Second Language Research Fo Illusory vowels in Spanish-English sequential bilinguals: Evidence that accurate L2 perception is neither necessary nor sufficient for accurate L2 production rP Journal: Second Language Research ee Manuscript ID SLR-18-0044.R3 Manuscript Type: Original Article rR phonotactic constraints, speech perception, speech production, Spanish, Keywords: English, L2 use, grammatical proficiency, age of L2 acquisition, executive control ev Spanish native speakers are known to pronounce onset /sC/ clusters in English with a prothetic vowel, as in esport for sport, due to their native language phonotactic constraints. We assessed whether accurate iew production of e.g. spi instead of espi, was related to accurate perceptual discrimination of this contrast in L2 speech of Spanish-English sequential bilinguals. A same-different discrimination task in stimulus pairs such as spi-espi assessed speech perception and a phonemic verbal fluency task elicited speech production. Logistic mixed model regressions revealed significant differences in accuracy between the bilinguals and the English monolinguals, although some bilinguals performed within the monolingual range. For the production task, but not for the perception task, bilinguals with more Abstract: exposure to English and greater grammatical knowledge of English performed significantly more accurately than those with less exposure and lower grammatical knowledge. There was no significant correlation between production accuracy and perception accuracy. Through examining phonotactic constraints, these results expand a growing body of research into single sounds which suggests dissociations between L2 perception and production. In contrast to predictions made by L2 speech models, the findings indicate that accurate L2 perception is neither necessary nor sufficient for accurate L2 production, and instead are interpreted to indicate that the two capacities recruit different executive control mechanisms and are acquired – at least to a certain extent - independently in L2 acquisition. http://mc.manuscriptcentral.com/SLR Second Language Research Page 2 of 64 1 2 3 Illusory vowels in Spanish-English sequential bilinguals: Evidence that accurate L2 4 5 perception is neither necessary nor sufficient for accurate L2 production 6 7 Esther de Leeuw1, Linnaea Stockall1, 8 Dimitra Lazaridou-Chatzigoga2, Celia Gorba Masip3 9 10 11 1Queen Mary University of London 12 13 2Humboldt-Universität zu Berlin and University of Cambridge 14 15 3Universitat 16 Autònoma de Barcelona 17 18 19 20 Acknowledgements 21 Fo 22 The authors are particularly grateful to Professor Juana Gil Fernández who welcomed 23 24 us to Madrid, where we collected most of the data at her Laboratorio de Fonética of rP 25 26 the University of Madrid, and to Professor Fernando Sánchez Miret at the University 27 28 of Salamanca. Without their help, we would never have been able to conduct this ee 29 study. We are also grateful to Annette Zhao for the data analysis, all of the 30 31 participants who took part in this study, and to the British Academy for funding the rR 32 33 project (BA grant SG112797: PI: EdeL; co-investigator: LS). 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 1 of 52 http://mc.manuscriptcentral.com/SLR Page 3 of 64 Second Language Research 1 2 3 Abstract 4 5 6 Spanish native speakers are known to pronounce onset /sC/ clusters in English with a 7 8 prothetic vowel, as in esport for sport, due to their native language phonotactic 9 10 constraints. We assessed whether accurate production of e.g. spi instead of espi, was 11 12 13 related to accurate perceptual discrimination of this contrast in L2 speech of Spanish- 14 15 English sequential bilinguals. A same-different discrimination task in stimulus pairs 16 17 such as spi-espi assessed speech perception and a phonemic verbal fluency task elicited 18 19 speech production. 20 21 Fo 22 Logistic mixed model regressions revealed significant differences in accuracy 23 24 between the bilinguals and the English monolinguals, although some bilinguals rP 25 26 performed within the monolingual range. For the production task, but not for the 27 28 ee 29 perception task, bilinguals with more exposure to English and greater grammatical 30 31 knowledge of English performed significantly more accurately than those with less rR 32 33 exposure and lower grammatical knowledge. There was no significant correlation 34 35 ev 36 between production accuracy and perception accuracy. 37 38 Through examining phonotactic constraints, these results expand a growing iew 39 40 body of research into single sounds which suggests dissociations between L2 41 42 perception and production. In contrast to predictions made by L2 speech models, the 43 44 45 findings indicate that accurate L2 perception is neither necessary nor sufficient for 46 47 accurate L2 production, and instead are interpreted to indicate that the two capacities 48 49 recruit different executive control mechanisms and are acquired – at least to a certain 50 51 52 extent - independently in L2 acquisition. 53 54 55 56 57 58 59 60 Page 2 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 4 of 64 1 2 3 Introduction 4 5 6 While there is some research into bilinguals’ perception and production of single 7 8 sounds, there is little work investigating L2 acquisition of phonotactic constraints, and 9 10 to our knowledge no work systematically investigating the relationship between 11 12 13 perception and production of phonotactic constraints. Our study examined phonotactic 14 15 constraints, i.e. language specific licit versus illicit sequences of sounds, in the speech 16 17 perception and production of Spanish-English sequential bilinguals. As mentioned by 18 19 Dupoux, Kakehi, Hirose, Pallier, & Mehler (1999), “in Spanish, /s/ + consonant clusters 20 21 Fo 22 are always preceded by a vowel and we have informally heard reports by Spanish 23 24 speakers who maintain that they hear the vowel [e] preceding English words that begin rP 25 26 with an /sC/ cluster” (p. 1568). Accordingly, Spanish learners of English often 27 28 ee 29 mispronounce initial /sC/ clusters with a prothetic initial vowel (Hualde, 2005). 30 31 The main question of this study was whether accurately perceiving the /sC/ rR 32 33 34 cluster in English (and therefore not perceiving an illusory preceding [e]) was 35 ev 36 37 associated with the accurate production of the /sC/ cluster (and therefore not 38 iew 39 pronouncing a preceding vowel in English speech where it is not required) within a 40 41 group of Spanish–English sequential bilinguals. Furthermore, it was examined whether 42 43 44 increased and earlier exposure to the L2, and improved L2 grammatical proficiency as 45 46 measured through a C-Test (Klein-Braley, 1985; Raatz and Klein-Braley, 1981), would 47 48 help to improve accurate perception and production of /sC/ clusters. Bilingual data were 49 50 51 collected in Spain (Madrid and Salamanca, henceforth referred to as the Spain 52 53 bilinguals), and in London, UK (henceforth referred to as the UK bilinguals). This study 54 55 therefore aimed to shed light on how the “rules” of language specific sound sequences 56 57 are implemented in L2 speech acquisition - when they violate rules in the L1 - and feeds 58 59 60 Page 3 of 52 http://mc.manuscriptcentral.com/SLR Page 5 of 64 Second Language Research 1 2 3 into a growing body of research which examines the relationship between L2 perception 4 5 6 and production. 7 8 L2 speech acquisition models 9 10 Often, when considering L2 speech acquisition, the premise is that problems accurately 11 12 13 producing L2 speech arise from difficulties in accurately perceiving L2 speech. It seems 14 15 logical that perception abilities would need to be in place before accurate production 16 17 abilities are possible, and that difficulties producing L2 speech would have a perceptual 18 19 basis, such that incorrect perception would lead to incorrect production (see e.g. 20 21 Fo 22 Escudero, 2005; and Llisterri, 1995, for an overview of early studies). For example, 23 24 according to the Speech Learning Model (SLM), it is claimed that those L2 sounds rP 25 26 which have a similar (although not identical) counterpart in the L1 are the most difficult 27 28 ee 29 to learn to pronounce because a similar sound is more likely to be perceived within the 30 31 same category as that of the L1 (Flege, 1995; Flege, Schirru, & MacKay, 2003). A new rR 32 33 category, on the other hand, is more likely to be created for a dissimilar sound 34 35 ev 36 (originally termed ‘new’ (Flege, 1987)), which is not perceived to be the same sound. 37 38 Long-term pronunciation problems are, according to the SLM, more likely in the case iew 39 40 of similar sounds than in dissimilar sounds as perceptual equivalence classification 41 42 prevents experienced L2 learners from producing similar, but not dissimilar sounds 43 44 45 (Flege, 1995). This notion of equivalence classification has strong parallels with Kuhl’s 46 47 Native Language Magnet Theory which suggests that the nearer an L2 sound is to an 48 49 L1 sound (specifically to its prototype), the more it will be perceptually assimilated to 50 51 52 this L1 sound (Kuhl, 2004). 53 54 The notion of perceptual similarity is also intrinsic to Best’s Perceptual 55 56 Assimilation Model (PAM) (1995) (see also PAM-2 with regard to L2 acquisition of 57 58 59 prosody, So and Best, 2010, 2014). According to this model, the similarity between L1 60 Page 4 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 6 of 64 1 2 3 and L2 sounds is based on the perceived resemblance of articulatory gestures used to 4 5 6 produce L2 sounds, in comparison to those used to produce the closest L1 sound. As 7 8 such, perception entails the ability to detect articulatory properties of speech (i.e., 9 10 tongue movement, vocal tract size, etc.), suggesting that speech perception and 11 12 13 production are aligned. This model has strong parallels with the Motor Theory of 14 15 speech perception, proposed by Liberman and Mattingly (1985), which claims that “the 16 17 objects of speech perception are the intended phonetic gestures of the speaker, 18 19 represented in the brain as invariant motor commands that call for movements of the 20 21 Fo 22 articulators through certain linguistically significant configurations” (p. 2) (see also 23 24 Fowler (1996) and Ohala (1996)). rP 25 26 What unifies both the SLM and PAM is that perception is considered to be 27 28 ee 29 intrinsically linked to production, and that either perception precedes production 30 31 (SLM), or that they develop in tandem (PAM). However, as will be discussed, there is rR 32 33 a growing body of research examining the perception and production of individual L2 34 35 ev 36 sounds which suggests dissociations between L2 perception and production. Moreover, 37 38 languages differ not only because of differences in their respective phoneme iew 39 40 inventories, but also in their realisation of phonotactic constraints. Indeed, very little is 41 42 known regarding the acquisition of new permissible sound sequences in the L2 which 43 44 45 violate L1 phonotactic constraints. The results from this study therefore inform both 46 47 PAM and the SLM regarding the assertion that difficulties in producing L2 speech are 48 49 linked to difficulties in perception. 50 51 52 53 54 Research into perception and production of L2 speech 55 56 Some previous research into the question of whether accurate perception leads to 57 58 59 accurate production in L2 acquisition suggests that perception and production are 60 Page 5 of 52 http://mc.manuscriptcentral.com/SLR Page 7 of 64 Second Language Research 1 2 3 intertwined. For example, it has been found that non-native English speakers who were 4 5 6 considered to be experienced L2 speakers (who lived in the US for approximately 7 7 8 years) produced and perceived English vowels more accurately than relatively 9 10 inexperienced non-native participants (who lived in the US for under 1 year) (Flege, 11 12 13 Bohn, & Jang, 1997). Both production and perception accuracy appeared to depend on 14 15 the perceived relation between English vowels and vowels in the participants’ specific 16 17 L1 inventory. Such findings provide support for the SLM in suggesting that perceptual 18 19 similarity between the L1 and L2 is linked to speech production in the L2. 20 21 Fo 22 However, other research has not substantiated a clear dependency between 23 24 perception and production of L2 speech, and it has been found that some learners are rP 25 26 able to produce differences between L2 sounds that they cannot perceive (Baker & 27 28 ee 29 Trofimovich, 2006; Beach, Burnham, & Kitamura, 2001; Kartushina & Frauenfelder, 30 31 2014; Sheldon & Strange, 1982; Zampini, 1998). For example, early results from rR 32 33 Sheldon and Strange (1985) revealed that native Japanese speakers who had learned to 34 35 ev 36 produce /l/ and /r/ appropriately still made perception errors in perceiving the contrast. 37 38 The authors summarised that “perceptual mastery of a foreign contrast does not iew 39 40 necessarily precede adult learners’ ability to produce acceptable tokens of the 41 42 contrasting phonemes, and may, in fact, sometimes lag behind production mastery” (p. 43 44 45 254). Similar results from Zampini (1998) into perception and production of voiced and 46 47 voiceless plosives in Spanish by English native speakers confirmed that L2 production 48 49 may in some cases precede perception and that learners do not begin to adjust 50 51 52 perceptual boundaries until they have attained accurate production categories. Likewise, 53 54 research examining the production of the three-way voicing difference in Thai bilabial 55 56 stops by Greek-English bilinguals and English monolinguals has indicated that those 57 58 59 bilinguals who “exaggerated” the voicing difference in their Thai speech production 60 Page 6 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 8 of 64 1 2 3 also best perceived those differences when listening (Beach et al., 2001). The study 4 5 6 concluded that production profiles are an important adjunct to the assessment of 7 8 bilingual speakers, and have important implications for the interface between 9 10 perception and production. 11 12 13 The current investigation expands on such findings by examining whether 14 15 accurately perceiving syllable onset /sC/ clusters in English (although such clusters are 16 17 not allowed in Spanish) would likewise help to avoid production of a prothetic vowel 18 19 before such clusters in Spanish-English sequential bilinguals, e.g. would those 20 21 Fo 22 bilinguals who perceive a difference between stimuli such as spi and espi be less likely 23 24 to produce a prothetic vowel in a phonemic verbal fluency task eliciting onset /sC/ rP 25 26 clusters, whilst bearing in mind individual differences in English L2 exposure, age of 27 28 ee 29 acquisition, and grammatical proficiency in English? 30 31 This question is particularly interesting when viewed in relation to executive rR 32 33 control, which has been linked to bilingualism and L2 acquisition (Bialystok, 2009, 34 35 ev 36 2017; de Leeuw and Bogulski, 2016). Executive control processes are hypothesised to 37 38 supervise “the selection, initiation, execution, and termination” of multiple task iew 39 40 performance (Rubinstein et al., 2001: 763). Frequently postulated executive 41 42 mechanisms underlying the overall system are the (1) shifting of mental sets, (2) 43 44 45 monitoring and updating of working memory representations, and (3) inhibiting 46 47 competing stimuli (Miyake et al., 2000). It may be that speech production recruits 48 49 different executive control functions than speech perception, as the former is inherently 50 51 52 more active than the latter, i.e. when producing speech, the articulators must be co- 53 54 ordinated to match the intended output, whilst when perceiving speech, there is no 55 56 output. Different underlying executive control processes may have influenced the 57 58 59 60 Page 7 of 52 http://mc.manuscriptcentral.com/SLR Page 9 of 64 Second Language Research 1 2 3 findings from the aforementioned studies, and are therefore also considered to be 4 5 6 potentially relevant in relation to the present research. 7 8 Research into perception of L2 phonotactic constraints 9 10 In seminal research on the perception of illusory vowels by Japanese native speakers, 11 12 it was found in four experiments comparing French and Japanese hearers that 13 14 15 phonotactic properties of Japanese induced native Japanese listeners to perceive 16 17 “illusory” vowels inside consonant clusters in VCCV stimuli, where there were in fact 18 19 no vowels, as Japanese does not allow such word medial consonant clusters (e.g. ebzo 20 21 Fo 22 was perceived as ebuzo) (Dupoux, Kakehi, Hirose, Pallier, & Mehler, 1999; see also 23 24 Harris, 1983; Itô and Mester, 1995). rP 25 26 In contrast, French native speakers had no difficulty perceiving this contrast 27 28 ee 29 (French allows such consonant clusters, e.g. ‘observer’ /obzɛrve/), However, the French 30 31 native speakers had difficulties discriminating items that differed in vowel length (e.g. rR 32 33 ebuzo vs. ebuuzo), while Japanese native speakers had no difficulty discriminating such 34 35 ev stimuli (vowel length is contrastive in Japanese but not in French). The conclusion from 36 37 38 this series of studies was that models of speech perception should be revised to account iew 39 40 for language specific phonotactic constraints. However, although the research found 41 42 significant differences between the Japanese and French native speakers, it was not 43 44 45 further investigated whether, for example, some of the Japanese native speakers might 46 47 have been able to acquire the French phonotactic constraints, nor whether accurate 48 49 perception of the French phonotactic constraints (i.e. licit consonant clusters) may have 50 51 52 been associated with accurate production of the consonant clusters. 53 54 In more recent research, several populations of Japanese-Brazilian bilinguals 55 56 were tested in order to determine whether the perceived illusory vowel differed between 57 58 bilingual groups (Parlato-Oliveira, Christophe, Hirose, & Dupoux, 2010). The results 59 60 Page 8 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 10 of 64 1 2 3 from an explicit metalinguistic task showed that, as expected based on the phonotactics 4 5 6 of each language, monolingual Japanese participants mainly perceived the stimuli as 7 8 containing the epenthetic vowel /u/ (e.g. in items like ebna and ebuna) while 9 10 monolingual Brazilian participants mainly perceived the epenthetic vowel /i/ (e.g. in 11 12 13 items like ebna and ebina). Interestingly, first-generation immigrants, who had 14 15 Brazilian Portuguese as an L2, appeared to behave like Japanese monolinguals, 16 17 suggesting that their immersion within a Brazilian-speaking country, which started in 18 19 adulthood, “did not induce them to significantly modify their phonological settings” (p. 20 21 Fo 22 3742). In sharp contrast, second-generation immigrants behaved like Brazilian 23 24 monolinguals and were more likely to perceive the epenthetic vowel /i/. Moreover, in rP 25 26 an implicit perceptual task, it was found that all three groups of bilinguals behaved 27 28 ee 29 exactly like Brazilian monolinguals, “showing high confusability between ebna and 30 31 ebina, while the distinction between ebna and ebuna proved to be easy to for them” (p. rR 32 33 3745). Parlato-Oliveira et al. (2010) summarised that an effect of L2 exposure was most 34 35 ev 36 clearly observed in the explicit task but not in the implicit task and suggested that L2 37 38 experience is more likely to affect explicit, or metalinguistic, perceptual tasks. iew 39 40 Another study, specifically examining Spanish–English sequential bilinguals, 41 42 similarly found differences between bilinguals with regard to their capacity to perceive 43 44 45 phonotactic constraints in the L2 (Carlson et al., 2016), noting that English not only 46 47 permits #sC, the prohibited sequence in Spanish, but it favours this sequence over 48 49 #VsC, the preferred Spanish repair. In both the identification and discrimination tasks, 50 51 52 the Spanish–English bilinguals exhibited weaker perceptual repair effects relative to 53 54 Spanish monolinguals, who came from Cuetos et al. (2011). This was true even for the 55 56 bilinguals who were considered to be dominant in Spanish; however, the perceptual 57 58 59 repair effects were least pronounced for English-dominant bilinguals, suggesting that 60 Page 9 of 52 http://mc.manuscriptcentral.com/SLR Page 11 of 64 Second Language Research 1 2 3 the acquisition of English influenced the perception of phonotactic constraints in 4 5 6 Spanish (Carlson et al., 2016). These results support related research which has 7 8 similarly found that /sC/ onset clusters in English activate Spanish phonotactic 9 10 constraints in Spanish–English bilinguals (Freeman et al., 2016) and most recent 11 12 13 research suggesting that even the perception of L1 Brazilian Portuguese phonotactic 14 15 constraints are modified by the acquisition of English as an L2 (Cabrelli et al., 2019). 16 17 However, in none of these studies was speech production investigated. Indeed, 18 19 as it is quite often accepted that perception provides the foundation for production, it 20 21 Fo 22 might be – potentially prematurely - interpreted that such results regarding perception 23 24 provide insight into what those same participants might produce in their L2 speech. rP 25 26 However, based on the previously discussed research which shows weak to not existent 27 28 ee 29 relationships between L2 speech perception and production (see also Kartushina & 30 31 Frauenfelder, 2014), we cannot assume that such studies into the perception of L2 sound rR 32 33 sequences provide direct insight into the capacity of L2 learners to produce those same 34 35 ev 36 L2 sound sequences. 37 38 In the present study, by focusing on a single phonotactic feature, we were able iew 39 40 to systematically examine the degree to which perception and production of the 41 42 prothetic vowel were associated with one another in Spanish native speakers with 43 44 45 English as an L2. The research by Hallé et al. (2008) confirms that the phenomenon of 46 47 an illusory prothetic vowel preceding sibilant consonant clusters in Spanish native 48 49 speakers is real, but the study does not actually employ a straightforward, simple, 50 51 52 auditory discrimination task along the lines of Dupoux et al. (1999), and, moreover, 53 54 Hallé et al.’s (2008) research does not compare perception and production in the same 55 56 experiment and same speakers, which is the aim of the current study. 57 58 59 60 Page 10 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 12 of 64 1 2 3 4 5 6 Methodology 7 8 The study comprised two main tasks. In Task I, the perception of a prothetic vowel in 9 10 minimal pairs was examined in the same two participant groups: English monolinguals, 11 12 13 and Spanish-English bilinguals. In Task II, the production of prothetic vowels 14 15 preceding /sC/ consonant clusters in onset position was assessed to determine whether 16 17 there was an association between perceptual and production accuracy. Additionally, 18 19 predictor variables were examined which might have played a role in determining 20 21 Fo 22 production and perception in the sequential Spanish–English bilinguals, i.e. amount of 23 24 daily exposure to L2, as self-assessed; age of L2 acquisition; and grammatical rP 25 26 proficiency in the L2, as measured by a written C-Test (see Appendix I for questions 27 28 ee 29 and Appendix II for information). Grammatical proficiency was thought to be an 30 31 interesting variable to examine because it may have been that bilinguals with higher rR 32 33 grammatical knowledge were also more aware of the phonotactic rules of English, 34 35 ev 36 which allow /sC/ clusters in onset position (see Kivistö-de Souza, 2015 and Morales 37 38 Pech and Izquierdo, 2011 for related research). This metalinguistic awareness may have iew 39 40 improved their English speech production, and, potentially also their speech perception, 41 42 if the two domains are related. 43 44 45 46 47 Participants 48 49 In total, 42 participants were examined. Eight participants were native English speakers, 50 51 52 whose data were collected in London, United Kingdom. These participants listed no 53 54 fluency in other languages, although some did have knowledge of other languages due 55 56 to school education and vacations abroad. Seven participants were native Spanish 57 58 59 speakers with English as an L2 who were recorded in London, United Kingdom. These 60 Page 11 of 52 http://mc.manuscriptcentral.com/SLR Page 13 of 64 Second Language Research 1 2 3 participants were either working in London, or were students on a year abroad, and had 4 5 6 lived in the UK for between 3 and 16 months. The remaining 27 participants were also 7 8 all native Spanish speakers whose data were collected in Spain. These participants 9 10 reported that they had learned English as an L2 in Spain, and used English to varying 11 12 13 extents in their daily lives. Many of the Spain bilinguals had also spent time in an 14 15 English speaking country with the average time being 7 months, and both the United 16 17 Kingdom and Spain bilinguals were therefore collapsed into one group. All participants 18 19 were between 40 and 18 years of age and most were either university students, 20 21 Fo 22 researchers, administrators, or academic faculty. All of the bilinguals considered 23 24 Spanish to be their native language, and English to be the language in which they had rP 25 26 the greatest proficiency after Spanish. The mean amount of daily English use of the 27 28 ee 29 bilinguals, which was self-assessed, was 30% with a standard deviation of 18%; the 30 31 mean grammatical proficiency score was 38% with a standard deviation of 16%; and rR 32 33 the mean age of English acquisition was 8 years of age with a standard deviation of 5.4 34 35 ev 36 years. 37 38 English Spanish-English iew 39 monolinguals bilinguals 40 Number of participants 8 34 41 Average age at time of experiment, 23.6 (6.3) 27.15 (11.9) 42 43 standard deviation in brackets 44 Females : males 4:4 24:10 45 Table 1: Participant background information, see Appendix II. 46 47 It is relevant to note that more data were collected initially than included in the analysis. 48 49 50 This is because a number of English monolinguals turned out to have high proficiencies 51 52 in additional languages; and a number of participants in Spain cited that they were 53 54 bilinguals in either Catalan or Galician, and sometimes that they did not consider 55 56 themselves to be native speakers of Spanish, which we had not expected when 57 58 59 60 Page 12 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 14 of 64 1 2 3 originally conducting the experiment. To ensure the reliability of our study, we 4 5 6 therefore excluded 19 participants from the analysis. 7 8 9 10 Data collection procedure 11 12 13 The same procedure was followed in both Madrid, Salamanca and London, and in each 14 15 setting the recordings were conducted in a sound attenuated room at respectively the 16 17 Laboratorio de Fonética of the Centro de Ciencias Humanas y Sociales, the University 18 19 of Salamanca, or the Phonetics Laboratory of Queen Mary University of London. 20 21 Fo 22 Recordings were conducted entirely in English in all cases in order to ensure an English 23 24 mode (Grosjean, 1998), i.e. this investigation tested L2 English perception and rP 25 26 production; however, particularly with regard to the perception task, as the tokens were 27 28 ee 29 non-words, the participants’ accuracy also provides insight into perceptual capacities 30 31 more generally (Carlson et al., 2016) and therefore the results also have ramifications rR 32 33 on the mechanisms of L1 perception (Cabrelli et al., 2019) and its plasticity in the 34 35 ev 36 context of bilingualism (de Leeuw and Celata, 2019). 37 38 Participants entered into the sound attenuated room where the outline of the iew 39 40 procedure was described to them, and then filled in the participant consent form. 41 42 Thereafter, the research investigator (either EdeL or DLC) filled in an adapted version 43 44 45 of the MPI Language Background Questionnaire (Gullberg and Indefrey, 2003), by 46 47 asking the participant the questions on the form (see Appendix I). In general, the 48 49 completion of this form took between 15 and 20 minutes. Participants then completed 50 51 52 the C-test, for which they were given a maximum of 10 minutes (see Appendix I), the 53 54 onset elicitation task (i.e. the phonemic verbal fluency task), and then the perceptual 55 56 discrimination task. Although the perceptual discrimination task was completed last to 57 58 59 ensure /sC/ clusters were not made salient to participants before the production task, 60 Page 13 of 52 http://mc.manuscriptcentral.com/SLR Page 15 of 64 Second Language Research 1 2 3 the perception task is presented initially in this manuscript, and the production task 4 5 6 thereafter. In total, data elicitation took no more than 45 minutes. 7 8 9 10 Task 1: Perception Task Elicitation 11 12 13 The purpose of this task was to determine the extent to which Spanish native speakers 14 15 with English as an L2 perceived a difference between stimuli such as spi and espi (see 16 17 Table 3 for the full set of stimuli). To assess this question, participants were invited to 18 19 take part in a same-different perceptual discrimination task (AX task). Participants wore 20 21 Fo 22 Sennheiser over ear headphones, and were presented with two non-words directly after 23 24 one another using DMDX software (Forster and Forster, 2003), which also recorded rP 25 26 participant responses, e.g. (1) spi and spu or (2) spi and espi or (3) spi and spi (see Table 27 28 ee 29 2). For ease of exposition, we will continue to use these tokens as the model stimuli. In 30 31 line with what we know about Spanish (Hallé et al., 2008; Hualde, 2005), we expected rR 32 33 the Spanish-English bilinguals to perceive a difference between the stimuli in (2) less 34 35 ev 36 accurately than English monolinguals, but we did not expect a difference between the 37 38 Spanish-English bilinguals and English monolinguals in (1) and (3) (see Table 2). We iew 39 40 did not expect to find a difference in Condition (3) because the items presented in the 41 42 AX discrimination task were the same in these trials, and we did not expect a difference 43 44 45 in Condition (1) because /i/ and /u/ are contrastive in both English (e.g. beat versus boot) 46 47 and Spanish (e.g. sí versus su). However, as English favours /sC/ onset clusters over 48 49 /VsC/ onsets (Carlson et al., 2016), we also expected that English monolinguals might 50 51 52 have difficulties in Condition 2, albeit less than the Spanish native speakers. There were 53 54 approximately twice as many stimuli pairs for Condition 3 as for the other two 55 56 conditions in order to ensure an equal number of same stimuli pairs (Condition 3) versus 57 58 59 different stimuli pairs (Conditions 1 & 2), see Table 2, and 197 trials in total. 60 Page 14 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 16 of 64 1 2 3 4 5 6 Contrast example Nr. of trials Prediction 7 Condition 1 spi-spu 49 No difference between Spanish– 8 English bilinguals and English 9 monolinguals 10 Condition 2 spi-espi 49 Spanish–English bilinguals will 11 12 exhibit less accuracy than English 13 monolinguals 14 Condition 3 spi-spi 99 No difference between Spanish– 15 English bilinguals and English 16 monolinguals 17 Table 2: Conditions in perception task and general predictions 18 19 20 21 Stimuli Fo 22 23 24 Stimuli were produced by the first author, who is a native speaker of English. The rP 25 26 author produced all of the espi tokens first, listed in column b in Table 3. 27 28 ee 29 30 31 a. Edited token with no b. Original token with rR 32 prothetic vowel prothetic vowel 33 34 spi espi 35 spu espu ev 36 sli esli 37 slu eslu 38 iew ski eski 39 40 sku esku 41 smi esmi 42 smu esmu 43 sni esni 44 snu esnu 45 46 sti esti 47 stu estu 48 Table 3: Tokens used in same-different discrimination task. 49 50 Thereafter, the initial vowel was edited from the stimulus, such that a new stimulus was 51 52 created: spi. These consonant clusters (i.e. /sp/, /st/, /sk/, /sm/, /sn/, /sl/) were chosen 53 54 55 because they are the possible /sC/ clusters in English, and therefore also the same onsets 56 57 which were elicited in the phonemic verbal fluency task. 58 59 60 Page 15 of 52 http://mc.manuscriptcentral.com/SLR Page 17 of 64 Second Language Research 1 2 3 Delimiting the initial vowel from the following sibilant was not difficult based 4 5 6 on analysis of the spectrogram and waveform. The boundary was marked where the 7 8 periodicity of the vowel ended and the aperiodicity of the fricative commenced. 9 10 Therefore, these stimuli (a. spi and b. espi) were exactly the same except for the vowel 11 12 13 in b (see Figure 1 and 2 of /spi/ and /espi/). Note that it may have been that formant 14 15 remnants from the initial vowel remained in the sibilant due to coarticulation effects, 16 17 although all actual periodicity from the vowel was cut out (see Fig. 1 & 2), which might 18 19 potentially have led to more perceived instances of /e/ on the part of the bilinguals; 20 21 Fo 22 however, all bilinguals heard the same items, and therefore this would have affected 23 24 their discrimination of spi versus espi equally. rP 25 26 27 28 ee 29 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 Figure 1: Token of spi used in perception task. 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 16 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 18 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 Figure 2: Token of espi used in perception task. rP 25 26 27 28 As this study examined the L2 acquisition of English, the initial vowel in b was ee 29 30 the open-mid front unrounded vowel /ɛ/, which is present in the first syllable of English 31 rR 32 33 34 words like escape and estate. This vowel is also similar to Spanish /e/, and in some 35 ev 36 transcriptions has been considered to be an allophone of this phoneme (Tomás Navarro, 37 38 1918), although other transcriptions refute this (Martínez-Celdrán et al., 2003). This iew 39 40 41 vowel was considered to be the most appropriate for the purpose of this study, i.e. in 42 43 contrast to schwa, because (1) it is present in English words preceding /sC/ clusters, 44 45 and hence the bilinguals would have been exposed to this sound sequence in their 46 47 acquisition of English, and because (2) it is similar to Spanish /e/, and hence more likely 48 49 50 to be identifiable on the part of the bilinguals than English schwa would have been. A 51 52 final reason rationalising the choice of /ɛ/ was that (3) it was more naturalistic for the 53 54 55 speaker to produce, as this vowel indeed occurs in English words, and therefore the 56 57 58 recordings were likewise more likely to also sound naturalistic for the listeners. In 59 60 essence, had we chosen schwa as the initial vowel, it may have been more likely that Page 17 of 52 http://mc.manuscriptcentral.com/SLR Page 19 of 64 Second Language Research 1 2 3 the quality of the vowel impeded discrimination, rather than the presence of the vowel 4 5 6 itself. Sound files are online for interested readers. 7 8 The duration of the vowel approximated 80ms to 85ms, and was therefore 9 10 considered to be controlled yet naturalistic. As in Table 2, only high vowels followed 11 12 13 the /sC/ cluster as in stressed position these English vowels were thought to be more 14 15 similar to the Spanish high vowels than English low vowels to Spanish low vowels. For 16 17 example, in Spanish there is just one low vowel, /a/, whilst in English there are 18 19 20 numerous, e.g. depending on dialect /a/, /ɑ/ and /ɒ/ (Hualde, 2005). As such, the task 21 Fo 22 was considered to focus on the discrimination of the vowel prothesis in the original 23 24 token, and no additional challenges were present which might have additionally rP 25 26 27 encumbered the perception of the stimuli. Note that the speaker produced 28 ee 29 monophthongal realisations of the vowels in the espi items with low second formant 30 31 frequencies (see Fig. 3 & 4), i.e. no GOOSE-fronting, which has been reported in new rR 32 33 34 varieties of British English (Cheshire et al., 2011; Mills, 2014). 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 Figure 3: Token of spu used in perception task. 58 59 60 Page 18 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 20 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 Figure 4: Token of espu used in perception task. rP 25 26 27 28 Procedure ee 29 30 The experiment comprised three conditions, which were presented randomly rather 31 rR 32 than in blocks. In each condition participants had to distinguish between the pairs of 33 34 stimuli presented immediately after the other in conditions (1), (2) and (3). Participants 35 ev 36 37 were not specifically told what language the items would be in, but all of the 38 iew 39 instructions for this specific task were in English, and the entire experimental procedure 40 41 was in English, as well as the questionnaire, so participants were considered to have 42 43 44 been in an English mode (Grosjean, 1998). As such, it was thought that they would be 45 46 more likely to discriminate spi-espi pairs than had the experiment been conducted 47 48 entirely in Spanish. 49 50 51 Condition 1 contained stimulus pairs like spi-spu, which we expected to be 52 53 relatively easy for both Spanish native speakers and English native speakers. Condition 54 55 2 was the target condition in which participants had to distinguish between spi-espi 56 57 pairs. This condition was predicted to be more difficult for the Spanish native speakers 58 59 60 with English as an L2 than for the English monolinguals. In Condition 3, participants Page 19 of 52 http://mc.manuscriptcentral.com/SLR Page 21 of 64 Second Language Research 1 2 3 had to determine whether spi-spi type pairs were the same or different, and this, like 4 5 6 Condition 1, was expected to be relatively easy for both the Spanish-English bilinguals 7 8 and the English monolinguals. However, because all three conditions were presented 9 10 randomly to the participants, the task was on the whole expected to be more difficult 11 12 13 for the bilinguals than for the monolinguals, such that the Spanish-English bilinguals 14 15 may also have made more errors, or have had slower response times in Conditions 1 16 17 and 3. 18 19 Participants indicated their response using a handheld video game controller by 20 21 Fo 22 pressing the left button for same and the right button for different. Participants were 23 24 given 2500ms to complete their decision, and then the next stimulus pair was presented rP 25 26 automatically. A 2500ms duration was allowed because it worked well in the pilot 27 28 ee 29 project as a duration that was long enough to allow responses from the full range of 30 31 participants, but fast enough to keep the pace of the experiment proceeding without rR 32 33 long lags, and to force a rapid, automatic response (rather than a more reflective slow 34 35 ev 36 response). Note that Dupoux et al. 1999 exp3 used a 4 sec cut-off, but we found that 37 38 we lost less than 3% of trials with a 2500ms cut-off in the pilot (in the current iew 39 40 experiment, we lost 4.5% of all trials due to no response within the time window). 41 42 There was a short practice block, consisting of 6 trials: 3 different, 3 same. 43 44 45 Participants were given feedback after the practice block, and could ask questions if the 46 47 instructions were not clear, although this never happened. The timing ‘clock’ started 48 49 with the onset of the second wav file, so participants could press a button to indicate 50 51 52 ‘same’ or ‘different’ as soon as the second file began. There were only 2 trials where a 53 54 response was given before 300ms passed across the entire data set, so we included all 55 56 trials in the analysis. There was no beep between stimuli; participants just heard the two 57 58 59 60 Page 20 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 22 of 64 1 2 3 sound files and then silence until they pressed the response button or 2500ms elapsed 4 5 6 with no response (following Dupoux et al. 1999). 7 8 In total, the perception task lasted 12 minutes, which included two short breaks 9 10 equally dispersed. Including the explanation of the perceptual discrimination task, the 11 12 13 entire procedure took 15 minutes. 14 15 16 17 Task II: Prothetic Vowel Production Elicitation 18 19 For the prothetic vowel production task, participants took part in a phonemic verbal 20 21 Fo 22 fluency task. In the standard version of this task, participants are asked to produce as 23 24 many unique words as possible which start with a given letter within a given timeframe rP 25 26 (Newcombe, 1969). The vast majority of studies that use this task use single letters as 27 28 ee 29 cues, but there is at least one study that used double-letter cues (e.g. “fa”, “sm”) when 30 31 comparing Spanish-English bilinguals to English monolinguals (Sandoval et al., 2010). rR 32 33 In the standard test, the participant’s score is the number of unique words produced for 34 35 ev 36 each given letter, but for our purposes, the accuracy score reflected accurate /sC/ 37 38 pronunciation (i.e. no prothetic vowel) for each unique word. iew 39 40 The phonemic verbal fluency task, also called the letter fluency task, is a 41 42 standard task used in both normative and non-normative language assessments of native 43 44 45 and non-native languages, often implemented together with the semantic verbal fluency 46 47 task for which word categories are listed, e.g. words like cat and dog would be listed if 48 49 “animal” were given as a category (Grogan et al., 2009; Newcombe, 1969). 50 51 52 Interestingly, some research has shown that monolinguals and fluent bilinguals (in their 53 54 L2) score similarly on the phonemic verbal fluency task, i.e. they are able to list 55 56 approximately the same amount of words, but that monolinguals obtain higher scores 57 58 59 on the semantic verbal fluency task (Gollan et al., 2002; Portocarrero et al., 2007, but 60 Page 21 of 52 http://mc.manuscriptcentral.com/SLR Page 23 of 64 Second Language Research 1 2 3 see also Bialystok et al., 2008). For the purposes of our study, the bilinguals were 4 5 6 considered to be capable of the phonemic verbal fluency task, but that the task would 7 8 also be slightly more demanding than e.g. a word list reading task, for which we thought 9 10 that the full orthographic representation of the word would enable them to focus on 11 12 13 their pronunciation, and in this way not necessarily reflect what they would produce in 14 15 normal conversation. 16 17 In the instructions for the phonemic verbal fluency task, it was stated that they 18 19 would see one or two letters, like ‘b’ or ‘gr’ and should name as many English words 20 21 Fo 22 as possible which begin with the sounds which those letters spell. An example of ‘pr’ 23 24 was given, for which one should say ‘print’ but not ‘pint’ “because ‘pint’ does not begin rP 25 26 with ‘pr’. Likewise, if they saw ‘sk’, it was instructed that they could say ‘scream’ or 27 28 ee 29 ‘sky’ since “they both begin with the ‘sk’ sounds”, but that ‘scene’ would be incorrect. 30 31 Participants were given 10 seconds for each onset, and they were asked to only name rR 32 33 unrelated words, e.g. they should avoid naming both “print” and “printed”. A short 34 35 ev 36 amount of time of 10 seconds was given for each onset so that participants were not 37 38 able to reflect too much on their pronunciation. In total, 22 onsets were elicited, but iew 39 40 only the /sC/ clusters were of interest to the study at hand: /sp/, /st/, /sk/, /sl/, /sn/ and 41 42 /sm/. As there were six clusters, participants were given in total 1 minute to produce 43 44 45 the /sC/ clusters, which is the standard total time in a phonemic verbal fluency task 46 47 (Newcombe, 1969). 48 49 The phonemic verbal fluency task is often used to assess executive control 50 51 52 abilities, as participants need to retrieve words, which requires them to access their 53 54 mental lexicon, whilst focusing on the task, selecting words meeting certain constraints 55 56 and inhibiting repetition (Miyake et al., 2000). However, the amount of words produced 57 58 59 is considered to be dependent on not only executive control abilities, but also on e.g. 60 Page 22 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 24 of 64 1 2 3 vocabulary size (Sauzéon et al., 2011). As such, this task was considered to be a more 4 5 6 demanding task than other potential production elicitation techniques, and more 7 8 representative of normal everyday speech production, which would likewise require 9 10 both monolinguals and bilinguals to access their mental lexicon whilst monitoring 11 12 13 pronunciation. 14 15 Moreover, the task was also considered to be potentially more cognitively 16 17 demanding and more reflective of “real world” language tasks than the previously 18 19 described perception task. In this way, it could be argued that the tasks were not 20 21 Fo 22 balanced, but we thought that it was interesting to investigate whether the “raw capacity” 23 24 to discriminate spi-espi in the perception task, unimpeded by other cognitive tasks, rP 25 26 would be shown to facilitate accurate production of /sC/ clusters in English. A crucial 27 28 ee 29 question was therefore whether bilinguals who scored higher in the production task 30 31 would likewise also necessarily score higher in the perception task, if accurate rR 32 33 perception is indeed a requirement for accurate production. 34 35 ev 36 37 38 Acoustic analysis iew 39 40 In order to determine whether a prothetic vowel was present in the Spanish-English 41 42 bilinguals’ production of word-initial consonant clusters beginning with /sC/ in English, 43 44 45 an acoustic analysis was conducted on recordings obtained from all participants taking 46 47 part in the production task. 48 49 Using Praat (Boersma and Weenink, 2010), a textgrid file was created which 50 51 52 included interval tiers (respectively, “word”, “vduration”) and one point tier (“notes”). 53 54 In the top “word” tier, real words which were produced by participants were input 55 56 directly while non-words were spelt out according to participants’ actual pronunciations 57 58 and were marked with an “X” at the end of the created word. The tier called “vduration” 59 60 Page 23 of 52 http://mc.manuscriptcentral.com/SLR Page 25 of 64 Second Language Research 1 2 3 was used to annotate the prothetic vowel, if there was one at all, and the point tier was 4 5 6 used to make comments on individual items, when considered necessary. To delimit 7 8 the words, the cursor to indicate the start of the word was placed at the closest 0- 9 10 crossing in the waveform where either phonation began from the prothetic vowel, or 11 12 13 where aperiodic noise began from the sibilant. The cursor to indicate the end of the 14 15 word was placed at the closest 0-crossing where there was considered to be a significant 16 17 drop in waveform amplitude. 18 19 Both the spectrogram and the waveform were used to determine the onset and 20 21 Fo 22 offset of the prothetic vowels preceding the onset of the sibilant. Where necessary, 23 24 Praat’s recognition of a pitch contour (i.e. lack of recognition for the voiceless sibilant) rP 25 26 was also considered as an indicator of voicing in the case of the potential vowel. In 27 28 ee 29 most cases, the onset and offset of the potential prothetic vowel were located at the 30 31 beginning and the end of periodicity where the waveform crossed the 0-axis (some rR 32 33 prothetic vowels started with a glottal stop, or some other form of plosion – and this 34 35 ev 36 was also counted as part of the vowel). The presence of formants in the spectrogram 37 38 was also used as an indicator of vowel prothesis. However, in some cases, the prothetic iew 39 40 vowel was produced with creaky voice. In such cases, auditory perception of an initial 41 42 vowel was the main cue which was used to determine whether there was a prothetic 43 44 45 vowel. The offset of the vowel was determined by the onset of the following sibilant. 46 47 Determination of the sibilant was relatively uncontroversial; marked by the onset of a 48 49 high intensity band of aperiodic frication in the waveform and spectrogram. As was the 50 51 52 case for the onset of the potential prothetic vowel, this transition from vowel to sibilant 53 54 was marked at the nearest 0-crossing in the waveform. The duration of the prothetic 55 56 vowel was then measured in milliseconds (ms), and data regarding duration can be 57 58 59 found in the appendix; however, here due to space constraints, we only present binary 60 Page 24 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 26 of 64 1 2 3 data of both the perception and production results (findings from continuous durational 4 5 6 analyses confirmed binary data). 7 8 9 10 Statistical analysis 11 12 For both the perception and production results, data were organised in CSV files using 13 14 15 Excel software. Thereafter, R software (R Core Team, 2017) was used for the 16 17 analyses and a series of binomial mixed-effects regression models were built for the 18 19 accuracy results of the perception and production tasks and a series of linear 20 21 Fo 22 regression models were built for the reaction time results of the perception task (using 23 24 the lme4 package in R; Bates, Maechler, Bolker, & Walker, 2015) for the examination rP 25 26 of the influence of fixed and random factors on the response. For the analysis of the 27 28 ee 29 perception data, timed out responses (i.e. trials which were not discriminated within 30 31 2500ms) were excluded resulting in the exclusion of 371 trials out of 8274 trials. For rR 32 33 the analysis of the production data, 616 individual word tokens were analysed. We 34 35 ev present the perception results initially, then the production results, then the results 36 37 38 regarding perception as a predictor of speech production. iew 39 40 41 42 Results 43 44 45 46 47 Task I: Perception 48 49 50 51 52 The accuracy and reaction time results of the perception experiment are summarised 53 54 per group and condition in Table 4 and the binary accuracy results are visualised in 55 56 Figure 5, revealing that Spanish–English bilinguals performed least accurately in 57 58 59 60 Page 25 of 52 http://mc.manuscriptcentral.com/SLR Page 27 of 64 Second Language Research 1 2 3 condition 2 compared to conditions 1 and 3, and less accurately than the English 4 5 6 monolinguals. 7 8 English monolinguals Spanish-English bilinguals 9 10 Percent of Response Percent of Response 11 12 correct times for correct times for 13 responses correct responses correct 14 responses responses 15 Condition 1: spi- 96.05 1175.4 95.17 1193.48 16 spu (19.5) (217.12) (21.45) (252.26) 17 Condition 2: spi- 87.4 1240.89 62.29 1311.41 18 espi (33.23) (302.94) (48.48) (384.38) 19 20 Condition 3: spi- 97.36 1197.19 93.11 1228.09 21 spi (16.04) (308.67) (25.33) (341.35) Fo 22 Table 4: Percentage of correct responses and response times for each group 23 (standard deviations in brackets). 24 rP 25 26 27 28 ee 29 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 Figure 5: Perception accuracy in English monolinguals versus Spanish-English 53 bilinguals. 54 55 For the data analysis of the accuracy results, the modelled response was accuracy (i.e. 56 57 58 yes or no, whether the participant was accurate in the AX discrimination or not), and 59 60 the fixed factors were language background (L1, English vs. Spanish) and condition Page 26 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 28 of 64 1 2 3 (C1: spi-spu, C2: spi-espi, C3: spi-spi). Random intercept was consonant, i.e. second 4 5 6 consonant in the sC cluster (e.g. /p/ in /sp/) (participant was initially included but this 7 8 model failed to converge). Models were manually stepped-down (using likelihood 9 10 ratio tests) from maximal models containing all factors and possible interactions to the 11 12 13 ‘best’ model that only contained significant predictors or predictors that participated 14 15 in significant interactions (Barr et al., 2013). The best fitting model determined by 16 17 step-down pairwise model comparison was acc ~ L1 * condition + (1 | consonant). 18 19 The model parameters are in Table 5. 20 21 Fo 22 Estimate Std. Error z value Pr (>|z|) 23 24 (Intercept=L1=English, 3.1955 0.2689 11.884 < 2e-16 *** rP 25 26 Cond=1) 27 28 ee 29 L1=Spanish, Cond=1 -0.2099 0.2888 -0.727 0.46741 30 31 L1=English, Cond=2 -1.2577 0.3060 -4.111 3.95e-05 *** rR 32 33 L1=English, Cond=3 0.4174 0.3484 1.198 0.23084 34 35 ev 36 L1=Span x Cond=2 -1.2300 0.3318 -3.707 0.00021 *** 37 38 L1=Span x Cond=3 0.7967 0.3742 -2.129 0.03325 * iew 39 40 41 Table 5: Estimates, standard errors, z values and p values of the best fitting 42 model for the perception experiment. 43 44 45 As displayed in Table 5, the main finding was that, as predicted, the magnitude of the 46 47 difference in the Spanish speakers between condition 1 and condition 2 was greater 48 49 50 than the difference in the English speakers between condition 1 and condition 2 (β = - 51 52 1.2300, z =-3.707, p < 0.001), although responses to condition 2 were also less 53 54 accurate than for condition 1 for English speakers (β = -1.2577, z = -4.111, p < 55 56 57 0.0001). Furthermore, the magnitude of the difference in the Spanish speakers 58 59 between condition 1 and condition 3 was also greater than the difference in the 60 Page 27 of 52 http://mc.manuscriptcentral.com/SLR Page 29 of 64 Second Language Research 1 2 3 English speakers between condition 1 and condition 3 (β = 0.7967, z = -2.129, p < 4 5 6 0.05). No other comparisons were significant. 7 8 In a second analysis, we investigated possible differences among the Spanish– 9 10 English bilinguals in their responses to the critical condition 2 (e.g. spi-espi) using the 11 12 13 predictor variables of percentage of English daily use, age of English acquisition, and 14 15 C-Test proficiency score, i.e. grammatical score, which were scaled using the scale 16 17 function in R. However, none of these predictors proved to be significant (ps > 0.05). 18 19 For the reaction time results, in our first analysis, the modelled response was 20 21 Fo 22 log-transformed RT and the fixed factors were language background (L1, English vs. 23 24 Spanish) and condition (C1: spi-spu, C2: spi-espi, C3: spi-spi). Random intercepts rP 25 26 were participant and again consonant. None of the predictors or their interaction 27 28 ee 29 proved to be significant (ps > 0.05). Nevertheless, the numerical RT pattern was 30 31 consistent with the accuracy results pattern with the Spanish–English bilinguals rR 32 33 having longer RTs in condition 2 (e.g. spi-espi). 34 35 ev 36 In a second analysis, we investigated possible differences among the Spanish– 37 38 English bilinguals in their responses to the critical condition 2 (e.g. spi-espi) using the iew 39 40 predictor variables of percentage of English daily use, age of English acquisition, and 41 42 C-Test proficiency score, which were scaled using the scale function in R; however, 43 44 45 none of the predictors proved to be significant (ps > 0.05). 46 47 48 49 Task II: Production 50 51 52 53 54 Table 6 summarises the rates of prothesis, for each of the six cues of interest, for the 55 56 two participant groups and Figure 6 visualises the results. 57 58 59 60 Page 28 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 30 of 64 1 2 3 Spanish–English bilinguals 4 5 6 Cluster English monolinguals (%) (%) 7 8 sk 8.70 (28.8) 55.56 (50.0) 9 10 sl 11.1 (32.3) 55.13 (50.1) 11 12 13 sm 0.00 (0.0) 57.53 (49.8) 14 15 sn 6.67 (25.8) 52.86 (50.3) 16 17 sp 4.35 (20.9) 62.63 (48.63) 18 19 20 st 8.33 (28.2) 67.37 (47.1) 21 Fo 22 Table 6: Percentage of words produced with prothesis for each sC cue, by 23 participant group (standard deviations in brackets). 24 rP 25 26 27 28 ee 29 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 Figure 6: Production accuracy in English monolinguals versus Spanish-English 46 bilinguals. 47 48 49 50 51 For the production experiment, the modelled response was whether a prothetic vowel 52 53 was produced or not (i.e. prothesis), for each word produced, and the fixed factor was 54 55 language background (L1 English vs. L1 Spanish). Random intercepts included 56 57 58 participant and consonant cluster (i.e. /st/, /sp/, /sk/, /sm/, /sn/ and /sl/). The maximal 59 60 Page 29 of 52 http://mc.manuscriptcentral.com/SLR Page 31 of 64 Second Language Research 1 2 3 model motivated by our hypotheses was prothesis ~ L1 + (1 | participant) + (1 | 4 5 6 consonant.cluster). 7 8 Estimate Std. Error z value Pr(>|z|) 9 10 (Intercept) (L1 = 11 12 13 English) -3.7248 0.8353 -4.459 8.23E-06 *** 14 15 L1 = Spanish 4.2509 0.886 4.798 1.61E-06 *** 16 17 18 Table 7: Estimates, standard errors, z values and p values of the best fitting 19 20 model for the production experiment. 21 Fo 22 As displayed, for the production analysis, the main finding was that as predicted, there 23 24 was a significant difference between the English monolinguals and the Spanish- rP 25 26 27 English bilinguals (β = 4.2509, z =4.798, p < 0.00001) with the bilinguals producing 28 ee 29 significantly more prothetic vowels than the monolinguals. To exemplify this 30 31 difference, individual words (i.e. speak, slain, smile) produced by the Spanish– rR 32 33 English bilinguals are displayed in figures 7-9. 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 30 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 32 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 30 31 Figure 7: The word “speak” produced by a Spanish-English bilingual with rR 32 prothesis. 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 31 of 52 http://mc.manuscriptcentral.com/SLR Page 33 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 Figure 8: The word “slain” produced by a Spanish-English bilingual with 30 31 prothesis. rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 32 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 34 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 30 Figure 9: The word “smile” produced by a Spanish-English bilingual with 31 prothesis. rR 32 33 34 35 ev 36 Perception as a predictor for production 37 38 For the next analysis stage, we calculated a mean perception accuracy score on the iew 39 40 critical Condition 2 contrast (e.g. spi-espi) in the perception experiment for each 41 42 43 bilingual speaker. This score was included as a possible predictor in this second 44 45 analysis of the production results, along with percentage of English daily use, age of 46 47 English acquisition, and grammatical proficiency score (C-test result) in order to 48 49 examine whether these factors would be influential in determining accuracy of 50 51 52 pronunciation (i.e. either producing the prothetic vowel in the phonemic verbal 53 54 fluency task or not). The maximal model tested was: prothesis ~ Percent of English in 55 56 Daily Use + Age of Acquisition + English Proficiency + 57 58 59 60 Page 33 of 52 http://mc.manuscriptcentral.com/SLR Page 35 of 64 Second Language Research 1 2 3 Condition2PerceptionAccuracy + (1 | participant) + (1 | consonant.cluster)1. As 4 5 6 before, all continuous predictor variables were scaled using the scale function in R. 7 8 Models were manually stepped-down (using likelihood ratio tests) from maximal 9 10 models containing all factors to the best fitting model that only contained significant 11 12 13 predictors (Barr et al., 2013). The best fitting model determined by step-down 14 15 pairwise model comparison was prothesis ~ Percent of English in Daily Use + English 16 17 Proficiency + (1 | participant) + (1 | consonant.cluster). Neither age of English 18 19 20 acquisition nor accuracy rate on condition 2 of the perception task was a significant 21 Fo 22 predictor in the maximal model, and removing those terms from the maximal model 23 24 did not decrease the model fit (as assessed by likelihood ratio tests) (all ps > 0.05). rP 25 26 27 The parameters of the winning model are displayed in Table 8. As shown, both 28 ee 29 percentage of daily English use (β = -0.5454, z =-2.523, p < 0.05) and English 30 31 proficiency (β = -0.6648, z =-3.057, p < 0.01) significantly influenced the amount of rR 32 33 prothesis produced by the Spanish-English bilinguals. N. b. an analysis of duration of 34 35 ev 36 prothesis revealed similar overall findings. 37 38 Estimate Std. Error z value Pr(>|z|) iew 39 (Intercept) 40 41 Prothesis 0.03271 0.41492 0.079 0.93716 42 Daily Eng Use -0.5454 0.2162 -2.523 0.01165* 43 Eng 44 Proficiency -0.6648 0.2174 -3.057 0.00223** 45 46 47 Table 8: Estimates, standard errors, z values and p values of the best fitting model 48 for prothesis produced by the Spanish–English bilinguals. 49 50 51 52 Figures 10 and 11 plot the relationships between the two significant predictor 53 54 variables (respectively English daily use and English grammatical proficiency) and 55 56 57 58 59 1Interaction terms were not included in the maximal model as there were not enough observations to 60 allow for so many parameters. Page 34 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 36 of 64 1 2 3 the proportion of words produced with prothesis for each speaker, with the grey 4 5 6 ribbon representing 95% confidence interval. As displayed in Figure 10, a higher 7 8 amount of English use was correlated with less prothetic vowel productions, i.e. a 9 10 more English-like pronunciation (t = -2.1289, df = 32, p < 0.05, r=-0.35). As 11 12 13 displayed in Figure 11, a higher grammatical proficiency in English was correlated 14 15 with less prothetic vowel productions, i.e. a more English-like pronunciation (t = - 16 17 2.5228, df = 32, p = 0.05, r=-0.41). 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 Figure 10: Production of prothetic vowel over daily English use (%). 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 35 of 52 http://mc.manuscriptcentral.com/SLR Page 37 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 Figure 11: Production of prothetic vowel over grammatical proficiency C-test rP 25 26 result (%). 27 28 ee 29 Figure 12 plots the non-significant relationship between perception accuracy and 30 31 rR 32 prothesis proportion (t=1.4906, df = 32, p=0.1458). If the four quarters of the scatterplot 33 34 are inspected, it is possible to see that the bottom right-hand quarter is almost empty, 35 ev 36 indicating that hardly anyone displayed accurate production and perception. 37 38 iew 39 Alternatively, in the top left-hand corner, many bilinguals displayed perception 40 41 accuracy rates below chance level (i.e. beneath 50%), as well as high prothetic vowel 42 43 rates. In the top right-hand quarter, bilinguals are visible who showed perception 44 45 46 accuracy rates above chance, combined with high prothetic vowel rates, indicating that 47 48 accurate perception was not sufficient for accurate production. In the bottom left-hand 49 50 corner, bilinguals are visible who showed perception accuracy rates below chance, 51 52 combined with low prothetic vowel rates, indicating that perception was not necessary 53 54 55 for accurate production. 56 57 58 59 60 Page 36 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 38 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 30 31 Figure 12: Production over perception of prothetic vowel (non-significant rR 32 relationship). 33 34 35 ev 36 Discussion of findings 37 38 In this study, speech perception was assessed through a same-different discrimination iew 39 40 task in stimulus pairs such as spi-espi and speech production of vowel prothesis was 41 42 43 assessed through a phonemic verbal fluency task in Spanish-English sequential 44 45 bilinguals. The primary objective was to examine whether accurately perceiving the 46 47 prothetic vowel preceding /sC/ clusters would likewise help to not produce this 48 49 50 prothetic vowel in the L2. In general, the findings revealed that there was a significant 51 52 difference between the monolingual English speakers and the Spanish-English 53 54 bilinguals, and that there was no clear relationship between the perception and 55 56 57 production of /sC/ clusters in onset position (which are restricted by Spanish 58 59 phonotactic rules) by the Spanish-English bilinguals. Results indicated that the 60 Page 37 of 52 http://mc.manuscriptcentral.com/SLR Page 39 of 64 Second Language Research 1 2 3 bilinguals were less successful at discriminating the difference between tokens like spi- 4 5 6 espi than the English monolinguals, although, interestingly, the monolinguals were, like 7 8 the bilinguals, also more accurate in the spi-spu and spi-spi trials than in the spi-espi 9 10 trials. The reason the monolingual English speakers may have performed less 11 12 13 accurately on the spi-espi trials than the spi-spu and spi-spi trials is potentially because, 14 15 as observed by Carlson (2016), English favours #sC over #VsC. Nevertheless, the 16 17 English monolinguals performed significantly more accurately than the Spanish 18 19 bilinguals in discriminating spi-espi in comparison to the other two conditions. 20 21 Fo 22 Furthermore, in this population of sequential Spanish-English bilinguals accurately not 23 24 producing the prothetic vowel in the phonemic fluency task was not associated with rP 25 26 accurate perception of the prothetic vowel. 27 28 ee 29 Moreover, although a greater amount of English daily use and improved 30 31 grammatical proficiency were associated with better production (i.e. not producing the rR 32 33 prothetic vowel), these same predictor variables did not lead to better discriminatory 34 35 ev 36 abilities in the perception task. Therefore, the results suggest that with regard to this 37 38 particular phonotactic constraint, production appears to be modified by increased iew 39 40 exposure to the L2 and grammatical proficiency, whereas perception does not. 41 42 Moreover, the analysis of the individual bilinguals did not fully verify the 43 44 45 assertion that accurate speech perception would be a prerequisite for accurate speech 46 47 production. For example, some participants scored poorly on the production task, but 48 49 above chance level on the perception task; alternatively, many participants scored 50 51 52 highly on the production task, but below chance level on the perception task. Although 53 54 many participants differed with regard to their accuracy on the perception versus 55 56 production tasks, some sequential Spanish-English bilinguals achieved high accuracy 57 58 59 results on both the perception and production tasks. Alternatively, other participants 60 Page 38 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 40 of 64 1 2 3 scored poorly on both tasks. These different profile patterns suggest that accurate L2 4 5 6 perception was neither necessary nor sufficient for accurate L2 production in the case 7 8 of these sequential Spanish-English bilinguals, but that, as previously noted, increased 9 10 exposure to English and improved grammatical proficiency could improve production 11 12 13 results. 14 15 Therefore, as in previous research into the relationship between speech 16 17 perception and production, no clear overall pattern was evident in terms of whether 18 19 perception or production necessarily “comes first” (Baker & Trofimovich, 2006; Beach, 20 21 Fo 22 Burnham, & Kitamura, 2001; Llisterri, 1995; Kartushina & Frauenfelder, 2014; 23 24 Sheldon & Strange, 1982; Zampini, 1998). With regard to increased L2 exposure, it rP 25 26 seems plausible that bilinguals exposed to more English would have been more likely 27 28 ee 29 to develop English-like pronunciation, and, likewise, it seems plausible that bilinguals 30 31 with higher grammatical proficiency would likewise be more metalinguistically aware rR 32 33 that in English /sC/ clusters in onset position are licit, and that this knowledge and 34 35 ev 36 exposure would enhance their L2 speech production, but not necessarily their 37 38 performance in the speeded perceptual discrimination task. iew 39 40 41 Why did the bilinguals seem to be able to produce English L2 speech without 42 43 the vowel in front of /sC/ clusters, but increased English use and improved English 44 45 46 grammatical proficiency did not seem to improve their perception of this contrast? 47 48 Firstly, in answering this question, it is important to remember that some participants 49 50 did perceive a difference but nevertheless produced /sC/ clusters in the Spanish manner 51 52 53 – so it would not be entirely correct to say that perceptual processes are always more 54 55 robust and pervasive as far as L1 influence in L2 processing than is speech production. 56 57 If that were the case then we would only have found evidence for enhanced speech 58 59 production, but not for speech perception. What is more likely is that either the tasks 60 Page 39 of 52 http://mc.manuscriptcentral.com/SLR Page 41 of 64 Second Language Research 1 2 3 assessed were actually quite different - and assessed speech perception and speech 4 5 6 production differently - or, alternatively, other predictor variables, not assessed within 7 8 this study, impact speech perception, rather than L2 use and L2 grammatical 9 10 proficiency, which proved to enhance speech production. For example, it may be that 11 12 13 some individuals simply hear differences between sounds better than others. We know 14 15 as well that some people suffer in musical terms from amusia, i.e. tone deafness (Powell, 16 17 2016). Similarly, someone who is less “apt” at hearing differences between sounds 18 19 might nonetheless be able to not produce the prothetic vowel in front of English /sC/ 20 21 Fo 22 clusters with the metalinguistic knowledge that this sound combination is possible in 23 24 English, and, with practice, become quite good at consistently not producing the vowel. rP 25 26 Note that what is different about this speech production task is that accuracy was 27 28 ee 29 reached when nothing was pronounced, rather than when a sound was produced more 30 31 “native-like”, e.g. such as a more similar vowel quality. Alternatively, someone who is rR 32 33 very “apt” at hearing differences between sounds (Powell, 2016) might essentially have 34 35 ev 36 a talent for hearing differences, which would be revealed in the perception task, but, 37 38 when taking part in the production task, that same person might not perform particularly iew 39 40 highly if not focussed, or if he or she didn’t know the grammatical rule in English, or 41 42 had not internalised that rule through practice. The point here is that different factors 43 44 45 might lead to improved speech perception than lead to improved speech production, 46 47 and the current study might only have examined factors which lead to improved speech 48 49 production, i.e. no production. 50 51 52 53 It may also be that the striking interpersonal differences between bilingual 54 55 participants arose due to different individual approaches to the tasks. Parlato-Oliveira 56 57 et al. (2010) summarised in their work that an effect of L2 exposure was most clearly 58 59 observed in their explicit task, but not in the implicit task, and suggested that L2 60 Page 40 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 42 of 64 1 2 3 experience is more likely to affect explicit, or metalinguistic, perceptual tasks. With 4 5 6 regard to our AX discrimination task, participants may have been metalinguistically 7 8 aware of the phonotactic constraint it assessed (note that grammatical knowledge was 9 10 correlated with production of the prothetic vowel), and the interpersonal differences 11 12 13 may have been a result of some participants simply being more aware of this 14 15 metalinguistic difference than others. 16 17 18 Furthermore, AX discrimination tasks require low-level perceptual judgments 19 20 on whether auditory input is the same or different. This is arguably easier to 21 Fo 22 23 discriminate than, for example, whether a vowel is present at the beginning of /sC/ or 24 whether the auditory input forms a word or non-word, and hence the perception task rP 25 26 27 may have assessed a rather low-level ability to acoustically discriminate, rather than the 28 ee 29 30 prothetic vowel - as a grammatical constraint - as such. Therefore, the interpersonal 31 rR 32 differences in the perception results may have been the result of some participants 33 34 simply acoustically discriminating better than others, with no direct insight into their 35 ev 36 grammatical knowledge; however, the bilinguals were just as good as the English 37 38 iew 39 monolinguals at perceiving the spi-spu contrast, whereas there was a significant 40 41 difference in their ability on the spi-espi task, so it cannot simply be the case that some 42 43 participants simply acoustically discriminated better than others. 44 45 46 It may also be that some of the prothetic vowels were actually hesitation 47 48 49 markers, and not reflective of grammatical constraints. It has been found that in 50 51 spontaneous speech hesitation markers are often used and there are discrepant 52 53 interpretations regarding their function, e.g. whether “um” and “uh” are used to “hold 54 55 56 the floor” or whether they are symptomatic of cognitive processes on the part of the 57 58 speaker, or both (Clark and Fox Tree, 2002; de Leeuw, 2007; Fox Tree, 2001; Maclay 59 60 and Osgood, 1959; Shriberg and Lickley, 1993). The present methodology included all Page 41 of 52 http://mc.manuscriptcentral.com/SLR Page 43 of 64 Second Language Research 1 2 3 vocalic utterances preceding the sibilant as prothetic, and thus, a future analysis may 4 5 6 consider including a read task as well, where hesitations are less likely. 7 8 9 It may also be, moreover, that the perception and production tasks recruited 10 11 different cognitive functions. As previously discussed, the phonemic verbal fluency 12 13 task is often used to assess executive control, as participants need to retrieve words, 14 15 16 which requires them to access their mental lexicon, whilst focusing on the task, 17 18 selecting words meeting certain constraints and inhibiting repetition (Miyake et al., 19 20 2000). This particular production task of eliciting words which begin with /sC/ clusters 21 Fo 22 23 would require the Spanish L1 participants to shift, as they moved from one cluster to 24 the next, monitor, to ensure that they didn’t repeat words, and to remember new words; rP 25 26 27 as well as inhibit their native Spanish language phonotactic constraint, as well as 28 ee 29 30 previously named words. In contrast, it could be argued that the perception task might 31 rR 32 have recruited a shifting mechanism, but that working memory and inhibition would 33 34 have been less involved, i.e. the participants only had two stimuli to remember. 35 ev 36 Likewise, it may be more generally that the two capacities – L2 speech production and 37 38 iew 39 L2 perception - recruit different executive control mechanisms to varying extents, 40 41 which could be an interesting research avenue to pursue in the future. 42 43 44 Essentially, the perception task in the current study was a ‘can you do it’ type 45 46 task, whilst for the production task the participants were required to think quickly 47 48 49 within a time constraint, which takes their focus off of their pronunciation. As such, in 50 51 the current experiment, the production task was cognitively more costly, and it therefore 52 53 seems possible that the relationship between production and perception might have 54 55 56 been more clear cut if the tasks had required similar cognitive demands. That said, if it 57 58 is accepted that the production task was more cognitively demanding than the 59 60 perception task, and accurate perception is indeed a prerequisite for accurate production, Page 42 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 44 of 64 1 2 3 we should nevertheless not have observed bilinguals who had high accuracy rates on 4 5 6 the production task, but low accuracy rates on the perception task (i.e. bottom left-hand 7 8 quarter in Figure 12). The fact that this quarter was indeed quite full of participants, 9 10 indicates that accurate L2 speech perception is not necessary for accurate L2 speech 11 12 13 production. 14 15 16 Perhaps it was not only the task which varied in terms of different executive 17 18 control mechanisms, or whether it was explicit or implicit (although this is of course 19 20 the case too), but that the individuals approached the same tasks differently. For 21 Fo 22 23 example, in the case of the perception task, participants may have differed in the amount 24 of attention they devoted to the task. Those participants who were very focused may rP 25 26 27 have performed more accurately than those who were less focused, regardless of how 28 ee 29 30 proficient they were in their L2. Likewise, those participants who were more focussed 31 rR 32 on their pronunciation in the production task might have produced no prothetic vowel, 33 34 whilst those participants who focussed more on the task of naming as many words as 35 ev 36 possible could have more frequently produced a prothetic vowel. Again, this would 37 38 iew 39 have little to do with their proficiency in English as such, but rather with how they 40 41 approached the task, and may help to explain the differences observed between 42 43 participants. Nonetheless, at face value, the results do not lend direct support to the 44 45 46 notion that perception precedes production in L2 acquisition. 47 48 49 As such, the findings suggest that L2 speech models such as SLM and PAM, 50 51 which postulate to different extents that perception is linked to production, may need 52 53 revisiting. Indeed, some of the participants in the present study were able to not produce 54 55 56 the prothetic vowel (i.e. accurately, as in English), but performed below chance with 57 58 regard to the perception of the prothetic vowel in the same-different discrimination task. 59 60 The findings from the correlational analysis, which indicated no significant relationship Page 43 of 52 http://mc.manuscriptcentral.com/SLR Page 45 of 64 Second Language Research 1 2 3 between perception and production, as well as in the mixed model analyses, which 4 5 6 likewise did not indicate that perception abilities predicted production, both suggest 7 8 that perception and production were not clearly related to one another with regard to 9 10 the tasks at hand. 11 12 13 In brief, the results confirm a growing body of research examining the 14 15 perception and production of individual L2 sounds which suggests dissociations 16 17 between L2 perception and production. All in all, the results reveal that it is possible 18 19 for sequential bilinguals to acquire knowledge of new permissible phoneme sequences, 20 21 Fo 22 which violate their L1 phonotactic constraints, but that there is a great amount of 23 24 interpersonal variability within regard to how the L2 acquisitional process unfolds. rP 25 26 27 28 ee 29 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 44 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 46 of 64 1 2 3 References 4 5 6 Baker W and Trofimovich P (2006) Perceptual paths to accurate production of L2 7 vowels: The role of individual differences. IRAL – International Review of 8 Applied Linguistics in Language Teaching 44(3): 231–250. 9 10 Barr D J, Levy R, Scheepers, C and Tily, HJ (2013) Random effects structure for 11 confirmatory hypothesis testing: Keep it maximal. Journal of Memory and 12 Language, 68(3): 255-278. 13 14 15 Beach, E. F., Burnham, D. & Kitamura. C. (2001). Bilingualism and the relationship 16 between perception and production: Greek/English bilinguals and Thai bilabial 17 stops. International Journal of Bilingualism, 5(2), 221–235. 18 19 Best CT (1995) A direct realist view of cross-language speech perception. In: Strange 20 W (ed.) Speech Perception and Linguistic Experience: Theoretical and 21 Methodological Issues. Baltimore: York Press, pp. 171–204. Fo 22 23 Bialystok E, Craik F and Luk G (2008) Cognitive control and lexical access in 24 younger and older bilinguals. Journal of experimental psychology. Learning, rP 25 26 memory, and cognition 34(4): 859–873. 27 28 Boersma P and Weenink D (2010) PRAAT. University of Amsterdam. Available at: ee 29 http://www.praat.org. 30 31 Cabrelli J, Luque A and Finestrat-Martínez I (2019) Influence of L2 English rR 32 phonotactics in L1 Brazilian Portuguese illusory vowel perception. Journal of 33 34 Phonetics 73: 55–69. 35 ev 36 Carlson MT, Goldrick M, Blasingame M, et al. (2016) Navigating conflicting 37 phonotactic constraints in bilingual speech perception. Bilingualism: 38 Language and Cognition 19(5): 939–954. iew 39 40 Cheshire J, Kerswill P, Fox S, et al. (2011) Contact, the feature pool and the speech 41 community: The emergence of Multicultural London English. Journal of 42 Sociolinguistics 15(2): 151–196. 43 44 45 Clark HH and Fox Tree JE (2002) Using uh and um in spontaneous speaking. 46 Cognition 84(1): 73–111. 47 48 Cuetos F, Hallé P, Dominguez A, et al. (2011) Perception of prothetic /e/ in #sC 49 utterances: Gating data. In: ICPhS XVII Conference Proceedings, Hong Kong, 50 17 August 2011. 51 52 de Leeuw E and Bogulski CA (2016) Frequent L2 language use enhances executive 53 control in bilinguals. Bilingualism: Language and Cognition 19(5): 907–913. 54 55 56 de Leeuw E (2007) Hesitation markers in English, German, and Dutch. Journal of 57 Germanic Linguistics 19(2): 85–114. 58 59 60 Page 45 of 52 http://mc.manuscriptcentral.com/SLR Page 47 of 64 Second Language Research 1 2 3 de Leeuw E and Celata C (2019) Plasticity of native phonetic and phonological 4 5 domains in the context of bilingualism. Journal of Phonetics 75: 88–93. 6 7 Dupoux E, Kakehi K, Hirose Y, et al. (1999) Epenthetic vowels in Japanese: A 8 perceptual illusion? Journal of Experimental Psychology: Human Perception 9 and Performance 25(6). American Psychological Association: 1568. 10 11 Escudero P (2005) Linguistic Perception and Second Language Acquisition: 12 Explaining the Attainment of Optimal Phonological Categorization. 13 University of Utrecht, Utrecht. 14 15 16 Flege JE (1987) The production of ‘new’ and ‘similar’ phones in a foreign language: 17 Evidence for the effect of equivalence classification. Journal of Phonetics 15: 18 47–65. 19 20 Flege JE (1995) Second language speech learning: Theory, findings, and problems. 21 In: Strange W (ed.) Speech Perception and Linguistic Experience: Theoretical Fo 22 and Methodological Issues. Maryland: York Press, pp. 233–277. 23 24 Flege JE, Bohn O-S and Jang S (1997) Effects of experience on non-native speakers’ rP 25 26 production and perception of English vowels. Journal of Phonetics 25(4): 27 437–470. 28 ee 29 Flege JE, Schirru C and MacKay IRA (2003) Interaction between the native and 30 second language phonetic subsystems. Speech Communication 40(4): 467– 31 491. rR 32 33 Forster KI and Forster JC (2003) DMDX: A Windows display program with 34 35 millisecond accuracy. Behavior Research Methods, Instruments, & Computers ev 36 35(1): 116–124. 37 38 Fowler CA (1996) Listeners do hear sounds, not tongues. The Journal of the iew 39 Acoustical Society of America 99(3): 1730–1741. 40 41 Fox Tree JE (2001) Listeners’ uses ofum anduh in speech comprehension. Memory & 42 Cognition 29(2): 320–326. 43 44 Freeman MR, Blumenfeld HK and Marian V (2016) Phonotactic constraints are 45 46 activated across languages in bilinguals. Frontiers in Psychology 7. 47 48 Gollan TH, Montoya RI and Werner GA (2002) Semantic and letter fluency in 49 Spanish-English bilinguals. Neuropsychology 16(4): 562–576. 50 51 Grogan A, Green DW, Ali N, et al. (2009) Structural correlates of semantic and 52 phonemic fluency ability in first and second languages. Cerebral Cortex 53 19(11): 2690–2698. 54 55 Grosjean F (1998) Studying bilinguals: Methodological and conceptual issues. 56 57 Bilingualism: Language and Cognition 1(02): 131–149. 58 59 Gullberg M and Indefrey P (2003) Language History Questionnaire. Developed in 60 The Dynamics of Multilingual Processing. Max Planck Institute for Page 46 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 48 of 64 1 2 3 Psycholinguistics, Nijmegen. Available at: 4 5 http://www.mpi.nl/departments/other-research/research-projects/the- 6 dynamics-of-multilingual-processing/tools/Lang-Hist-Quest-Engl.pdf. 7 8 Hallé PA, Dominguez A, Cuetos F, et al. (2008) Phonological mediation in visual 9 masked priming: Evidence from phonotactic repair. Journal of Experimental 10 Psychology: Human Perception and Performance 34(1): 177–192. 11 12 Hualde JI (2005) The Sounds of Spanish. Cambridge, UK. Available at: 13 http://www.cambridge.org/gb/academic/subjects/languages- 14 15 linguistics/semantics-and-pragmatics/sounds-spanish (accessed 3 January 16 2018). 17 18 Kartushina N and Frauenfelder UH (2014) On the effects of L2 perception and of 19 individual differences in L1 production on L2 pronunciation. Frontiers in 20 Psychology 5. 21 Fo 22 Kivistö-de Souza H (2015) Phonological awareness and pronunciation in a second 23 language. PhD. Universitat de Barcelona, Barcelona. Available at: 24 https://core.ac.uk/download/pdf/78632484.pdf. rP 25 26 27 Klein-Braley C (1985) A cloze-up on the C-Test: a study in the construct validation of 28 authentic tests. Language Testing 2(1): 76–104. ee 29 30 Kuhl PK (2004) Early language acquisition: Cracking the speech code. Nature 31 Reviews Neuroscience 5(11): 831–843. rR 32 33 Liberman AM and Mattingly IG (1985) The motor theory of speech perception 34 35 revised. Cognition 21(1): 1–36. ev 36 37 Llisterri J (1995) Relationships between speech production and speech perception in a 38 second language. In: Proceedings of the 13th International Congress of iew 39 Phonetic Sciences, Stockholm, 1995. 40 41 Maclay H and Osgood CE (1959) Hesitation Phenomena in Spontaneous English 42 Speech. WORD 15(1): 19–44. 43 44 Martínez-Celdrán E, Fernández-Planas AM and Carrera-Sabaté J (2003) Castilian 45 46 Spanish. Journal of the International Phonetic Association 33(2): 255–259. 47 48 Mills L (2014) An articulatory study of GOOSE-fronting in varieties of English. PhD. 49 Queen Margaret University, Edinburgh. Available at: 50 https://eresearch.qmu.ac.uk/handle/20.500.12289/7678 (accessed 4 September 51 2018). 52 53 Miyake A, Friedman NP, Emerson MJ, et al. (2000) The unity and diversity of 54 executive functions and their contributions to complex “Frontal Lobe” tasks: 55 56 A latent variable analysis. Cognitive Psychology 41(1): 49–100. 57 58 Morales Pech D and Izquierdo J (2011) L2 phonology learning among young adult 59 learners of English: Effects of regular classroom-based instruction and L2 60 Page 47 of 52 http://mc.manuscriptcentral.com/SLR Page 49 of 64 Second Language Research 1 2 3 proficiency. Revista Electrónica de Investigación Educativa 13(1). Available 4 5 at: http://redie.uabc.mx/vol12no2/contentsmoralesizquierdo.html. 6 7 Newcombe F (1969) Missle Wounds of the Brain. A Study of Psychological Deficits. 8 Oxford, UK: Oxford University Press. 9 10 Ohala JJ (1996) Speech perception is hearing sounds, not tongues. The Journal of the 11 Acoustical Society of America 99(3): 1718–1725. 12 13 Parlato-Oliveira E, Christophe A, Hirose Y, et al. (2010) Plasticity of illusory vowel 14 perception in Brazilian-Japanese bilinguals. The Journal of the Acoustical 15 16 Society of America 127(6): 3738–3748. 17 18 Portocarrero JS, Burright RG and Donovick PJ (2007) Vocabulary and verbal fluency 19 of bilingual and monolingual college students. Archives of clinical 20 neuropsychology : The official journal of the National Academy of 21 Neuropsychologists 22(3): 415–422. Fo 22 23 Powell J (2016) Why We Love Music: From Mozart to Metallica - The Emotional 24 Power of Beautiful Sounds. London: John Murray. Available at: rP 25 26 http://www.hodder.co.uk/HodderStoughton/books/detail.page?isbn=97814736 27 13751 (accessed 10 September 2018). 28 ee 29 Raatz U and Klein-Braley C (1981) The C-Test--A Modification of the Cloze 30 Procedure. Not available separately; see FL 013 014. 31 rR 32 Rubinstein JS, Meyer DE and Evans JE (2001) Executive control of cognitive 33 processes in task switching. Journal of Experimental Psychology: Human 34 35 Perception and Performance 27(4): 763–797. ev 36 37 Sandoval TC, Gollan TH, Ferreira, VS and Salmon, DP (2010). What causes the 38 bilingual disadvantage in verbal fluency: The dual-task analogy. Bilingualism: iew 39 Language and Cognition, 13: 231–252. 40 41 Sauzéon H, Raboutet C, Rodrigues J, et al. (2011) Verbal knowledge as a 42 compensation determinant of adult Age differences in verbal fluency tasks 43 over time. Journal of Adult Development 18(3): 144–154. 44 45 46 Sheldon A and Strange W (1982) The acquisition of /r/ and /l/ by Japanese learners of 47 English: Evidence that speech production can precede speech perception. 48 Applied Psycholinguistics 3(3): 243–261. 49 50 Shriberg EE and Lickley RJ (1993) Intonation of Clause-Internal Filled Pauses. 51 Phonetica 50(3): 172–179. 52 53 So CK and Best CT (2010) Cross-language perception of non-native tonal contrasts: 54 Effects of native phonological and phonetic influences. Language and Speech 55 56 53(2): 273–293. 57 58 So CK and Best CT (2014) Phonetic influences on English and French listeners’ 59 assimilation of mandarin tones to native prosodic categories. Studies in Second 60 Language Acquisition 36(2): 195–221. Page 48 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 50 of 64 1 2 3 Tomás Navarro T (1918) Manual de Pronunciación Española. Reprinted in 1982. 4 5 Madrid: CSIC. 6 7 Zampini ML (1998) The relationship between the production and perception of L2 8 Spanish stops. Texas Papers in Foreign Language Education: 1–18. 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 49 of 52 http://mc.manuscriptcentral.com/SLR Page 51 of 64 Second Language Research 1 2 3 Appendix I 4 5 6 Language use question regarding daily use 7 8 Consider your present amount of English and Spanish language use. Divide 9 100% into English and Spanish use, e.g. if you only speak English presently, 10 write "English=100%; Spanish=0%". If you also use another language, or other 11 languages, include them in your division of 100%. 12 13 Alternatively, if you speak both English and Spanish equally, write "English=50%; 14 Spanish=50%". Divide according to your own estimates, considering all aspects of 15 language use, e.g. both passive (watching television) and active (speaking with 16 parents). 17 18 _____________________________________________________________________ 19 ______ 20 21 Fo 22 23 C-tests 24 rP 25 On the next pages you will find 2 small texts in total. Each text contains gaps where 26 27 parts of some words haves been left out (no whole words are missing, though). 28 Please try and fill in the gaps. In many cases there are several possibilities, so there ee 29 are no right or wrong answers. Thank you very much for your help. 30 31 Text 1: rR 32 Two former US navy ships contaminated with chemicals were expected to arrive in the 33 34 English Channel last night. The Maritime and Coastguard Agency 35 sa____________________ the ves____________________ , at the ev 36 cen____________________ of an enviro____________________ row, 37 we____________________ being to____________________ through the 38 cha____________________ , before hea____________________ up the east iew 39 co____________________ to Hartlepool. Pl____________________ to 40 41 dism____________________ them in north-east England have been 42 she____________________ after being dee____________________ to 43 fl____________________ international ru____________________. Last 44 we____________________ , the gover____________________ said the ships could be 45 sto____________________ in Hartlepool before go____________________ back 46 acr____________________ the Atlantic. 47 Two former US navy ships contaminated with chemicals were expected to arrive in the 48 49 English Channel last night. The Maritime and Coastguard Agency says the vessels, at 50 the centre of an environmental row, were being towed through the channel, before 51 heading up the east coast to Hartlepool. Plans to dismantle them in north-east England 52 have been shelved after being deemed to flaunt international rules. Last week, the 53 government said the ships could be stowed/stored in Hartlepool before going back 54 across the Atlantic. 55 56 57 Text 2: 58 Don’t get me wrong. I love magazines. I’ve been addicted to them since my teenage 59 years. There’s some____________________ about wom____________________ 60 magazine superfi____________________ that I of____________________ enjoy. But Page 50 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 52 of 64 1 2 3 oh b__________________ , they are ju____________________ so, so 4 5 frustr____________________ predictable. I rec____________________ you 6 co____________________ cobble o____________________ together very 7 eas____________________ in five min____________________. Take the 8 co____________________ for example: the cover im____________________ , get a 9 he____________________ and shou____________________ shot of a 10 smi____________________ , heavily make-uped and airbr____________________ 11 model (or optio____________________ a fam____________________ person). 12 13 Don’t get me wrong. I love magazines. I’ve been addicted to them since my teenage 14 years. There’s something about women’s magazine superficiality that I often enjoy. But 15 oh boy, they are just so, so frustratingly predictable. I reckon you could cobble one 16 together very easily in five minutes. Take the cover for example: the cover image, get 17 a head and shoulder shot of a smiling, heavily make-uped and airbrushed model (or 18 optionally a famous person). 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Page 51 of 52 http://mc.manuscriptcentral.com/SLR Page 53 of 64 Second Language Research 1 2 3 Appendix II 4 5 6 7 Participant Recorded in Amount of Daily C-Test Age of English 8 Spain (2) vs UK English Use (%) Result Acquisition 9 (1) 10 1 2 25 32.5 4 11 12 2 2 20 44 6 13 3 2 30 55 6 14 4 2 20 79 7 15 5 2 5 46.5 6 16 17 6 1 40 51.5 5 18 7 2 70 31.5 6 19 8 1 10 16.5 30 20 9 2 40 28 11 21 10 2 35 45 8 Fo 22 23 11 1 60 70 7 24 12 2 40 40 1 rP 25 13 1 20 6.5 10 26 27 14 2 25 60 3 28 15 1 40 40.25 5 ee 29 16 2 20 37.5 8 30 17 2 30 46.5 8 31 18 2 30 30.5 10 rR 32 33 19 1 30 33 7 34 20 2 20 21.5 6 35 ev 21 2 20 11.5 7 36 37 22 2 40 27.5 7 38 23 2 40 30 3 iew 39 24 2 1 7.5 5 40 25 2 30 31.5 5 41 42 26 2 40 62.5 7 43 27 1 80 51.25 7 44 28 2 35 34.5 8 45 29 2 35 33 6 46 47 30 2 5 30.5 6 48 31 2 40 32.5 10 49 32 2 0 46.5 10 50 33 2 0 44 11 51 52 34 2 30 41.5 4 53 54 55 56 57 58 59 60 Page 52 of 52 http://mc.manuscriptcentral.com/SLR Second Language Research Page 54 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 36 iew 37 38 39 40 41 Figure 1 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Page 55 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 36 iew 37 38 39 40 41 Figure 2 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Second Language Research Page 56 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 36 iew 37 38 39 40 41 42 Figure 3 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Page 57 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 36 iew 37 38 39 40 41 42 Figure 4 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Second Language Research Page 58 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 Figure 5 36 iew 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Page 59 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 Figure 6 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 36 iew 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Second Language Research Page 60 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 Figure 4: Example of a prothetic vowel, ‘speak’, produced by Participant 18. 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Page 61 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 Figure 5: Example of a prothetic vowel, ‘slain’ produced by Participant 16. 30 31 rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Second Language Research Page 62 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Fo 22 23 24 rP 25 26 27 28 ee 29 30 31 Figure 6: Example of a prothetic vowel, ‘smile’ produced by Participant 1. rR 32 33 34 35 ev 36 37 38 iew 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Page 63 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 Figure 10 35 36 iew 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Second Language Research Page 64 of 64 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 36 Figure 11 iew 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR Page 65 of 64 Second Language Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Fo 20 21 22 23 rP 24 25 26 ee 27 28 29 rR 30 31 32 ev 33 34 35 36 Figure 12 iew 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 http://mc.manuscriptcentral.com/SLR View publication stats