cues present in synth etic speech systems are not helpful to the listener, so removing them causes no additi onal decrement in intelligibility. The researchers argued that when these cues are not modele d correctly extra burden is placed on working memory that can exceed its capacity. ListenersÂ’ attention, they contend, is drawn towards more superficial acoustic inform ation and is directed away from deeper linguistic analyses. They argued that as inte lligibility decreases, context
23 becomes increasingly important because listeners must de pend on other sources of information for accurate word recognition. Context ma y be used as a Â“compensatory mechanismÂ” that listeners use when intelligibility is degraded or poor, as with synthetic speech, and by extension non-native speech. Thus, prosody is only helpful when the overall intelligibility is relatively good, as in syn thetic speech or like non-native speakers in the case of second language learning. In investigating these issues, Paris, Gilson, Thomas, and Silver (1995) also found that performance on text comprehension tasks was better for natural voices as contrasted with synthetic and for easy than for hard passages. The y argued that stimulus encoding and comprehension processes share a common pool of resour ces. If listening to synthetic speech requires that a greater proportion of cognitive re sources be allocated to analyzing the initial-acoustic structure of the signal, the resea rchers assert, fewer resources are then available for comprehending and processing the semantic con tent. One explanation of these findings is that decoding the a coustic and phonetic characteristics of synthetic speech may require more cognitive effort than decoding natural speech. This may be due to the relatively small number of acoustic cues present in synthetic speech than in natural speech, which is re dundant and contains many cues that may help to specify a particular phoneme. With l ess redundancy in the acoustic signal, the listener has fewer converging sources of evide nce regarding the identity of the phoneme or word in question and thus may have a more diff icult time differentiating the target word from phonologically similar neighbors. It is hypothesized that the reduction in redundancy of acoustic cues in synthesized speech leads to more effortful processing of the speech, which may go unnoticed in conditions of qu iet and when task demand is
24 low (Logan et al., 1989). When noise masks portions of th e acoustic cues or processing demand is increased by other factors, however, the effec ts of the greater demand placed on the system by the synthetic speech are seen. Pisoni, Nusbaum, and Greene (1985) hypothesized that perceptio n of synthetic speech requires more cognitive effort than perception of natural speech for both words and non-words. This means not only that lexical retriev al that is more difficult, but also that the extra processing effort appears to be related t o the process of extracting the acoustic-phonetic information from the signal. They re asoned that synthetic speech requires more short-term memory capacity and should int erfere with other cognitive processes because it imposes greater capacity to process i t. The consequences might mean that listeners who are trying to encode an impover ished signal, such as that found in synthetic speech, speech presented in noise, or non-nat ive speech, could perform worse on simultaneous or subsequent cognitive tasks. Likewise, Ral ston, Pisoni, Lively, Greene, and Mullennix (1991) found that on-line processing, as a ssessed by word monitoring and sentence-by-sentence listening was worse in a ll tasks with synthetic speech than for natural speech. In line with the reaso ning offered by Pisoni et al. (1985), these authors suggested that poorer comprehension is due in par t to the greater encoding demands required for the perception of synthetic speech. In addition to the difficulties encountered in the perce ption of synthetic speech relative to natural speech, there seem to be even mor e deleterious effects found when the listenerÂ’s task requires increased capacity demands. For example, Luce, Feustel, and Pisoni (1983) compared recall for synthetic speech and natura l speech using a memory pre-load paradigm: subjects were visually presented with ze ro, three, or six digits and
25 then heard a list of words in either synthetic or nat ural speech. Subjects were instructed to recall the digits in the exact order and then recal l as many of the words as they could. Results showed that fewer listeners were able to reca ll the digits accurately when they were followed by synthetic speech than by natural speech; participants also showed poorer free recall and poorer ordered recall for digits a nd words in the synthetic speech condition. Luce et al. (1983) argued that these difference s in recall performance between synthetic and natural speech occur because synthetic spee ch has fewer redundant acoustic cues than natural speech, leading to impoverished representa tions in short-term memory. These impoverished representations mean that short-term memory has to work harder to maintain the signals in memory. Additionally, rehears al of the digits in short-term memory may be interrupted by the greater encoding effort necessary for synthetic speech than for natural speech. The researchers argued that deg raded input may require spare capacity in short-term memory, thus supporting the proposal that decrements in recall for degraded stimuli are the result of both encoding difficu lties and short-term memory limitations. If the difficulty encountered in processing synthetic s peech results from differences between listenersÂ’ expectations of acoust ic cues and the acoustic cues encoded in the signal, then it seems reasonable to assum e that the same type of difficulty may arise for bilingual listeners, who may not have acqui red all of the cues used by native listeners or who may weight these cues differe ntly from native listeners (Flege, 1995; Imai, Walley, & Flege, 2005). That is, if the perceptio n deficits encountered by non-native listeners are due to encoding difficulties a t early processing stages, then there should be measurable increases in the demands placed on th e resources available in short-
26 term memory for the non-native listeners, relative to native listeners. Another factor that may contribute to increased processing demand for bilinguals i s the need to suppress the non-active language during processing of the active language in order to decrease interference (Grosjean, 1997). Taken together, the two f actors of mismatch between phonetic expectations and input and the need to suppress the non -active language may result in substantially greater processing demands for bil ingual listeners. Another goal of the study is to directly investigate the effects of cognitive load (noise and memory load) on speech processing by bilinguals to help confirm or disconfi rm these hypotheses and, if they are found to be true, allow for some estimate of the magnitude and conditions of the increase in processing demand for bilinguals relative to mono linguals. Semantic Characteristics In addition to considering the phonological neighborhood characteristics, semantic network characteristics of words may also af fect the access one has to words in recognition tasks. No previous studies of the effects of pho nological word characteristics on word recognition have taken into account the potential effects of these characteristics. Whereas phonological neighborhood characteristics prov ide information about the relationships among words based on their sound patterning, se mantic network characteristics provide information about the relationshi ps among words based on their meanings (Nelson, McEvoy, & Schreiber, 1998). Three indic es were considered: cue set size, connectivity, and concreteness. The first index considered, cue set size, refers to the number of different cued associates for a particular wor d. It is calculated by presenting individuals with a word and then counting the number of dif ferent responses or targets given by two or more participants in a given sample (Nel son, McEvoy, & Schreiber,
27 1998). Cue set size provides a relative index of the set size of a target word by giving a reliable measure of how many strong associates it has. In the words used for this study, this index did not differ between easy and hard word lists. Connectivity is an index of the average associate-to-associate connectivity among the associates of the cue and of the target. In other words, it indicates the density and leve l of semantic association within cues and targets. This measure differed between the easy and hard words in this study, with the hard words being less connected than the easy words. These findings appear to be contrary to the Neighborhood Activation Model (Luce & Pisoni, 1998). The phonologically hard words are less semantically connected and thus, more easily accessed due to less competition relative to the easy wo rds. Thus, an effect in which the hard words are more difficult to recognize cannot be attr ibutable to the difference in semantic connectivity. Alternatively, perhaps having a hig h degree of connectivity does not have the same consequences for word recognition as does having a large number of phonologically similar neighbors. Having lots of neighbor s may make recognition for a specific word relatively difficult compared to a word wit h few neighbors because the neighbors compete with each other based on the way the y sound to a listener. On the other hand, having a high degree of semantic connectivity coul d work to support the activation of a particular word because the semantical ly associated words work to prime the target word for recognition. The final semantic in dex considered, concreteness, which is a measure of the ease with which a word can b e imagined as measured on a scale from 1-7, did differ between easy and hard word lis ts. This effect also seems to go against the predictions of the Neighborhood Activation Mode l (Luce & Pisoni, 1998) because the phonologically hard words are more concrete and presumably more easily
28 accessed than the easy words. Therefore, an effect i n which the hard words are more difficult to recognize relative to the easy words could no t be easily attributable to the difference in concreteness. It should be noted here that a limitation of this study is that not all the words had data points for each of the sema ntic features, especially the hard words. For example, only 62 of the easy words had values for connectivity, while the hard words only had 34. Therefore, it is difficult to specu late on what any findings regarding the effects of semantic characteristics on word recognition accuracy might mean, as they should be interpreted with caution. This was purely an exploratory analysis and provides direction for future study. Based on these issues, the first goal of the present study is to compare the size of the easy hard word effect in recognition of spoken words by three groups of listeners: monolingual English speakers, high-proficiency bilinguals, and l ow-proficiency bilinguals. The second goal of this study is to determine the effects of increasing cognitive demand (in this case increasing number of digits to be re called) on speech recognition and working memory during speech perception tasks to determine whether the effect was greater for non-native than for native listeners. A s noted earlier, in their study on recall of synthetic versus natural words, Luce et al. (1983) found th at the pre-load memory technique placed increased demands on the encoding and/or re hearsal processes in shortterm memory when the participants were simultaneously e ngaged in another task that also required short-term memory capacity. This decrement in performance was worse for synthetic speech than for natural speech. Further, the stimuli in the present study consisted of the easy and hard words used by Bradlow and Pi soni (1999). In their study,
29 easy words had higher intelligibility than hard words, and this was especially true for the non-native listeners. They argued that the ability to m ake the fine acoustic-phonetic distinctions required to discriminate the hard words is a skill that develops with knowledge of the sound-based system of the language. The findings of this study should have important implicati ons for the teaching and assessment of non-native speakers. Specifically, working t o increase the vocabulary level of second-language learners may indirectly help improve the ir recognition and subsequent comprehension of spoken material. As their L2 vo cabulary grows, bilingualsÂ’ ability to make fine phonetic distinctions also appears to get better. Typically, classrooms are noisy and distraction-filled environments and the task of comprehending a lecture is made more difficult when the subject matte r is advanced. That holds true when all individuals are native speakers, but may become exa ggerated when non-native listeners are involved. Clearly, we need more studies t o elucidate the factors that have the greatest impact on intelligibility, especially for non-native talkers and listeners. The specific research objectives were (1) to compare the effect of phonological neighborhood characteristics on word recognition betwe en native listeners and nonnative listeners, (2) to examine the effect of non-nat ive listenersÂ’ age of immersion and vocabulary level on recognition of words and recall of digits in noise and in quiet, (3) to compare the effects of increasing number of digits to r ecall on the recognition of easy and hard words for native and non-native listeners in quie t and noise, and (4) to explore the effects of semantic characteristics of words on thei r recognition. It was predicted that the easy/hard word effect on rec ognition would be greater for non-natives, especially under conditions of increas ed digits to recall. More
30 specifically, since it is argued that the ability to discr iminate and subsequently encode hard words is a skill that develops with knowledge of the language (i.e., vocabulary level), I predicted that the early learners (EL) liste ners in this study would have better recognition of the hard words compared to the later learn ers (LL) listeners. Finally, the effects of vocabulary level and noise were predicted to have an effect on the recall of digits such that the LL group would experience more delete rious effects of noise on their ability to recall the digits than would the EL group. This would result from the additional short term memory capacity needed to recollect the digi ts, which would subsequently leave less capacity for the encoding of words in the re cognition task. There are no predictions regarding the effects of the semantic chara cteristics of the words on word recognition since I did not have data for all the words nor did I control for the factors. Instead, the analyses were exploratory in nature with no a-priori expectations. It was hypothesized that vocabulary level would predict word recognition, which supports the premise that increasing vocabulary level may result in greater attention to fine phonetic detail. The more words one has in his or her vocabulary, the more necessary it becomes to be able pay attention to fine phonetic detail in order to make distinctions among them. It was further hypothesized tha t positive correlations between the vocabulary level of the non-native listeners an d their word recognition scores would suggest that lexical development precedes and influences phon ological knowledge of the L2 or a bi-directional or interactionist theory in whic h lower-level, phonological knowledge and higher level, lexical knowledge influence one an other.
31 Method Design The experimental design is a mixed model with three dig it recall conditions (0, 3 or 6 digits), two types of words to be recognized (easy ver sus hard), and noise (quiet versus noise) as within subjects variables. Proficien cy (monolingual, earlier-learner nonnative, later-learner non-native) varied between subjec ts. The dependent variables were number of words correctly recognized and the number of di gits correctly recalled. Participants Two groups of listeners participated in this experiment. Thi rty-six monolingual English speakers (MO) born in the United States comprise d the first group. According to self-report, they did not have spoken or written fluenc y with any language besides English (see Appendix A). The listeners in the other group consisted of sixty non-native speakers whose second language is English. This group was divided into 36 earlier and 24 later age-of-immersion (AOI) categories (see Tables 1 a nd 2) based on the participantsÂ’ age of immersion in an English-speaking envir onment and other information gathered via the language-background questionnaire (see Appendix B). Theoretically, it was not necessary to control for first language (L1) b ecause the effects of vocabulary size, noise, and number of digits to be recalled should be present in the pairings of any L1 with any L2. Listeners were between the ages of 18 and 50 years. Fifty years was chosen as the upper limit in age because, beyond this poin t, age-related hearing loss and age-related decreases in cognitive processing abilities are more likely to occur. The
32 listeners included both males and females (M=15, F=81). Th ey were recruited from the University of South Florida Departments of Psychology an d Communication Sciences and Disorders and the English Language Institute. Partici pants were compensated with extra credit points or were paid for their participation either with cash or gift certificates. Listeners were screened to exclude those with a history of speech, language, or hearing disorders. Potential participants were also required to pa ss a pure tone hearing screening prior to their participation. Native listeners did not hav e a strong regional accent as judged by the investigator, a native English speaker. The earlier learners (EL) were those who were immer sed in an English-speaking environment at age 10 or earlier, rated themselves as rel atively balanced in proficiency in their L1 and L2 in a variety of contexts, and, according to a screening by the experimenter had at most a mild foreign accent. The l ater learners (LL) were those who were immersed in an English-speaking environment at age 14 or later, rated themselves as dominant in their L1 in a variety of contexts, and h ad a moderate to strong degree of foreign accent in the experimenterÂ’s judgment. The c ut-off ages for the EL and LL groups, although relatively arbitrary, were selected becau se they provide a good separation between the groups in terms of age of immersi on in the L2. As shown in Table 3, the EL and LL groups differed significantly in the r eported percent of time spent speaking English at home and the reported amount of time sp ent speaking their L1 with others. They also differed in how they rated their pro ficiency in their L1 and L2. The EL bilinguals gave themselves significantly higher ratings than the LL bilinguals in the areas of comprehension, fluency, vocabulary, and pronunciation in English. On the other hand, the LL bilinguals gave themselves significantly higher ratin gs than the EL bilinguals in
33 the areas of comprehension, fluency, vocabulary, pronun ciation, and grammar in their L1. The groups rated themselves similarly in grammar in Engl ish. Table 1. Demographic information of individual earlier-lear ner bilingual participants. Age AOI L1 Country of Origin LOR Age AOI L1 Country of Origin LOR 20 2 German Germany 0.2 21 7 Spanish US 21 18 5 Tagalog Philippines 13.5 19 5 Spanish US 19 24 10 Arabic Egypt 14.2 19 5 Gujarati US 19 20 5 Vietnamese US 20 18 6 Creole US 18 22 7 Spanish Puerto Rico 17 20 4 Spanish Dominican Republic 16 20 3 French Creole Canada 16.5 18 10 Spanish Cuba 7.5 18 4 Vietnamese US 18 19 5 Spanish US 19 18 9 Hindi India 9.5 18 5 Urdu Pakistan 13 20 5 Spanish Mexico 19 24 1 Spanish US 24 18 2 Spanish Cuba 15 19 7 Spanish US 10.8 30 1 Greek US 30 20 4 Spanish US 19 19 8 Urdu Pakistan 11.5 19 6 Creole US 5 20 5 Spanish US 20 19 6 Tagalog Philippines 3 20 5 Spanish US 20 20 8 Spanish US 19 19 6 Spanish US 19 23 4 Creole US 23 21 2 Creole US 21 19 4 Spanish US 19 20 4 Spanish US 20 22 9 Spanish Puerto Rico 10 20 5 Serbian Serbia 12.3 27 5 Spanish US 7.3
34 Table 2. Demographic information of individual later-learner bilingual participants. Age AOI L1 Country of Origin LOR 25 21 Japanese Japan 4.3 25 20 Japanese Japan 5 19 18 Spanish Colombia 1 20 19 Portuguese Angola 1 18 17 Spanish Colombia 0.75 23 15 Creole Haiti 9.25 18 14 Russian Russia 5.5 22 14 Serbian Bosnia 10.5 20 15 Spanish Peru 4.75 38 28 Spanish Colombia 10 29 20 Spanish Nicaragua 10 22 17 Creole Haiti 4.25 21 14 Spanish Colombia 9 34 19 Spanish Puerto Rico 14.3 20 19 Polish Poland 1.25 22 14 Spanish Peru 12 23 16 Spanish Colombia 7 22 14 Spanish Colombia 8.6 23 16 Albanian Albania 7.5 49 14 Bulgarian Bulgaria 13 23 14 Spanish Cuba 9.3 21 14 Spanish Colombia 7 27 18 Serbo-Croatian Serbia 9.6 22 14 Japanese Japan 8
35 Table 3. Demographic information for the EL and LL bilingual groups. Means are presented with standard deviations and ranges in parenthesis Ratings are based on a scale from 1-5 from 1 (not proficient) to 5 (very proficient). Earlier Learners Later Learners Chronological Age 20.28 (2.59; 18-30) 24.42 (7.04; 18-49) Age of Immersion ** 5.25 (2.29; 0-10) 16.83 (3.36; 14-28) Length of Residense ** 15.81 (6.21; 0-24) 7.2 (3.84; 1-14.3) % of time spent speaking English at home 55.36 (30.88; 0-100) 36.63 (38.42; 0-100) % of time spent speaking English at work 82.60 (33.44; 0-100) 72.95 (36.37; 0-100) % of time spent speaking English in all other situations 70.15 (31.68; 0-50) 65 (30.32; 10-100) % of time spent with speakers of their L1 40.19 (27.43; 0-90) 59.5 (32.29; 0-100) Comprehension in English ** 4.58 (.65; 3-5) 3.88 (.68; 3-5) Comprehension in L1 ** 4.35 (.61; 3-5) 4.83 (.48; 3-5) Fluency in English ** 4.61 (.64; 3-5) 3.71 (.64; 3-5) Fluency in L1 ** 4 (.93; 2-5) 4.88 (.45; 3-5) Vocabulary in English ** 4.28 (.74; 2-5) 3.63 (.65; 3-5) Vocabulary in L1 ** 3.75 (.87; 2-5) 4.58 (.72; 3-5) Pronunciation in English ** 4.44 (.81; 2-5) 3.25 (.85; 1-4) Pronunciation in L1 ** 4.06 (.95; 2-5) 4.92 (.28; 4-5) Grammar in English 4.25 (.87; 2-5) 3.96 (.75; 3-5) Grammar in L1 ** 3.47 (1.18; 1-5) 4.54 (.78; 3-5) *= significant difference between groups at p <.05 *= significant difference between groups at p <.005
36 Stimuli Speakers Two monolingual speakers of English recorded lists of wo rds. The speakers were 24and 26-year-old women who were recruited fr om the University of South Florida and were judged to have no strong regional d ialect by the experimenter. Word lists. In the main production task, the speakers read aloud 144 words from a list provided by the investigator. All the words came from the stimuli used by Bradlow and Pisoni (1999) as described previously with a few modificati ons. Bradlow and PisoniÂ’s (1999) easy and hard word sets consist of 75 easy and 75 hard words each, but, the present study used only 72 words (see Appendix C and D) fro m each list for two reasons. First, a multiple of 12 words was needed to fit the design of the study. Second, the word lists used by Bradlow and Pisoni (1999) overlapped on al l three lexical characteristics. By omitting three words from the eas y list (Â“foolÂ”, Â“washÂ”, and Â“wasÂ”) and three from the hard list (Â“mainÂ”, Â“wrongÂ” and Â“whiteÂ”) there was a more defined separation between the lists. For example, Â“foolÂ” a nd "wash" have quite low target frequencies. Easy words should have lower neighborhood frequency, but Â“washÂ” is about one standard deviation above the mean for the easy word list. Easy words should also have lower neighborhood density, yet Â“foolÂ” was about one standard deviation above the mean for the easy words on this measure. Â“WasÂ” is an outlier for frequency, even though in the expected direction, and also has an extremely lo w neighborhood density. For the hard words, Â“mainÂ”, Â“wrongÂ” and Â“whiteÂ” are all relativel y high in target frequency and are the only words to overlap in target frequency with th e frequency of the "easy" words. Additionally, Â“wrong" is actually on both lists which wa s an admitted mistake on Bradlow and Pisoni's (1999) part. It also better fits the c riteria for the easy words. For "white" the neighborhood density is actually lower than the mean for the easy words
37 (hard words should have higher density); for "main" the nei ghborhood frequency is substantially below average for the hard words. In summ ary, the easy words are those that occur frequently in the language and have few phonetica lly similar neighbors that are mostly low-frequency. The hard words, on the other hand, occur less frequently in the language and have many similar neighbors that are mostly hi gh in frequency. In addition to considering the phonological neighborhood characteristics, semantic network characteristics of the easy and hard word lists were examined. It should be noted that data on these variables was not ava ilable for many of the words. The first index, cue set size, refers to the number of different cued associates for a particular word (Nelson, McEvoy, & Schreiber, 1998). The easy and hard word lists did not differ significantly on this characteristic (see A ppendix D). Connectivity is an index of the average associate-to-associate connectivity amon g the associates of the cue and of the target (Nelson, McEvoy, & Schreiber, 1998) (see Appendi x D). This measure differed between the easy and hard words, with the hard wo rds being less connected than the easy words. The final semantic index considered, co ncreteness, which is a measure of the ease with which a word can be imagined as measured o n a scale from 1-7, did differ between easy and hard word lists (see see Appendix D ). Recording procedures. The stimuli were recorded in a sound-attenuated booth in the Department of Communication Sciences and Disorders at the University of South Florida. The speakers were given the stimulus words to r ead over to allow for familiarization and to ensure fluent speech during recordi ng. These words were provided on a sheet of paper, and the speakers were instructed to re ad the words at a normal conversational pace and to leave about two seconds between words. The speakers were
38 instructed to repeat a word when they made any type of mista ke, such as a hesitation, mispronunciation, or dysfluency. To avoid ambiguity, th e experimenter demonstrated an acceptable pace by reading a list of practice, non-stimulus words to the speaker. Once the speakers demonstrated understanding of the task and had fa miliarized themselves with the stimuli, recording began. The experimenter exi ted the booth and returned to the recording equipment to monitor the recording levels while t he speakers read the words she instructed the speakers to repeat any target items if ne eded. During recording, the speakers first read a practice list. The practice list contained 10 items and familiarized the speakers with the task and allowed the experimenter to monitor and adjust the recording level (see Appendix E). Finally, the speakers were recorded reading the main word list. Distracter (non-target) words were added at t he beginning of the list, the end of the list, and at the end of each column in order to avo id prosodic differences in pronunciation due to list beginning and end effects (see Append ix F). After reading all 150 words (6 distracter words and 144 stimuli words) once, the participants had completed the speaking task. This procedure took approximately 30 minutes to complete. Ten practice words were recorded by an additional female, native English speaker in the same manner as described above. These words wer e used in the practice tasks for the listeners prior to the main experiment. The speakers were recorded digitally at a sampling rate of 44.1 kHz, with 16-bit amplitude resolution, using a digital audio workstation (Ro land VS890HD) and a highquality microphone (Audio-Techinica, AT4033). The words produce d by the speakers were saved to the workstation and transferred digitally to computer for subsequent digital editing. Each target word was edited from the list usi ng acoustic editing software, saved
39 to a separate file, and then peak normalized to a pre-speci fied RMS level (approximately 20 dB less than the system maximum amplitude). Noise mixing. Pilot testing was conducted in order to determine the noise level at which a relatively equal challenge would be present for eac h of the listening groups in the main experiment. This was done by looking at the percent correct responses for each listener group in the pilot study at various signal-to-noi se ratios (SNR) and estimating the SNR at which there was a 25% reduction in performance com pared to the quiet condition. It was predicted that the LL group would need le ss noise to achieve the same decrement in performance relative to the EL and monoli ngual groups. As with the main experiment, two groups of listeners par ticipated in the pilot study: one group of monolingual English speakers (n = 12) and t wo groups of bilinguals (n = 12) who differed according to age of immersion in an E nglish-speaking environment. The stimuli included the words used in the mai n experiment, divided into six lists of 24 words. The words were spoken by a female, monolingual speaker of English. The noise used consisted of multi-talker babble from the Speech Perception in Noise (SPIN) sentences (Bilger, Neutzel, Rabinowitz, & Rzeczkowski, 1984). To avoid any potential learning effects that might result from us ing the same segment of noise for all the words, a two-minute segment was selected from t he SPIN sentences babble which had relatively stable levels of noise throughout. The n oise was then mixed with each target word by using a computer program that first randomly chose a section of the twominute babble that was equal to the duration of the target w ord plus 1000ms (500 ms lead and 500 ms lag). The program then scaled the noise to achiev e the desired SNR, based
40 on peak amplitude of the two items (word and noise), mixed the noise and word, and then rescaled the combined file to the original RMS amplitude. During pilot testing, each listener heard one list of w ords at a time presented either in quiet or mixed with noise at several SNR tha t decreased from 18dB to 2dB in four dB steps (e.g., 18dB, 14dB, 10dB, 6dB, and 2dB). The SNR needed t o obtain 75% of performance in quiet was calculated for each group. T he SNRs chosen for the groups to be used in the main experiment were as follows: monol inguals had +6 dB SNR, EL had +5 dB SNR, and the LL group had +13 dB SNR. Materials Receptive vocabulary size was measured using the Peabody Picture Vocabulary Test Â–Third Edition (PPVT-R; Dunn & Dunn, 1997). The PPVT is a measure of receptive vocabulary of English as well as a screening test of verbal ability for individuals aged 2-90+ years. For each target word, spoken by the test administrator, the participant must select from among four black and white draw ings. It took approximately ten to 15 minutes to administer. The PPVT correlates we ll with other measures of receptive language, including an average correlation of .69 wi th the Oral and Written Language Scale Listening Comprehension subtest. Additional ly, it has even higher correlations with some measures of verbal ability such the WISC-III VIQ (.91) and the KBIT (.81) (Dunn & Dunn, 1997). In addition, receptive language was assessed for all part icipants using the Listening Comprehension subtest of the Oral and Written L anguage Scale (OWLS; Carrow-Woolfolk, 1995). For this test, the examiner read a stimulus sentence to the participant who then indicated the correct picture from among four choices that
41 corresponded to the stimulus. It is designed for individuals aged three through 21 years. Items probe lexical knowledge, understanding of syntactic c onstructions such as embedded sentences and subordination, knowledge of supra-li nguistic structures such as figurative language, and other higher-order thinking skills. The OWLS correlates well with other measures of language including the CELF-R Oral Comprehension subtest (.91). The OWLS Listening Comprehension subtest also corr elates well with cognitive measures that assess both verbal (WISC-III Verbal IQ .77; K-BIT Vocabulary subtest, .76) and non-verbal ability (WISC-III Performance IQ, .70; K-BIT Matrices subtest, .59). Its correlations with global measures of cognitive abili ty are .76 for the WISC-III Full Scale IQ and .72 for the K-BIT Composite (Carrow-Woolfol k, 1995). These language tests were administered to examine the relationship betwee n vocabulary and word recognition skills and more general linguistic competence i n bilinguals and monolinguals. A detailed language background questionnaire was provided to the no n-native participants for collection of data on age of acquisitio n, language dominance, language usage, and history of speech and hearing impairment. A si milar but less detailed language background questionnaire was provided to the native speaker s to ensure that they were indeed monolingual and did not have a history of speech or hearing impairment (see Appendices A and B). Procedure Participants in the listening task were tested individuall y or in groups of up to four in a speech perception lab in the Department of Communic ation Sciences and Disorders. Upon arrival in the lab, they were greeted, told about t he nature of the study, and given informed consent materials. Participants completed all consent forms, language
42 background questionnaires, and a basic hearing screening prior to the main experimental task. After all consent forms, hearing screenings, and language measures were completed, participants were directed to have a seat in f ront of one of the computers. Each session began with one practice list of ten words ( with no memory pre-load) consisting of non-target words (i.e., not from the easy and hard lists) spoken by the same speaker as for the main test stimuli. For the main experimental task, participants heard words spoken by one of the native-English speaking females, in sets of 24, 12 easy a nd 12 hard in each set. Following the procedure used in Luce et al. (1983), prior to the presentation of each word sub-list, the participants saw either zero, three, or six digits displayed visually, one at a time, on a computer screen positioned directly in front of them at a distance that allowed for easy viewing. The participants were instructed to rem ember the digits (if any) in the same order as they were presented. Each digit, sample d without replacement from the digits one through nine, remained on the screen for two seconds. The interval between the presentations of each digit was one second. Nex t, the words in the sub-list were presented and participants typed in what they recognized aft er each word. The six sets of 24 words were counterbalanced across listening conditions ( 3 digit conditions and 2 noise conditions) using a Latin Square design with six condition s. For example, a listener might have the quiet condition first and start with thr ee digits, so that their order of presentation would be: three digits quiet, three digits noise, zero digits quiet, zero digits noise, six digits quiet, and six digits noise. All t okens were presented over headphones for each listener at approximately 65 dB SPL. Each partic ipant heard all the words with none repeated. The subsets of 24 words and noise condition s were counterbalanced
43 across listeners within each group so that each subset wa s heard under each digit condition and each noise condition an equal number of times within each group, according to the Latin square design. At the end of the s ub-list presentation, the participants typed in the digits they remembered seeing. The procedure continued in this manner until the six 24-item sub-lists were completed. The words were automatically scored by a customized comput er program The misspelling of a word did not necessitate its being counted as incorrect. Rather, after all words were scored automatically by the computer program, the experimenter went through the responses and counted correct any word that was obviously misspelled or that was an obvious typo based on the position of the letters on the keyboard. Data for the receptive vocabulary measure were collec ted after the main experimental task. For the PPVT, an answer book was us ed that had four pictures per page, with one that corresponded to the target word. The inv estigator said a word and the participant was instructed to say the number or point to the picture that corresponded to the word. Next, the OWLS was administered. This test a lso used an answer book that had four pictures per page, with one corresponding to the c orrect answer. For this test, however, a short sentence or paragraph was read. Parti cipants were instructed to point to the picture or say the number of the picture that corres ponded to the situation described. Following the language testing, the 144 words presented in the main experimental task were presented again in random order ove r headphones along with the written counterparts in the same order. In this post-te st, listeners rated each word for familiarity on a scale from 1 (not known) to 9 (very familiar). After rating all 144 words for familiarity, the participants were compensated for their participation and dismissed.
44 Results The specific research objectives for this study were (1) to compare the effect of phonological neighborhood characteristics on word recognit ion between native-speaking listeners and non-native listeners, (2) to examine the ef fect of non-native listenersÂ’ age of immersion and vocabulary level on recognition of words and recall of digits in noise and in quiet, (3) to compare the effects of increasing number of digits to recall on the recognition of easy and hard words for native and non-nativ e listeners in both quiet and noise, and (4) to explore the effects of semantic chara cteristics of words on their recognition. It was predicted that the easy/hard word effect on reco gnition would be greater for non-natives, especially in noise and with increased n umber of digits to recall. It was also predicted that vocabulary level would be the best pred ictor of hard word recognition compared to age-of-immersion and length-of-residence and that positive correlations would exist between the receptive vocabulary level of th e non-native listeners and their word recognition scores. The effects of age of immersi on and noise were also predicted to have an effect on word recognition and the recall of d igits such that the later-learning (LL) group would experience more deleterious effects of no ise on their ability to recognize the words and recall the digits than would the early-learning (EL) group. The possible effects of the semantic characteristics of t he words on word recognition was explored.
45 In order to answer some of these questions, a four-way a nalysis of variance was conducted to analyze the percent-correct scores for word re cognition task. The dependent variables were percent of words recognized correctly. Li stener group (three levels: native (MO), EL non-native, and LL non-native) was the betw een-subjects variable; digit recall condition (three levels: 0, 3 or 6 digits), word type (two levels: easy and hard), and noise condition (quiet and noise) were the within-subjects va riables. Data were converted to Rationalized Arcsine Units (RAU) (Studebaker, 1985). Doing l inear tests on proportional data can be difficult since the distributi ons of these values are not strictly Gaussian, especially when the proportions are near 0 or 1. The Rationalized Arcsine Transform linearizes the proportions and converts them to rational arcsine units so that linear tests can be performed on the RAU values. Arcsi ne transformations have been used in research to transform proportions to make them more appropriate for statistical analysis, but the arcsines did not always show a clear relationship to the original proportions, making them difficult to interpret. The RAU Transform, on the other hand, produces values that are numerically close to the original per centage values over most of the range while retaining all of the desirable statistical properties of the arcsine transform. Word Recognition There were significant main effects of group, noise cond ition, and word type but not digit condition, see Table 4 and 5 for means. The M O group correctly recognized more words than did the EL group and the EL group correctly r ecognized more words than did the LL group ( F (2, 93) =32.603, p <.000, h p 2 = .412). Further, all groups performed better in quiet than in noise, ( F (1, 93) = 888.16, p < .000, h p 2 = .905). Finally,
46 all groups had better word recognition for easy words than for hard words ( F (1,93) = 299.23, p < .000, h p 2 = .763). Table 4. Means of Word Recognition for Easy and Hard Words by Listener Group. Table 5. Means of Digit Recall by Listener Group. The interactions addressed the research questions concer ning the effect of age of immersion of the non-native listeners on recognition of words in quiet and noise and the effect of phonological neighborhood characteristics o n word recognition for listeners in quiet and noise. The analysis showed differences among groups as a function of the noise condition ( F = (2, 93) = 48.73, p < .000, h p 2 = .512), see Figures 1 and 2. The post hoc Bonferroni analysis revealed that the MO did not dif fer from the EL group. However, they both recognized more words than did LL and these diff erences were greater in quiet than noise. There was also an interaction of word type by group ( F (2, 93) = 31.37, p < .000, h p 2 = .403) such that the LL recognized fewer words than the MO and the EL and Monolingual Early Learners Later-Learners Easy Qt Easy Ns Hard Qt Hard Ns Easy Qt Easy Ns Hard Qt Hard Ns Easy Qt Easy Ns Hard Qt Hard Ns Means 97.4 74.5 92.7 62.0 94.9 73.1 88.7 58.1 84.3 73.1 55.0 45.5 SEM 2.65 8.95 5.48 9.32 4.58 8.96 7.97 13.29 11.01 13.10 23.25 21.82 SD .44 1.49 .91 1.55 .76 1.48 1.33 2.21 2.25 2.67 4.75 4.45 Monolingual Early Learners Later-Learners 3 Qt 3 Ns 6 Qt 6 Ns 3 Qt 3 Ns 6 Qt 6 Ns 3 Qt 3 Ns 6 Qt 6 Ns Means 2.81 2.78 4.81 4.61 2.81 2.72 4.97 4.61 2.78 2.46 4.5 4.69 SEM .10 .11 .25 .29 .13 .12 .21 .30 .08 .23 .40 .34 SD .62 .64 1.47 1.74 .76 .70 1.25 1.77 .41 1.1 1.7 1.66
47 this difference was greater for the hard words than for the easy words. Figure 1. Percent correct word recognition for easy and hard words for all listener groups in quiet. (Error bars represent one standard error of th e mean). Figure 2. Percent correct word recognition for easy and hard words for all listener groups in noise. (Error bars represent one standard error of t he mean).
48 Another objective of the present study was to determine if there was an easy-hard word effect that differed by group and whether this effect was moderated by the addition of noise. It was predicted that there would be an interac tion of age of immersion with word difficulty such that the LL group would show an even gr eater easy/hard word effect than EL group, especially in conditions of noise. The hy pothesis was confirmed by a significant three-way interaction of noise, word type and group ( F (2, 93) = 3.32, p < .040, h p 2 = .07). An examination of the paired contrasts showed tha t in both quiet and noise, all three listener groups had better word recognitio n for easy words than for hard words. However, the difference between the easy and h ard word recognition accuracy was most pronounced for the LL group. In other words, the LL group showed a greater easy-hard word effect compared to the MO or EL groups. In terestingly, the difference between easy and hard word recognition scores among the gr oups lessened in the noise condition (see Figures 1 and 2). To further explore the question regarding the effects of phonological neighborhood on word recognition, analyses were conducted t o determine the frequency with which participants chose a neighbor versus a non-ne ighbor when they incorrectly identified the target word. The motivation for these analyses was the need to gain some insight into the phonological neighborhoods of non-nati ve listeners compared to native listeners. The Neighborhood Activation Model (Luce & P isoni, 1998) predicts that lowfrequency words with many and more frequent neighbors will be more difficult to recognize because of competition from the neighbors. This supposes that, when in error, a listener is likely to choose a neighbor, especially for the hard words. I wanted to know whether this was as true for non-native listeners as it is for native listeners, and to see to
49 what extent vocabulary level moderates this process. T wo sets of analyses were conducted: one compared the groups in the rates at which part icipants chose neighbors, and the other compared the word type with regard to the rates at which a neighbor was chosen. For the first, a one-way analysis of varianc e, with listener group (three levels: MO, EL, and LL) the between-subjects variable, was perf ormed on the percent of neighbors chosen when a target was not accurately recogni zed. Â“NeighborÂ” was operationalized as a word that differed from the target by one phoneme, such as Â“ratÂ” for Â“cat.Â” Non-word neighbors, such as Â“datÂ” for Â“cat,Â” w ere not counted as neighbors. There was a significant difference among the groups in t he percentage of neighbors chosen for the easy words ( F (2, 93) = 6.693, p < .002) and the hard words ( F (2.93) = 12.30, p < .000). Tukey Honestly Significant Difference (HSD) post hoc analysis revealed that, for the easy words, the MO and the EL gr oups did not differ from each other, but they both chose more neighbors than non-nei ghbors relative to the LL group, see Table 6 for means. For the hard words, the MO chose neighbors significantly more than did the EL group, who chose more neighbors than did the LL group. Table 6. Mean percentage of times groups chose a neighbor w hen in error during word recognition task. A paired samples t-test was conducted to determine whethe r a difference existed in the rate at which a neighbor versus a non-neighbor wa s chosen between the easy and hard words for each of the groups overall. It was found t hat participants were more likely Monolingual Early Learners Later-Learners Easy Hard Easy Hard Easy Hard Means 57.89 84.82 57.02 78.91 44.19 71.48 SEM 2.62 1.52 2.40 1.80 3.35 2.23 SD 15.73 9.13 14.40 10.78 16.39 10.90
50 to choose a neighbor in the hard condition than the ea sy condition. This makes sense considering that there are fewer neighbors from which to choose for the easy words, and the target tends to be more frequent than those neighbors. In contrast, the hard-word neighborhoods have many more neighbors from which to choo se, and those neighbors tend to be more frequent. Finally, an analysis was con ducted to determine whether the difference in likelihood to choose a neighbor between th e easy and hard words varied among groups. This was done by calculating the difference i n the percentage of times a neighbor was chosen when in error between the easy a nd hard words and then comparing this difference among the groups using a one-way analysis of variance. This difference was not significant. Anecdotally, the LL group was obse rved to choose neighbors that are not English words (and possibly non-words in their L1 also), something that the MO and EL groups did not do. Vocabulary, Language Scores, and Familiarity Ratings It is argued that the ability to discriminate and subsequentl y encode hard words is a skill that develops with knowledge of the language (i.e., vocabulary level), so, it was predicted that the EL listeners in this study would have b etter recognition of the words than the LL listeners. It was further hypothesized that vocabulary level would be the best predictor of hard word recognition. To investigate the eff ect of vocabulary level on recognition of words, a set of one-way analyses of var iance was conducted comparing the groupsÂ’ vocabulary scores, language scores, and familiarit y ratings. All three groups differed significantly from each other in their vocab ulary scores, with the MO group scoring the highest, followed by the EL bilinguals, and then the LL bilinguals ( F (2, 93) =26.54, p <.000), as shown in Figure 3. The groups also differed signi ficantly from one
51 another in their listening comprehension scores, again wi th the MO group scoring the highest, followed by the EL bilinguals and the LL bilinguals ( F (2,93) = 21.87, p =.0001). Figure 3. PPVT and OWLS standardized scores for all liste ner groups. A familiarity rating of 5.5 or higher on the nine-point scal e was given by the MO group for 143 of the 144 words, 138 of the words by the EL group, an d 118 of the words by the LL group. All but two of the lower-rated items were hard words. A one-way analysis of variance was conducted to determine whether d ifferences existed among the groups in how they rated the words for familiarity. Tho ugh all groups rated the easy words as more familiar than the hard words, there were group differences in how they rated these words for familiarity, ( F (2, 93) = 4.304, p = .016) (see Table 7 for means). Post hoc LSD tests revealed that the MO and the EL gro ups did not differ significantly from one another in how they rated the easy words. How ever, the LL group rated the easy words as significantly less familiar than both the MO and EL groups. For the hard words,
52 the same pattern of results was found. That is, the MO and EL groups did not differ from each other in how the rated the hard words for familia rity, but both groups differed significantly from the LL ( F (2,93) = 29.56, p <.001). Table 7. Vocabulary and Language Tests, Familiarity Rati ngs, and Demographic Variables *FE= average familiarity rating for the easy words **FH= average familiarity rating for the hard words Correlational Analyses. In order to further explore the relationship between the word recognition scores and various demographic factors, vo cabulary, and language skills, a series of correlational analyses (Pearson Â’s r) was performed. It was predicted that vocabulary level would be the best predictor of har d word recognition and that positive correlations would exist between the vocabulary level of the non-native listeners and their word recognition scores. Thus, it was necessar y to look at the correlations of age-of-immersion (AOI), length-of-residence (LOR), voc abulary level, and receptive language scores with word recognition in order to determine h ow strongly correlated each of these variables was with word recognition accuracy. T hen, regression analyses were conducted to find which variables predicted word recognition accur acy. Monolingual Early Learners Later-Learners PPVT OWLS FE* FH** PPVT OWLS FE* FH** PPVT OWLS FE* FH** Means 99.22 102.44 8.80 8.21 89.94 93.94 8.85 7.87 78.17 78.58 8.64 6.31 SEM 2.07 2.35 .05 .161 1.82 2.45 .02 .12 1.73 2.34 .06 .26 SD 12.43 14.07 .33 .97 10.90 14.70 .13 .70 8.46 11.47 .32 1.27 Range 55 51 1.22 4.40 48 52 .47 2.78 36 48 1.35 5.15 Min 72 77 7.78 4.6 72 73 8.53 6.14 60 53 7.65 3.39 Max 127 128 9 9 120 125 9 8.92 96 101 9 8.54
53 In each case, a correlation was run with all the groups together and in some cases, separate correlations were run for each group. The corr elations among the different variables for all groups considered together can be seen i n can be seen in Table 4. Table 8. Correlations among overall word recognition, voca bulary level (PPVT), receptive language scores (OWLS), and demographic variables ( AOI and LOR) for all listeners. Average Word Recognition PPVT OWLS AOI LOR Average Word Recognition .601* .572* -.702* .478* PPVT .548* -.526* .396* OWLS -.460* n.s. AOI -.656* LOR p < .005, (2-tailed) As shown in Table 8, AOI was significantly negatively c orrelated with overall word recognition scores (collapsed across easy and hard words in both quiet and noise) ( r = -.702, p < .001) for the non-native groups (note that the MO group did not have values for AOI, as with LOR, so this statistic reflects tha t of the non-native groups only). PPVT scores were also significantly correlated with overal l word recognition for the samples as a whole ( r = .601, p < .001). Receptive language scores, as measured by the OW LS, were also significantly correlated with overall word r ecognition scores ( r = .572, p < .001). Finally, LOR was significantly correlated with ov erall word recognition scores for the bilingual groups ( r = .478, p < 001).
54 Tables 9 and 10 show the correlations among the word rec ognition scores in the various conditions (easy, hard, quiet, and noise) with vocabulary level, receptive language scores, and familiarity ratings for the non-nati ve groups. Word recognition scores were not significantly correlated with any variabl es for the MO group. Table 9. Correlations among spoken word recognition, vocabul ary, listening comprehension scores, word familiarity, and demographic var iables for EL group Easy Word Recognition Hard Word Recognition Word Recognition in Quiet Word recognition in Noise PPVT .425** .339* .572** n.s. OWLS n.s. .412** .452** n.s. Familiarity Easy Words n.s. n.s. n.s. Familiarity Hard Words n.s. .324* n.s. AOI -.335* -.281* n.s. -.320* LOR .352* n.s. n.s. n.s. p < .05 (2-tailed) ** p < .005 Table 10. Correlations among spoken word recognition, vocabu lary, listening comprehension scores, word familiarity, and demographic var iables for LL group Easy Word Recognition Hard Word Recognition Word Recognition in Quiet Word recognition in Noise PPVT .499** .606** .565** .579** OWLS .485** .464* .527** .424* Familiarity Easy Words .559** .723** .659** Familiarity Hard Words .411* .391* .355* AOI -.531** -.661** -.582** -.658** LOR n.s. n.s. n.s. n.s. p < .05 (2-tailed) ** p < .005
55 Word recognition scores for the two non-native groups, how ever, were significantly correlated with many of these variables. For the EL, vocabulary scores, LOR, and AOI were moderately correlated with easy-word recognition. The same was true for the LL group, except that their easy-word recogni tion scores were also correlated with their familiarity ratings of the easy words and n ot with LOR. For the hard words, the EL and the LL groupsÂ’ vocabulary level, receptive language skills, and AOI were moderately correlated with word recognition. However, for the LL group, familiarity ratings were also correl ated with hard-word recognition. In looking at all the words in the quiet condition, the EL and LL groupsÂ’ vocabulary level, receptive language scores, and hard-word familiarity ratings were significantly correlated with word recognition. The LL groupÂ’s word recognition in quiet was also significantly correlated with easy-word famil iarity ratings and AOI. In the noise condition, AOI was significantly correl ated with word recognition for the EL group. However, the LL groupÂ’s word recognition in no ise was significantly correlated with AOI, vocabulary and receptive language sco res, and familiarity ratings. Next, a correlation was conducted that investigated the re lationship among the size of the easy-hard word effect (calculated as the diff erence between the easy and the hard word scores) and vocabulary level, receptive languag e, and demographic variables. As noted, the size of the easy-hard word effect was fo und to be greatest for the LL group relative to the NS and EL groups. Vocabulary and language scores as well as the demographic variables were all significantly correlated wit h the size of the easy-hard word effect for all groups: PPVT ( r = -.464, p < .001), AOI ( r = .690, p < .001), LOR ( r = -.373, p = .003), and OWLS ( r = -.496, p < .001). In other words, as vocabulary level,
56 receptive language skills, and length of residence increas e, the size of the easy-hard word effect decreases. However, as age-of-immersion increa ses, (i.e., the later listeners were immersed in an English-speaking environment) the size of t he easy-hard word effect gets larger. Regression Analyses. It was expected that vocabulary level of the non-na tive speakers would predict their word recognition accuracy. In order to confirm this prediction, stepwise multiple regression analyses were conducted in order to find the variables most predictive of performance on the word reco gnition task. An analysis was first performed which collapsed across group, word type, and noise condition. In other words, it looked for the predictive variable for the over all word recognition scores for all listeners. The independent variables were PPVT scores (vocabulary), OWLS scores (receptive language), and word familiarity ratings for the easy and hard words. For this analysis, the variables AOI and LOR could not be entere d into the equation because the NS group did not have values for them. PPVT accounted for 35.5 % of the variance ( = .601, p < .001), with OWLS ( = .347, p < .001) contributing an additional 8.4% of the variance, and easy-word familiarity ratings ( = .262, p = .001) contributing an additional 6.3%. The next set of analyses considered the non-native s peaker group only. Again, four stepwise multiple regression analyses were perform ed with word recognition in quiet, noise, easy words, or hard words as the dependent va riables, as described below. The independent variables were PPVT scores (vocabulary level), OWLS scores (receptive language), age-of-immersion, length of residenc e, and word familiarity. For the EL group, PPVT was found to be a significant predictor of word recognition for the
57 easy words ( = .425, p = .010) contributing 15.6% of the variance, with LOR ( = .363 p = .017) contributing an additional 13.2% of the variance. For the LL group, word familiarity was found to be a significant predictor of ea sy word recognition ( = .559, p = .005) contributing 28.1% of the variance. For the hard words, the EL groupÂ’s OWLS scores were found to be a significant predictor of hard word recognition ( = .412, p =.013), contributing 14.5% of the variance. For the LL group, AOI ( = -.504, p = .003) contributed 41.2% of the variance and, PPVT scores ( = .416, p = .012) contributed an additional 14.8% of the variance. In looking at all the words in the quiet condition, the EL groupÂ’s PPVT scores ( = .572, p < .001) accounted for 30.7% of the variance, with receptive langua ge ( = .323, p = .024) accounting for an additional 9.7% of the variance. For t he EL group, 50.2% of the variance was accounted for by their familiarity wit h the easy words ( = .583, p = .001), and their AOI accounted for an additional 9.7% of the var iance ( = -.342, p = .013). For the noise condition, the EL groupÂ’s vocabulary leve l accounted for 10.6% of the variance in the word recognition scores ( = .362, p = .030). For the LL group, two variables were found to be significant predictors of easy word recognition, familiarity ( = .468, p = .005) and AOI ( = -.465, p = .005), accounting for 40.9% and 17.9% of the variance, respectively. Digit Recall Another set of research questions concerned the effects of age-of-immersion, word-type, and noise on the recall of digits using the s ame memory-preload technique as Luce and Pisoni (1983). This task consisted of presenting the s ubjects with a list of digits
58 that they were to rehearse throughout the primary wordrecognition task. It was predicted that the LL group would experience more deleterious effect s of noise on their ability to recall the digits than would the EL group. A three-way analysis of variance was conducted to analyze the number of digits correctly reca lled and the percent correct words recognized in the various digit conditions. The depen dent variables were the number of digits recalled correctly and the percentage o f words accurately recognized. Listener group (three levels: MO, EL, and LL) was a be tween-subjects variable. Noise condition (two levels: quiet and noise) and digit recal l condition (three levels: 0, 3, or 6 digits) were within-subjects variables. The prediction was not supported because there were no significant effects found for the recall of dig its in this study. Semantic Features There were no predictions regarding the effects of the semantic characteristics of the words on word recognition because I did not have data for all the words nor did I control for the factors. Instead, the analyses were exploratory in nature with no a-priori expectations. Correlational analyses were conducted in order to investigate the relationship between the semantic features of the words and the accuracy with which they were recognized. There were no significant correlation s found between the word recognition scores and any of the semantic characterist ics for the subset of stimuli for which semantic characteristics could be calculated.
59 Discussion This study assessed the contributions of phonological ne ighborhood characteristics and memory load on spoken word recognition by monolingual English listeners and two groups of non-native bilingual listeners w ho differed in their age of immersion in an English-speaking environment. The predic tion that the easy/hard word effect on recognition would be greater for non-natives w as supported. The prediction that earlier-learning listeners would have better recognition o f the words compared to the later-learning listeners was confirmed and supports the hypoth esis that the ability to discriminate hard words is a skill that develops with kno wledge of the language. Further, the prediction that vocabulary level would correlate wi th word recognition accuracy such that those with lower vocabulary scores would also do more poorly on the word recognition task was also supported. The predictions regard ing the effects of the recall of digits on word recognition and the effects of phonologi cal neighborhood, noise, and ageof-immersion (AOI) on the recall of digits were not supported, which may have been due to a failure to manipulate memory load adequately. Finally, although the semantic characteristics of the word were not controlled and n o predictions regarding their effects were made, they were nevertheless explored. No signif icant effects of the semantic characteristics of the words on the accuracy with whic h they were recognized were found.
60 Phonological Neighborhood The stimuli in this study consisted of the easy and ha rd words used by Bradlow and Pisoni (1999). In their study, easy words had higher int elligibility than hard words, and this was especially true for the non-native listene rs. They argued that the ability to make the fine acoustic-phonetic distinctions required to discriminate the hard words is a skill that develops with knowledge of the sound-based syst em of the language. They found that all listeners identified words that were more easily discriminated from other words in their neighborhoods compared to words with many s imilar sounding neighbors, and this easy/ hard word effect was greatest for the bil ingual listeners. In the present study, a substantial easy-hard word effect w as obtained only for the earlier-learner listeners. Furthermore, oral vocabul ary size was significantly correlated with performance for the non-native listener groups only. Thus, the greater easy-hard effect for non-native listeners can be explained as a n effect of both phonetic proficiency and vocabulary size on the structure of lexical neighbo rhoods, and it seems that these skills are integrated. Garlock et al. (2001) noted that high-frequency words are mo re likely to overlap with many other words on a segmental basis, and the wo rds with which they overlap also tend to be high in frequency. The implications for this study are that the non-natives, especially the later-learners, may have different nei ghborhoods than the monolinguals, so that the words in their neighborhoods would overlap less (i.e., they have fewer English words in their neighborhoods, so less overlap is likely to result). However, these same English words might also overlap with words in their na tive language, causing greater difficulty with word recognition tasks. Later-learners may be likened to language-
61 learning children. Garlock et al. (2001) argued that children sho w a smaller competition effect than adults because they do not know as many words so there are fewer competitors in their neighborhoods. They maintained th at childrenÂ’s representations may not be as differentiated as adultsÂ’, so words from dense neighborhoods are not as impeded relative to words from sparse neighborhoods. Alternatively, because childrenÂ’s representations are un dergoing significant change, the recognition of hard words might be especially difficult and the recognition should be best for words that are most familiar and l ikely to be more stable and robust. This hypothesis could be extended to non-native speakers, es pecially for later-learners with relatively small vocabulary sizes. It seems r easonable to assume that, like children learning a first language, later-learnersÂ’ representations are also undergoing significant change. Following this argument, their representations for easy words might be more established than those of the hard words. If this were t he case, performance would be expected to be more similar between the earlier-learn ers and later-learners for the easy words compared to the hard words. The later-learners mig ht also have less competition for accurate word recognition because they do not have a s many words in their neighborhoods. As Marian and Blumfield (2006) explained, in a native language, word recognition tends to be better for words that are used of ten. In a non-native language, however, a L2 learner may have more limited exposure to and use of particular highfrequency words. This might give those words the same s tatus as low-frequency words, effectively making the neighborhood effects more pronoun ced in the L2. If later-learners have sparser neighborhoods overall, then the density of their neighborhoods might be expected to increase as their L2 v ocabularies increase. The
62 increased vocabulary (and by extension, proficiency) in the L2 would then allow greater ability to make fine-grained phonological distinctions bet ween neighbors (Imai et al., 2005). As evidenced by their lower PPVT scores, the non-nati ve listeners in this study had smaller L2 vocabularies, and thus, target words may have had fewer competitors. The results of this study support this line of reasoning: the easy-hard word effect was larger for the later-learner group compared to the earli er-learner and monolingual groups. Luce and Pisoni (1998) argued that the relatively sparse phonol ogical neighborhoods of children cannot be explained by smaller vo cabularies alone. They suggest that, because of the small size of their neighbo rhoods, children may use recognition strategies that are more holistic rather t han segmental. This is because the fine-grained phonetic discrimination strategies adults use d are not necessary given that childrenÂ’s neighborhoods are not as densely populated. The de scription of the developmental path of phonological neighborhood effects could shed light on the processes involved in word recognition for non-natives. N on-natives, especially those who are later learners, certainly have smaller L2 vocab ularies than native speakers. However, their neighborhoods may include words from their L1, which could compete with L2 words. The degree of phonological overlap betwee n the first and second language may impact bilingual word recognition in the L2. In fact, Boukrina and Marian (2006) manipulated cross-linguistic phonological overlap betwee n Russian and English in a lexical decision task and found that, as phonological o verlap increased, so did the speed and accuracy of responses in the L2, but not the L1. They suggested that facilitation of L2 lexical decisions occurred because of co-activation of wider L2 phonetic categories with similar L1 sounds.
63 The degree of activation of the target (in this case, E nglish) versus the non-target language (the listenersÂ’ various L1s) could vary with the level of proficiency of the individual. This was evident in the present study by the rate at which the groups chose neighbors versus non-neighbors when they incorrectly id entified the target word. All three groups were more likely to choose a neighbor for th e hard words than the easy words. This supports the findings of Roodenrys et al. (2002) who found that neighbors were more likely to be chosen if the target was less frequent and had many highfrequency neighbors. However, non-natives, especially the later-learner group, were more likely to choose a non-neighbor when they had incor rect word recognition. For the easy words, when in error, the monolingual and earlier-l earner group chose a neighbor 57% of the time, while the later-learner group chose a ne ighbor 44% of the time. For the hard words, the monolingual group chose a neighbor 84% of the t ime when in error, the earlier-learner group chose a neighbor 78% of the time, a nd the later-learner group chose a neighbor 71% of the time. Support for the Neighborhood Activation Model (Luce & Piso ni, 1998) comes from the finding that neighbors were more likely to be ch osen over non-neighbors for the hard words relative to the easy words. Easy words have fewer neighbors competing during recognition tasks, whereas hard words have many more n eighbors from which to choose that are higher in frequency compared to the tar get word. What is interesting is that the earlier-learner group chose neighbors at roughly t he same rate as the monolinguals for the easy words, and the later-learner gr oup chose significantly fewer neighbors than the other groups. For the hard words, howev er, the later-learner group chose significantly fewer neighbors compared to the oth er groups and the earlier-learner
64 group chose neighbors at a rate somewhat in between that of the native speakers and the later-learner group. So, the rate at which one chooses a neighbor when in error seems to increase with vocabulary size in the L2, especially fo r the easy words. The fact that the later-learner group chose more non-word neighbors than t he other groups is also interesting. Perhaps this group was less aware that som e of their responses were nonwords. It has been suggested that the degree of confusability for given words conveys information about the listenerÂ’s internal lexicon and t he relative accessibility of its component words (Goldinger et al., 1991). For this reason, r eceptive vocabulary size and listening comprehension scores were gathered to provide insi ght into their internal lexicons. Proficiency and Vocabulary Level Proficiency level of non-native speakers has not been c onsidered in studies that have investigated the easy/ hard word effect and cognitive demand using the digit recall technique. Spoken word recognition by non-native speakers depen ds largely on vocabulary development in the target language. Although st udies exist that have considered vocabulary level of children and adults in wor d recognition tasks (Garlock et al., 2001), no studies have considered vocabulary level as an index of L2 proficiency for non-natives in word-recognition tasks between easy and hard words. Imai et al. (2005) correlated proficiency (defined as the degree of accentednes s of the non-native speakers as measured by native listeners) with number of years o f English-language study and word recognition accuracy. Likewise, Bradlow and Pisoni ( 1999) performed correlational analyses of factors such as age of Engl ish study onset, number of years of
65 English study, and number of years in an English-speaki ng environment, but neither study directly measured the vocabulary level of their n on-native participants. The current study used several measures to classify non-na tive participants as higher or lower-proficiency. The PPVT was used to obtai n an objective measure of participantsÂ’ vocabulary level in English, the OWLS ga ve an index of their oral language comprehension, and the language background questionnaire provided inf ormation regarding their language use and dominance. Whereas the na tive English speakers scored higher than the non-natives on all language measur es, the earlier-learners scored higher on the PPVT and the OWLS and reported more frequen t use and better command of English than did the later-learners. It was predic ted that the vocabulary level of the bilinguals would influence their word recognition, with the e arlier-learners identifying more words correctly than the later-learners. Result s confirmed this prediction: monolinguals recognized more words than both bilingual groups, and the earlier-learners recognized more words than the later-learners. This find ing echoes the results of previous studies that demonstrated that later-learners h ave more difficulty in Englishword recognition tasks than do later-learners (Imai et a l., 2005; Rogers et al., 2004). The non-native groups in this study differed in the exten t to which their vocabulary levels were predictive of their word recogni tion scores. The PPVT was the best predictor of easy word recognition for the earlier-l earner group but familiarity with the easy words was the best predictor for the later-le arner group. The OWLS, which is a measure of receptive language skills, was the best predic tor of hard word recognition for the earlier-learners and age-of-immersion was the bes t predictor for the later-learners (though their vocabulary level accounted for additional v ariance for the latter group).
66 Moreover, vocabulary was strongly correlated with ea syand hard-word recognition for the earlier-learners, compared to age-of-immersion or le ngth-of-residence. For the laterlearner group, vocabulary and age-of-immersion shared si milar correlations with word recognition for the hard words. For the easy words, age -of-immersion and vocabulary level were similarly correlated with word recognition scores for this group. One explanation for these findings is that the earlier-le arners all had very low ages-ofimmersion with a relatively restricted range compared t o the later-learners, suggesting that if age-of-immersion is held comparatively constant, it is vocabulary that differentiates performance on word recognition tasks. Marian and Blumenfeld (2006) also used the PPVT as a measure of language proficiency in their study exploring the role of phonolog ical density in lexical access in native and non-native languages. They found it to be a bet ter predictor of naming accuracy than age of acquisition and suggested that bilinguals can improve L2 performance with increased proficiency, regardless of age of acquisition. Their findings and those of the present study support the argument that incr easing vocabulary level may result in greater attention to fine phonetic detail. Po sitive correlations were found between vocabulary level and word recognition scores, s uggesting that lexical development influences phonological knowledge of the target language or, at least, that lower-level, phonological knowledge and higher level, l exical knowledge influence one another. The relationship between vocabulary level and word recogni tion accuracy for the bilingual groups can be better understood when one considers th e literature on childrenÂ’s spoken word recognition abilities. Garlock et al. (2001) off ered two proposals regarding
67 the neighborhood competition effects for children as they relate to vocabulary level. On the one hand, the effects of competition might be sma ller because they do not know as many words as adults; therefore, the recognition of dense (hard) words relative to sparse (easy) words might not be as difficult for them compar ed to adults. On the other hand, when their phonological representations are undergoing s ignificant change, children might experience greater effects from competition, mak ing the hard words effectively harder. The authors argued that, in order to distinguish among increasing numbers of items in the mental lexicon, spoken word representations must become more segmentally structured. Thus, it is vocabulary growth that drives c hanges in the lexical representation of words. The applicability of Garlock et al.Â’s (2001) hypothesis to n on-native speakers is apparent. The authors state that words that are least robust and stable in terms of their familiarity and neighborhood status should undergo the great est developmental change in spoken word recognition. Hard words require fine-grained repres entations for accurate recognition. If differences in vocabulary development drive differences in performance between children and adults, then by extension, differen ces in L2 proficiency may drive differences in word recognition accuracy among non-native speakers. As with children and adults, non-native speakers who differ in L2 proficien cy should demonstrate more similar word recognition for easy words because they c an be recognized on a more segmental basis due to their more robust representation and less need for fine-grained distinctions. The segmental recognition of easy words is less hampered than that of hard words because they contrast with fewer words on a sing le phoneme basis.
68 Noise In this study, listeners identified English words spoken by two native English speakers and later rated the words for their familiarity. The word lists were presented in either quiet or in noise as it is expected that noise w ould make phonetic discrimination harder. The level of noise added to the stimuli was sele cted based on pilot testing and designed to cause relatively equal decrements to word rec ognition. Specifically, the noise level was intended to bring word recognition scores down t o approximately 70% and 75% of the word recognition scores in quiet. This was mo stly true for all groups: the groupsÂ’ easy word scores in noise averaged between 77% and 82% of that in quiet and the hard word scores in noise averaged between 67% and 72% of that in quiet. Thus, the same level of noise had a different impact on the eas y and hard words for all groups such that the hard words were more deleteriously affected, po ssibly because accurate recognition of hard words requires the ability to make fine ph onetic distinctions among phonetic cues, some of which might be masked in noise. Digit Recall It was predicted that the later-learners would experienc e more deleterious effects of noise on their ability to recall digits than the e arlier-learners because of the additional short term memory capacity needed to rehearse the digits which would subsequently leave less capacity for the encoding of words in the re cognition task. Groups showed no effect on the number of digits recalled under any cond itions. That is, presumably increasing the cognitive demand through the manipulation of no ise, word type, or number of digits to recall did not affect recall of the digi ts. This may have been because the manipulation of memory load failed. Anecdotally, the par ticipants in this study seemed
69 to not pay much attention to the digits during their presen tation and were guessing during the recall task. It is, therefore hard to say whether cognitive demand was actually manipulated. Luce and Pisoni (1983) were looking at the recall of syntheti c versus natural words when using the memory preload technique. In other words, they used a memory task to interfere with a subsequent memory task. It was used in this study to increase cognitive demand in a recognition task. Further, they did not find di fferential effects of digit preload across the natural and synthetic lists. Luce an d Pisoni (1983) did find that the number of subjects who recalled all the digits accurat ely decreased in the synthetic condition compared to the natural condition, especially for the six-digit condition relative to the three-digit condition. By extension, it was ex pected in this study that in the noise condition, subjects would recall fewer digits. That wa s not the case. The reasons for the null findings for digit recall might become clearer when considering the model of memory by Baddeley and Hitch (1974). Baddeley and Hitch (1974) proposed the idea that memory is composed of three m ain components: the central executive which controls the flow of information to and f rom its slave systems: the phonological loop, and the visuo-spatial sketch pad. The la tter two systems are shortterm storage systems for the verbal and visuo-spatial dom ains respectively. In 2000, Baddeley added a fourth system to his model, the episodic b uffer, which links information across domains with time sequencing and has as sociations with long-term memory and semantic meaning. The phonological loop deals with sound or phonological information and consists of two parts: the short-term phonological store which rapidly decays and an articulatory rehearsal component that keeps the memory traces active.
70 Auditory information is thought to enter into the phonol ogical store, whereas visually presented speech is transformed into a phonological code by silent articulation and thereby is encoded into the phonological store. The phono logical store remembers speech sounds in their temporal order, while the articulatory re hearsal component repeats the series of words to prevent them from decaying. Further, there seems to be an effect of phonological similarity such that lists of words that are similar in sound are harder to remember than words that do not sound alike. In contrast, semantic similarity does not seem to have an effect on memory, supporting the assumpt ion that verbal information is coded phonologically in working memory. Baddeley and Hitch (1974) found that performance of two simult aneous tasks which used two separate perceptual domains (e.g., a verbal and a visual task) is nearly as efficient as performance of the tasks individually. In contrast, performance of two simultaneous tasks requiring use of the same perceptual doma in is less efficient than when performing the tasks individually. Thus, there is le ss interference between visual and verbal tasks than between two visual tasks or two ve rbal tasks. The present study required participants to remember a visually presented list o f digits shown before each of the word lists was presented auditorally. Perhaps, a tas k in which the to-be-remembered material was an auditorally presented list of similar sounding words would have yielded an effect for the cognitive demand condition. It would a lso be of use to look at the reaction time data to determine if differences existed a mong the groups in the time it took to record the words and the digits under the various cond itions. According to the model presented by Baddeley and Hitch (1974), the presentation of ma terial through an auditory mode would have interfered more with the recognition of the auditorally presented words
71 because both tasks require use of the same perceptual domain To reiterate, however, the manipulation of cognitive load most probably failed in this study. Semantic Characteristics The null findings for semantic characteristics could be explained by the fact that many of the words did not have values and these character istics were not manipulated in this study. Semantic network characteristics provide inf ormation about the relationships among words based on their meanings. While some of the s emantic network characteristics differed between the word lists, a mor e controlled manipulation of these indices might reveal that the way words are semantical ly organized in the mental lexicon does influence their recognition. Future word recognition st udies should be conducted with words in which semantic and neighborhood character istics are varied orthogonally in order to tease apart the contributions of each. A limitation of this study is that not all the words had data points for each of the semantic feat ures, especially the hard words. For example, only 62 of the easy words had values for con nectivity, while the hard words only had 34. It should be noted that the semantic character istics may have an effect on word recognition accuracy, but because they were not con trolled in this study, it is difficult to speculate. Future Directions This study provided valuable information about differences bet ween native and non-native listeners in their recognition of English words and the contributions of proficiency level, neighborhood characteristics, and cogni tive demand. However, future work is needed in order to more fully understand the fa ctors that affect L2 word recognition. First, consider the stimuli. The target frequency of the words was based on
72 the frequency counts from the Brown Corpus of printed text (Kucera & Frances, 1967). Though these norms are somewhat old, their use reflects the notable absence of more recent frequency counts and the relative lack of spoken w ord frequency counts. The implications of using frequency counts based on written te xts is that the lists are very sensitive to the corpora from which they are drawn, parti cularly to the style, language, and content of the corpora. For example, a list gener ated from six million words of newspaper articles is likely to be significantly differ ent from a list generated from six million words of internet postings or magazines. The re is a newly created list of spoken word frequency counts available containing 1.6 million Ameri can-English words (Pastizzo & Carbone, 2007). This list was derived from the Michigan Corpus of Academic Spoken English (MICASE) which includes 152 transc riptions of lectures, meetings, advisement sessions, public addresses, and other educ ational conversations spoken by students, faculty, and other staff members and re corded at the University of Michigan. The authors found a moderately strong, positi ve correlation between log written frequency and log spoken frequency and suggested that a written measure can be replaced with spoken counts. Future work in second-language s peech perception should, therefore, consider using spoken frequency counts. Further consideration of the stimuli for future resear ch could involve investigating the contributions of the semantic characteristics of words to their recognition especially for bilingual populations. Although this study attempted to e xplore the effects of the semantic characteristics of the stimuli, they were not manipulated or controlled in any manner. A future study could manipulate both phonological ch aracteristics and semantic
73 characteristics (set size, connectivity, and concrete ness) to determine if these variables affect the recognition of words. Second, consider the types of errors that monolinguals versus bilinguals are making, specifically as they relate to neighborhoods and t he changes associated with increasing proficiency level. Considering the degree of ph onological and semantic (in terms of cognates) overlap between the target language a nd that of the non-native participants may also shed light on how different languag e backgrounds may affect the recognition of L2 words. Finally, investigating the contribution of probabilistic pho notactics would provide information about how non-native listenersÂ’ mental l exicon is organized compared to that of native listeners. Phonotactics refers to a syst em of rules or constraints that dictate the permissibility of the occurrence of segments within syll ables and words of a language (Auer & Luce, 2003). For example, in English, /la/ may l egally occur at the beginning of a syllable, whereas /lda/ may not. Further, these permi ssible segments and their sequences occur more or less frequently in a language (e.g., /tra/ occurs frequently in English, whereas /kwa/ occurs less frequently). Probabi listic phonotactics refers to the relative frequencies of segments occurring in a listene rÂ’s language (Auer & Luce, 2003). Vitevitch et al. (1998) explained that the neighborhood dens ity effects have a lexical focus, whereas probabilistic phonotactics effec ts have a sub-lexical focus. The facilitative effects of probabilistic phonotactics for non-words occur because non-words fail to activate competing lexical representations. The refore, the processing of highprobability non-words benefits from the absence of lexica l competition in the presence of high frequency segments. The easy-hard word effect could be minimized or reversed in
74 favor of probabilistic phonotactics by controlling the ne ighborhood density of the words while varying their phonotactic probability. For exampl e, consider a word that is so unfamiliar to an individual with very low proficiency in th e L2 that the word is effectively a non-word to the listener. Would recognition of the word be facilitated from its probabilistic phonotactics in ways that would not bene fit a native listener because of the native listenerÂ’s lexical focus? Or would the nonnative require greater proficiency in the L2 before probabilistic phonotactics shows its faci litative effects? Answers to these questions would have important implicatio ns for the teaching and assessment of non-native speakers and for the ways in which teachers can improve non-native speakersÂ’ comprehension of spoken material. B ilingual listeners have greater difficulty perceiving speech in their L2 than do native li steners, especially under adverse listening conditions and under conditions of increased cognit ive load, such as noise. Thus, the findings of this study for second language pedagogy ar e apparent. Classrooms can be quite noisy, and the task of comprehending a lectur e is made more difficult when the subject matter is advanced, especially for second-lan guage learners. Some specific recommendations for Speech-language Patholo gists and teachers of second-language learners may prove helpful. It is importa nt for those working with nonnative speakers to consider the mode through which lecture s and assignments are delivered. Increasing second-language learnersÂ’ vocabulary ma y help with their second language speech perception to the extent that a larger voc abulary may enhance phonetic discrimination skills. In other words, the more words on e has in his or her vocabulary, the more necessary it becomes to be able to make the f ine phonetic distinctions needed to discriminate among similar sounding words. But the vocab ulary instruction ideally
75 should be done orally rather than through written text and via multiple exemplars of the same word. Relying exclusively on written work deprives th e second-language learner the opportunities to hear correct pronunciation of target w ords and how the word contrasts with similar sounding words. Oral presentati on and practice during vocabulary instruction should also involve immediate feedback to he ighten the learnerÂ’s awareness of correct pronunciation and their own mispronunciation s. This study and others like it should be of interest to t hose who wish to promote intelligibility and comprehensibility in the classroom by incorporating communication strategies that offset the effects of noise when capa city demand is high. Clearly, we need more studies to elucidate the factors that have the grea test impact on intelligibility and comprehensibility for non-native speakers and listeners.
76 References Allen, R., & Hulme, C. (2006). Speech and language processing m echanisms in verbal serial recall. Journal of Memory and Language, 55, 64-88. Auer, Jr., E. T. & Luce, P. A. (2003). Probabilistic phonot actics in spoken word recognition. In C. T. McLennan, P. A. Luce, G. Mauner, & J. Charles-Luce (Eds.), University at Buffalo Working Papers on Language and Perception Vol. 2, (pp. 164 Â– 202) Buffalo, NY: University of Buffalo. Baddeley, A., & Hitch, G. (1974). Working memory. In G.H. Bower (Ed.). The psychology of learning and motivation: Advances in research and theory, (pp. 4789). New York, NY: Academic Press. Baddeley, A. (2000). The episodic buffer: A new component o f working memory? Trends in Cognitive Science, 4, 417-423. 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. 177-204). Timonium, MD: York Press. Bilger, R., Neutzel, J., Rabinowitz, W., and Rzeczkowski C. (1984). Standardization of a test of speech perception in noise. Journal of Speech and Hearing Research, 27 32-48.
77 Boukrina, O., & Marian, V. (2006). Integrated phonological pro cessing in bilinguals: Evidence from spoken word recognition. Proceedings of the TwentyEighth Annual Meeting of the Cognitive Science Society Mahwah, NJ: Lawrence Erlbaum. Bracken, B., & McCallum, R. (1998). Universal Nonverbal Intelligence Test (UNIT). Itasca, IL: Riverside Publishing. Bradlow, A., & Pisoni, D. (1999). Recognition of spoken words by native and nonnative listeners: talker-, listenerand item-related f actors. Journal of the Acoustical Society of America, 4, 2074-2085. Carrow-Woolfolk, C. (1996). OWLS: Oral and Written Language Scales Bloomington, MN: Pearson Assessments. Dunn, L., & Dunn, L. (1997 ). Peabody Picture Vocabulary Test-3 rd EditionRevised. Circle Pines, MN: American Guidance Service. Edwards, J, Fox, R.A., & Rogers, C.L. (2002). Final conso nant discrimination in children: Effects of phonological disorder, vocabulary size, and phonetic inventory size. Journal of Speech, Language, & Hearing Research, 45, 231-242. Ellis, L.W. & Fucci, D. J. (1992) Effects of listenersÂ’ experience on two measures of intelligibility. Perceptual and Motor Skills, 74 1099-1104. Flege, J. E. & Hillenbrand, J. (1984). Limits on the phonet ic accuracy in foreign language speech production. Journal of the Acoustical Society of America 76, 708-721. Flege, E., & Fletcher, K. (1992). Talker and listener effec ts on degree of perceived foreign accent. Journal of the Acoustical Society of America 91 (1), 370-389. Flege, J. E. (1995). Second language speech learning: Theory, findings, and problems. In
78 W. Strange (Ed.), Speech perception and linguistic experie nce: Issues in crosslanguage research (pp. 233-277). Timonium, MD: York Press. Flege, J., Bohn, O. & Jang, S. (1997). Effects of experienc e on non-native speakers' production and perception of English vowels. Journal of Phonetics, 25 (4), 437470. Flege, J., Schirru, C. & MacKay, I. (2003). Interaction between the native and second language phonetic subsystems. Speech Communication, 40 467-491. Garlock, V., Walley, A., & Metsala, J. (2001). Age-of-acqu isition, word frequency, and neighborhood density effects on spoken word recognition by chi ldren and adults. Journal of Memory and Language, 45 468-492. Goh, W., & Pisoni, D. (2003). Effects of lexical competiti on on immediate memory span for spoken words. The Quarterly Journal of Experimental Psychology, 56A(6), 929-954. Goldinger, S., Luce, P., & Pisoni, D. (1989). Priming lexical neighbors of spoken words: Effects of competition and inhibition. Journal of Memory and Language, 28, 501-518. Goldinger, S., Pisoni, D., & Logan, J. (1991). On the natur e of talker variability effects on the recall of spoken word lists Journal of Experimental Psychology, 12(1), 152-162. Grosjean, F. (1997). Processing mixed languages: Issues, findi ngs, and models. In A. M. de Groot & J. F. Kroll (Eds.), Tutorials in bilingualism: Psycholinguistic perspectives (pp. 225-253). Hillsdale, NJ: Erlbaum. Imai, Walley, & Flege, J. (2005). Lexical frequency and neig hborhood density effects
79 on the recognition of native and Spanish-accented words by na tive English and Spanish listeners Journal of the Acoustical Society of America, 117(2), 896-907. Kalikow, D., Stevens, K., & Elliot, L. (1977). Developm ent of a test of speech intelligibility in noise using sentence materials with controlled word predictability. Journal of the Acoustical Society of America, 61 1337-1351. Koul, R. K. & Allen, G. D. (1993). Segmental intelligibi lity and speech interference thresholds of high-quality synthetic speech in presence o f noise. Journal of Speech and Hearing Research, 36 790-798. Kucera, F., & Francis, W. (1967). Computational analysis o f present day American English. Providence, RI: Brown U.P. Lewellen, M., Goldinger, S., Pisoni, D., & Greene, B. (1993). Lexical familiarity and processing efficiency: Individual differences in naming, l exical decision, and semantic categorization. Journal of Experimental Psychology: General, 122(3), 316-330. Lisker, L., & Abramson, A. (1971). Distinctive features a nd laryngeal control. Language, 47(4), 767-785. Lively, S., Logan, J., & Pisoni, D. (1993). Training Japanese listeners to identify English /r/ and /l/. II: The role of phonetic environme nt and talker variability in learning new perceptual categories. Journal of the Acoustical Society of America, 94 (3), 1242-1255. Lively, S., Pisoni, D., VanSummers, W., & Bernacki, D (1993). Effects of cognitive workload on speech production: Acoustic analysis and perceptua l consequences. Journal of the Acoustic Society of America, 93 (5) 2962-2973.
80 Logan, J. S. Greene, B. G., & Pisoni, D. B. (1989). Segme ntal intelligibility of synthetic speech produced by rule. Journal of the Acoustical Society of America, 86(2) 566-581. Luce, P. (1986). Neighborhoods of words on the mental lexico n. Research on Speech Perception, Technical Report No. 6 Bloomington, IN: Indiana University Luce, P., Feustel, T., & Pisoni, D. (1983). Capacity deman ds on short-term memory for synthetic and natural speech. Human Factors, 25, 17-32. Luce, P. & Pisoni, D. (1998). Recognizing spoken words: The nei ghborhood activation model. Ear & Hearing, 19 1-36. Majerus, S., Van der Linden, M., Mulder, L., Meulemans, T. & Peters, F. (2004). Verbal short-term memory reflects the sublexical organiz ation of the phonological language network: Evidence from an incidental phonotactic lea rning paradigm. Journal of Memory and Language, 51 297-306. Majerus, S., Poncelet, M., Greffe, C., & Van der Linden M. (2006). Relations between vocabulary development and verbal short-term memory: The relative importance of short-term memory for serial order and item informa tion. Journal of Experimental Child Psychology, 93 95-119. Marian, V., and Blemenfeld, H. (2006). Phonological neighb orhood density guides lexical access in native and non-native language productio n. Journal of Social and Ecological Boundaries: Special Issue on Bilingualism, 2, 3-35 Mayo, L., Florentine, M., & Buus, S. (1997). Age of secon d language acquisition and perception of speech in noise. Journal of Speech and Hearing Research, 40 (3), 686-693.
81 McAllister, R. (1996). Perceptual foreign accent: L2 user sÂ’ comprehension ability. In A. James & J. Leather (Eds.) Second language speech structure and process (pp. 119-132). Berlin: Mouton de Gruyter. Meador, D., Flege, J., & Mackay, I. (2000). Factors affec ting the recognition of words in a second language. Bilingualism: Language and Cognition, 3, 55-67. Miller, G. & Nicely, P., (1955). Analysis of perceptual c onfusions among some English consonants. Journal of the Acoustical Society of America, 27 338-353. Munro, M. & Derwing, T. (1995). Foreign accent, comprehensi bility, and intelligibility in the speech of second language learners. Language Learning, 45 (1), 73-97. Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). The University of South Florida word association, rhyme, and word fragment norms. Available from: http://www.usf.edu/FreeAssociation/ Retrieved December, 2006. Paris, C., Gilson R., Thomas, M., & Silver, N. (1995). Effect of synthetic voice intelligibility upon speech comprehension. Human Factors, 37 335-340. Paris, C., Thomas, M., Gilson, R., & Kincaid, J. (2000). Linguistic cues and memory for synthetic and natural speech. Human Factors, 42 421-431. Pastizzo, M. and Carbone, R. (2007). Spoken word frequency counts based on 1.6 million words in American English. Behavior Research Methods, 39, 1025-1028. Pichura-Fuller, M.C., Schneider, B., & Daneman, M. (1995). How young and old adults listen to and remember speech in noise. Journal of the Acoustical Society of America, 97 (1), 593-608. Pisoni, D., Nusbaum, H., Luce, P., & Slowiaczek, L. (1985). Speech perception, word recognition, and the structure of the lexicon. Speech Communication, 4 75-95.
82 Pisoni, D. & Koen, E. (1981). Some comparisons of intellig ibility of synthetic and natural speech at different speech-to-noise ratios. Tech. Rep. 7, Research on Speech Perception Progress Report. Bloomington, IN: Indiana University. Pisoni, D., Manous, L., & Dedina, M. (1987). Comprehension of natural and synthetic speech: Effects of predictability on the verification o f sentences controlled for intelligibility. Computer Speech and Language, 2 303-320. Ralston, J., Pisoni, D., Lively, S., Greene, B., & M ullennix, J. (1991). Comprehension of synthetic speech produced by rule: Word monitoring and sent ence-by-sentence listening time. Human Factors, 33 (4), 471-491. Rochet, B. L. (1995). Perception and production of second-lan guage speech sounds by adults. In W. Strange (Ed.), Speech perception and linguistic experience: Issues in cross-language research (pp. 379-410). Baltimore, MD: York Press. Rogers, C., Dalby, T., & Nishi, K. (2004). Effects of nois e and proficiency on intelligibility of Chinese-accented English. Language and Speech, 47 (2), 139154. Rogers, C., Lister, J., Febo, D., Besing, J., & Abram s, H. (2006). Effects of bilingualism, noise, and reverberation on speech perception by listeners with normal hearing. Applied Psycholinguistics, 27, 465-485. Roodenys, S., Hulme, C., Lethbridge, A., Hinton, M., & Ni mmo, M. (2002). Wordfrequency and phonological-neighborhood effects on verbal short-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(6), 1019-1034. Sitler, R. W., Schiavetti, N. & Metz, D. E. (1983). Cont extual effects in the measurement
83 of hearing-impaired speakersÂ’ intelligibility. Journal of Speech and Hearing Research, 26 22-30. Sommers, M. (1995). The structural organization of the menta l lexicon and its contribution to age-related declines in spoken work recogniti on. Psychology and Aging, 11 (2) 333-341. Southwood, M.H., & Flege, J.E. (1999). Scaling foreign acc ent: direct magnitude estimation versus interval scaling. Clinical Linguistics and Phonetics, 13, 335349. Stine, E., & Wingfield, A. (1987). Process and strategy in memory for speech among younger and older adults. Psychology and Aging, 2 272-279. Strange, W. (1995). Cross-language studies of speech perception : A historical review. In W. Strange (Ed.) Speech perception and linguistic experience: Issues in crosslanguage research ( pp. 3-45). Timonium, MD: York Press, Strange, W. (1999). Perception of consonants: From varian ce to invariance. In J. M. Pickett (Ed.), The acoustics of speech communication: Fundamentals, speech perception theory, and technology (pp. 166-182). Needham Heights, MA: Allyn & Bacon. Strange, W., Bohn, O., Trent, S., & Nishi, K. (2004). Ac oustic and perceptual similarity of North German and American English vowels. Journal of the Acoustical Society of America, 115 (4), 1791-1807. Studebaker, G. (1985). A Â“rationalizedÂ” arcsine transform. Journal of Speech and Hearing Research, (28), 455-462. Suenobu, M., Kanzaki, K., & Yamanc, S. (1992). An experime ntal study of intelligibility
84 of Japanese English. IRAL: International Review of Applied Linguistics in Language Teaching, 30, 146-156. Takata, Y., & Nabelek, A. (1990). English consonant recognit ion in noise and in reverberation by Japanese and American listeners. Journal of the Acoustical Society of America, 88 (2), 663-666. TOEFL iBT Tour (n.d.). Retrieved March 28, 2007 from http://www.ets.org/Media/Tests/TOEFL/tour/highrez/startweb_content.html Vitevitch, M. (2002). The influence of phonological simil arity neighborhoods on speech production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(4), 735-747. Walley, A. (1993). The role of vocabulary development in childrenÂ’s spoken word recognition and segmentation ability. Developmental Review, 13, 286-350. Werker, J., Fennell, C., Corcoran, K., & Stager, C. (2002). InfantsÂ’ ability to learn phonetically similar words: Effects of age and vocabula ry size. Infancy, 3(1), 130. Wijngaarden, S. (2001). Intelligibility of native and non-na tive Dutch speech. Speech Communication, 35, 103-113. Wijngaarden, S., Herman, J., & Houtgast, T. (2002). Quantif ying the intelligibility of speech in noise for non-native listeners. Journal of the Acoustical Society of America, 111 (4), 1906-1916. Wright, R. (1997). Lexical competition and reduction in spe ech: A preliminary report. Research on Spoken Language Processing, Progress Report Num ber 21 Bloomington, IN: Indiana University.
86 Appendix A: Monolingual Language Background Questionnaire Participant Background Questionnaire Name: __________________ Age: _____ Address (town & state): ______________ 1. Is English your first (native) language? Circle one : Yes No 1a. If you answered Â“NoÂ” to (1) above, list your first language here. 2. Did you speak any languages other than English while growing up ( other than classroom instruction)? Circle one: Yes No 2a. If you answered Â“YesÂ” to (2) above, list those language s here __________ 3. List any languages you speak other than English and rate your degree of proficiency on a scale from Â“1Â” to Â“5Â” for each (1=beginner, canÂ’t h ave a conversation; 5=like a native speaker): _____________________________________________________ 4. Have you ever been diagnosed with a speech or hearing d isorder or had speech or hearing difficulties? Circle one: Yes No a. If you answered Â“yesÂ” to (4), above, please explain in th e space provided below (or on back if you need more room): _____________________________________________________ 5. How long have you lived in Florida (or current state)? ______________________ 6. What state were you born in and how long did you live th ere? __________________ (donÂ’t answer #Â’s 7 or 8 if youÂ’ve lived all your life in 1 state) 7. What state have you lived the longest in? ______________________ a. How many years did you live there? ___________________ 8. List any other states that youÂ’ve lived in for over a y ear (if more than 3, list top three): ______________________________________________ 9. On a scale from Â“1Â” to Â“7,Â” rate your experience with l istening to speakers with a foreign accent (1=little or no experience; 7=every day or very frequent): _______
87 Appendix B: Bilingual Language Background Questionnaire Participant Background Questionnaire Name: __________________ Age: _____ Address (town & state): ______________ 1. How many years have you lived in your current area (town & state)? __________ 2. Have you ever been diagnosed with a speech or hearing d isorder or had speech or hearing difficulties? Circle one: Yes No a. If you answered Â“yesÂ” to (2), above, please explain in the space provided below (or on back if you need more room): __________________________________________________________ 3. What language(s) did your parents speak with you? ________________ a. If you answered with more than one language in (1), above which language(s) did each parent speak with you? _________________________________________________________ 4. Where were you born (give city, state, country) ___________________________ a. How many years did you live there? ______ b. List other cities or regions youÂ’ve lived in for more tha n one year and note number of years you lived there for each. ___________________________________________________ c. What city and country are your parents from? Mother: _______________________ Father: ___________________ 5. How old were you when you began learning English? _________ a. Why did you begin learning English? _______________________ ______________________________________________________ 6. If you moved to the United States from another country, ho w much did you speak English before moving here (describe years of study, if you learned English in a classroom & percent of time speaking English)? _______________________________________________________
88 Appendix B: (Continued) 7. If you moved to the United States from another country, ho w long have you lived here? ___________ years, ______________ months. 8. On a typical day, what percent of your time do you spend s peaking English at work? _____ % At home? _________% Other (shopping, etc.)? _____% 9. On a typical day, what percent of your time do you spend speaking a language other than English at work? ______ % At hom e? _____ % Other (shopping, etc.)? ____% (if more than one, answer be low for each language) 10. What percent of your day do you spend with people with people who speak both (or more) languages that you do? ________ % 11. What language are you most comfortable speaking? ______________ a. How much more comfortable are you in speaking that language on a scale of 1 to 5? (1=equal or nearly equal comfort; 5=much more comfortabl e) _____ 12. What language are you most comfortable listening in? ______________ a. How much more comfortable are you in listening in that la nguage on a scale of 1 to 5? (1=equal or nearly equal comfort; 5=much more comfort able) ______ 13. What language are you most comfortable reading in? ______________ a. How much more comfortable are you reading in that language o n a scale of 1 to 5? (1=equal or nearly equal comfort; 5=much more comfortabl e) ______ 14. What language are you most comfortable writing in? ______________ a. How much more comfortable are you writing in that language on a scale of 1 to 5? (1=equal or nearly equal comfort; 5=much more comfortabl e) ______ 15. Do you think your ability in the language you are less com fortable in is still improving for any of the skills in questions 9-12? Circle one: yes no a. If you answered yes in 13 above, indicate which abilities you believe are still improving. Circle any that apply: speaking listening readingwriting
89 Appendix B: (Continued) 16. What academic degrees have you earned? (list language o f education for each) 17. For all languages that you speak, rate your level of ability on a scale of 1 to 5 (1=not proficient, like a child or beginner; 5=very proficient, l ike a well-educated native speaker) for each of the following areas: b. Comprehension __________________________________________ c. Fluency (ease of expression) ________________________________ d. Vocabulary: _____________________________________________ e. Pronunciation: ___________________________________________ f. Grammar: _______________________________________________
90 Appendix C: Easy and Hard Word Lists Easy Words Hard Words fig live dog ban rum pawn down move vote bead sane bun work food league bean soak gut long size thick bug suck lice both cause page bum tan mid does chief join cheer weed hurl put faith shop comb whore moat give pool roof cot wick teat young deep leg den con hash thing firm lose dune doom hid peace serve theme fade hick hoot god reach soil fin rut mace five mouth pull goat toot wad gave teeth chain knob wade moan death gas curve lad bud mum shall jack path mall dame rim real check dirt mat lace rout south king vice mitt lame wail job shape rough mole pad hum love learn balm pat chore sill full ship noise pet cod beak wife neck thought pup hack hag voice watch rat kin girl judge rhyme kit wrong hung chat wed
91 Appendix D: Target Words Phonological Neighborhood and Sema ntic Features Data Easy Word Frequency Neighborhood Density Neighborhood Frequency Connectivity Set Size Concreteness balm 36 13 17.77 4.59 17 5.91 both 730 13 22.38 cause 130 10 33.3 4.86 19 2.9 chain 50 19 39.42 3.36 14 5.85 check 88 15 15.93 4.28 13 4.38 chief 119 12 10.42 2.26 10 4.82 curve 45 13 15.23 4.1 20 4.4 death 277 10 30.7 4.53 15 3.86 deep 109 18 36.17 4.6 13 3.96 dirt 43 15 23.4 7.52 18 5.51 does 485 16 30.5 dog 75 8 11.875 3.56 5 5.75 down 895 20 38.7 3.78 10 3.23 faith 111 11 50.09 4.29 12 2.71 fig 72 16 44.4375 3.96 10 6.28 firm 109 13 13.69 6.65 20 3.96 five 286 12 46.5 3.34 14 3.53 food 147 11 24.91 5.31 18 5.84 full 230 15 59.53 3.52 11 3.74 gas 98 19 25.68 4.29 10 5.34 gave 285 18 47.67 girl 220 16 6.69 4.12 8 6.83 give 391 7 70.4286 5.9 13 3.18 god 318 19 77.32 7.07 23 3.61 hung 65 18 30.56 2.85 19 3.88 jack 92 17 74.41 17 5.2 job 238 19 5.32 5.45 7 4.11 join 65 8 27 4.49 20 2.88 judge 77 6 2.33 4.77 17 6.25 king 88 17 36.12 4.7 8 5.54 league 69 19 24.47 4.75 13 learn 84 16 51 7.46 19 3.66 leg 58 15 79.67 5.06 10 6.04 live 177 15 61.07 6.2 18 4.32 long 755 13 75.85 3.94 13 3.68 lose 58 17 65.76 love 232 11 42.45 6.45 18 3.51 mouth 103 7 41.86 7.34 19 5.47 move 171 8 16.38 10.19 27 3.81 neck 81 13 15.9231 4.4 19 5.83 noise 37 4 43.5 5.8 14 5.29 page 66 16 52.94 4.07 10 5.85 path 44 14 16.5714 4.17 12 4.93 path 327 19 18.16 3.19 19 2.98 peace 111 18 25.28 3.64 13 6.29 pool 51 16 64.81 4.61 10 3.4 pull 437 14 20 5.41 16 2.77 put 106 20 77.45 3.66 16 3.55 reach 260 16 23.44 14 3.77
92 Appendix D (continued) real 59 13 49.69 4.33 9 5.82 roof 41 20 19.5 5.39 17 4.48 rough 107 14 24.79 serve 267 13 3.85 shall 85 16 19.31 5.4 18 4.22 shape 83 19 18.2105 5.56 9 6.25 ship 63 16 40.38 5.04 14 5 shop 138 12 71.92 5.41 11 3.5 size 54 13 28.38 2.73 6 5.69 soil 240 5 22.2 2.74 7 3.39 south 103 12 19.33 6.35 16 6.14 teeth 55 8 45.375 5.01 22 3.32 theme 67 13 78.46 2.79 8 3.77 thick 333 11 72 20 3.46 thing 515 11 36.55 3.73 9 1.28 thought 42 16 31.13 13 4.09 vice 226 7 29 5.95 16 5.03 voice 75 15 29.6 6.69 21 3.85 vote 81 5 60.6 6.74 14 4.63 watch 228 15 72.93 3.96 8 5.8 wife 760 20 47 6.12 19 3.88 work 129 13 74.85 3.36 6 2.6 Hard words Frequency Neighborhood Density Neighborhood Frequency Connectivity Set Size Concreteness ban 1 26 299.19 4.3 20 5.61 bead 1 28 298.21 beak 5 25 396.36 19 6 bean 9 23 216.48 3.23 10 4.91 bud 4 26 190.58 6.38 16 6.4 bug 7 24 287.38 21 6.08 bum 1 28 410.5357 4.46 12 5.77 bun 5 22 743.36 chat 8 27 87.22 3.57 21 cheer 7 21 879.1 chore 6 27 91.7 3.39 11 6.13 cod 6 24 92.92 4.46 13 comb 9 21 104.8095 33 5.61 con 1 35 180.37 3.45 11 5.93 cot 7 23 117.52 18 2.38 dame 2 33 130.64 den 3 23 95.48 doom 1 27 120.33 dune 2 22 115.23 5.62 23 2.53 fade 3 30 825.33 5.03 12 goat 1 24 253.96 gut 3 30 438.8 hack 1 25 479.16 hag 1 22 544.36 hash 1 25 439.76 hick 6 25 711.44
93 Appendix D (continued) hid 9 23 146.09 5.09 13 hoot 5 25 232.52 4.6 11 hoot hum 3 22 199.18 hum hurl 2 33 796.27 5.37 12 hurl kin 2 35 281.86 19 kin kit 2 21 236.67 2.16 6 kit knob 7 29 92.34 4.18 24 knob lace 6 34 187.38 lace lad 2 28 89.29 lad lame 2 26 138.31 3.5 9 lame lice 1 29 175.14 lice mace 3 24 192.13 4.05 8 mace mall 8 30 636.03 3.32 12 mall mat 2 26 113.73 mat mid 1 33 321.18 mid mitt 1 26 233.96 mitt moan 1 31 161.39 4.98 15 moan moat 4 33 97.39 moat mole 1 23 144.74 mole mum 8 26 225.12 5.06 13 mum pad 35 39 444.72 4.65 19 pad pawn 2 21 366.95 3.05 11 pawn pet 8 30 96.63 4.21 11 pet pup 2 21 98.48 pup rat 6 37 480.27 4.56 19 rat rhyme 4 25 121.56 4.14 17 rhyme rim 5 26 129.31 5.04 20 rim rout 1 21 164.48 rout rum 3 29 256.28 6.1 10 rum rut 1 28 221.21 rut sane 8 33 90.33 sane sill 4 35 116.71 2.87 4 sill soak 7 23 108.91 7.41 16 soak suck 5 25 142.96 21 suck tan 9 25 379.28 3.49 13 4.18 teat 1 31 302.74 toot 3 27 1066.59 wad 1 22 163.14 wade 2 24 248.38 wail 3 32 153.78 wed 2 25 295.08 3.96 11 3.43 weed 1 24 287 5.02 14 5.96 whore 2 30 689.1 wick 4 26 432.69 2.26 3 5.45
94 Appendix E: Practice Words could pond ten frog mop dime beach ran gild train
95 Appendix F: Distracter Words more call take from band grass
96 Endnote Although it is recognized that the frequency norms by Kucer a and Francis (1967) are rather old, they are used in the proposed study for severa l reasons. First, studies which have considered wordfrequency use these norms (Bradlow & Pisoni, 1999; Imai et al., 2005; Lewellen et al., 1993; Nelson et al., 1998; Roodenrys et al ., 2002; Vitevich, 2002). In order to allow for comparison of results between t he proposed study and past studies which have considered word frequency, it was deemed best to use the same norms. Further, I know of no more recent word-frequency norms av ailable.
97 About the Author Astrid Doty received a BachelorÂ’s Degree in Humanities in 1993 and a M.L.A. in Humanities in 1995 from the University of South Florida. S he taught as a teaching assistant and later as an adjunct instructor at the Uni versity of South Florida in the Department of Humanities. She later went on to earn he r M.S. in Communications Sciences and Disorders in 2000 from the University of South Florida and worked as a speech-language pathologist in the public schools for severa l years before entering the interdisciplinary Ph.D. program in Psychology (Cognitive and Neural Sciences Program) and Communication Sciences and Disorders in 2001. While in the Ph.D. program at the University of South Fl orida, Mrs. Doty worked as a graduate teaching assistant and also taught Development al Psychology. She presented her research at several national and interna tional conferences such as the Acoustical Society of America and the American Speech Language and Hearing Association. Additionally, she worked as a speech-language pathologist in the public school system where she provided therapy for individuals aged three through 21.