Speech Communication 146 (2023) 93–108 Contents lists available at ScienceDirect Speech Communication journal homepage: www.elsevier.com/locate/specom The acquisition of L2 voiced stops by English learners of Spanish and Spanish learners of English Celia Gorba *, Juli Cebrian Departament de Filologia Anglesa i Germanıstica, Facultat de Filosofia i Lletres, Universitat Autonoma de Barcelona, Campus UAB, Bellaterra 08193, Spain A R T I C L E I N F O A B S T R A C T Keywords: Previous studies investigating the acquisition of L2 stops have found a positive effect of L2 experience, but few L2 acquisition have focused on voiced stops, particularly on prevoicing. This study investigates the acquisition of /b/ and /g/ by Voiced stops two populations, English learners of Spanish and Spanish learners of English. Three groups varying in amount of VOT L2 experience (mainly length of residence, but also L2 use and L2 instruction) were investigated for each L1 Prevoicing L2 experience population. Participants completed a carrier sentence reading task in their L1 and L2. Results showed that Length of residence amount of L2 experience had a positive effect on L2 stop production, as the least experienced groups were outperformed by the experienced ones. No clear effect of L2 experience was observed on the L1, as learners did not differ from monolingual controls, but some differences between learner groups emerged. Moreover, overall, L2 learners were able to produce L1 and L2 stops differently, which indicates that their L1 and L2 categories were not merged. Still, the L1-English L2-Spanish speakers produced L2 stops more accurately than the L1-Spanish L2- English groups, suggesting that learning to rely on an existing L1 cue may be easier than learning to use a cue associated with a different L1 category. 1. Introduction the production of L2 and L1 voiced stops, and (2) to contrast two groups of learners, whose L1 systems differ in the use of phonetic cues to Adult learners of a second language (L2) often produce L2 sounds implement the voicing contrast for initial stops (English learners of differently from native speakers, as their first language (L1) may have Spanish and Spanish learners of English), as explained below. influenced their L2 speech (Best and Tyler, 2007; Flege, 1995; Kuhl, Testing these two populations is of special theoretical interest given 1991). Gaining experience in the L2 – which may involve length of the different distribution of the cues used to signal stop voicing in the residence in an L2 setting (e.g., Flege 1987, 1995, Flege et al. 1997; two languages. Specifically, on the one hand, Spanish voiced stops are Gorba 2019; Gorba and Cebrian 2021; Jun and Cowie 1994; Levy and prevoiced, and are contrasted with short-lag VOT (voiceless stops) to Law 2010), L2 use (e.g., Kartushina and Martin 2019; Piske et al. 2001) signal voicing. On the other hand, in English prevoicing and short-lag or L2 instruction (e.g., Dmitrieva et al. 2020; Nagle 2019) – may help L2 VOT do not contrast and are both used to produce voiced stops, short- learners improve their pronunciation in the target language. Moreover, lag VOT being more common, particularly in initial post-pausal posi­ the L1 of an L2 learner may also be affected in the process of learning an tion (e.g., Lisker and Abramson 1964; see Section 1.2.1. for a detailed L2. This may happen as a result of phonetic drift from the L1 towards the review of VOT use in the languages under study). Thus, the underlying L2, when L1 sounds or features become more L2-like, (e.g., Chang 2012, question is whether the acquisition of L2 voiced stops by these two 2013; Dmitrieva et al. 2020; Flege 1987; Herd et al. 2015; Kartushina populations may differ due to the distributional differences of VOT be­ et al. 2016), or away from the L2, when L1 structures are modified to tween English and Spanish. Several scenarios can be considered. For differentiate them from L2 structures (Flege and Eefting, 1987). Thus, instance, one possibility is that learners would benefit from L1 transfer. crosslinguistic influence – i.e., the way the first (L1) and second lan­ In the case of the English learners of Spanish, positive transfer would guage (L2) of an L2 learner interact – may occur in either direction (e.g., involve using prevoicing in the L2, as voice-lead VOT is a possible, albeit Flege 1987; Major 1992; Sancier and Fowler 1997). The aim of the less common, cue to stop voicing in initial position, used in free variation present study is twofold: (1) to investigate the effect of L2 experience on with short-lag VOT. In the case of the Spanish speakers, producing L2 * Corresponding author. E-mail address: celia.gorba@uab.cat (C. Gorba). https://doi.org/10.1016/j.specom.2022.12.003 Received 9 December 2021; Received in revised form 17 November 2022; Accepted 23 December 2022 Available online 25 December 2022 0167-6393/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by- nc-nd/4.0/). C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 stops with prevoicing, the cue used in the L1, could be regarded as ac­ voiceless stops to a greater extent than learners studying in the US. curate, as prevoicing is a possible and unequivocal cue to voiced stop Generally, the longer the length of residence, the more target-like L2 production in English, even if a more target-like speech would be ach­ productions may be (e.g., Flege 1987; Flege et al. 1997; Lev-Ari and ieved with a reduction in the frequency of use of this cue. A second Peperkamp 2013; Levy and Law 2010). Moreover, immersive experience scenario involves negative transfer and would take place if English tends to have a greater impact on L2 learning than L2 instruction. For learners simply used short-lag VOT predominantly in Spanish voiced instance, Mora (2008) conducted a longitudinal study in which stop productions, and Spanish learners used the same amount and fre­ L1-Spanish/Catalan learners of English were tested before and after quency of prevoicing as in their L1. The former case would be more receiving formal instruction and, also, after a three-month stay abroad. problematic than the latter, as short-lag VOT stops signal voiceless stops Results suggested that instruction had only a small effect on L2 VOT in Spanish. Furthermore, acquiring the L2 categories for voiced stops productions, but production became more target-like after the stay may also influence the L1 – i.e., it may result in a phonetic drift towards abroad. L2 values. A third possibility is thus that the learners’ L1 is affected by On the other hand, a number of studies have also shown that L2 the L2 if the learning of the L2 category results in a drift of the L1 values experience may result in less native-like productions in the L1 due to to more L2-like values, for instance, by reducing the amount of pre­ phonetic drift towards L2-like values (e.g., Bergmann et al., 2016; voicing in L1 Spanish or increasing it in L1 English – i.e., category Chang 2012, 2013; Flege 1987; Harada 2003; Kartushina et al. 2016, see assimilation in Flege’s (2003) terms. The opposite scenario, category Section 1.2.1). In short, it appears that, overall, living in an L2 setting dissimilation, is also a possibility. In the case of the English learners of may accelerate the acquisition of L2 phones and that length of residence Spanish, this would involve making a greater distinction between the modulates the amount of crosslinguistic influence. However, if living in two languages by reducing the frequency of use of prevoicing in the L1, an immersion setting does not go hand in hand with other factors such as while in the case of the Spanish learners of English, by increasing the amount of L2 use, the effects of length of residence on both the L2 and amount of prevoicing in the L1. Examining the L1 and L2 production of the L1 may be lower. Flege and Bohn (2021) analyzed the data in Flege speakers of both languages and considering the differences between the (1995) as well as data collected in 2003 from the same population – i.e., two phonetic systems thus allows us to assess these different possible native Italian speakers living in Canada – and observed that, even learning scenarios. though participants were comparable in terms of length of residence, Another important contribution of the present study is that it in­ some improved in their production of English /p, t, k/, whereas others vestigates the acquisition of voiced stops, which has received little showed the opposite pattern. This difference was attributed to differ­ attention in previous studies compared to the acquisition of voiceless ences in language use, as those who improved used the L2 to a greater stops. Moreover, as discussed above, the inclusion of both Spanish extent. The results also showed that the L2 may keep changing over time learners of English and English learners of Spanish allows for the eval­ as a function of L2 use. Thus, the revised version of the SLM (SLM-r, uation and comparison of the acquisition of two different main cues to Flege and Bohn, 2021) claims that the effect of length of residence alone voicing: short-lag VOT for Spanish learners of English and predominant is not an appropriate predictor of L2 learning, and points to the use of prevoicing for English learners of Spanish. Finally, the current importance of the amount of L1 and L2 use. In this regard, the SLM-r study also presents a methodological innovation, given that a categori­ proposes a variable, referred to as Full Time Equivalent (FTE), that cal measure, namely the presence (yes/no) of prevoicing, is used to combines length of residence in the L2 setting and proportion of L2 use study voiced stops, whereas most previous research use VOT duration (e. to quantify L2 input. Furthermore, previous studies show that L2 use, as g., Baese-Berk 2019; Casillas 2019; Dmitrieva et al. 2020; Flege and well as L2 instruction, can have an effect not only on the L2 but also on Eefting 1987; Nagle 2019; Zampini 1998, amongst others). The the L1 (e.g., Chang 2012; Herd et al. 2015; Nagle 2019), even in a non L2 following section will review aspects that are relevant to the present setting (e.g., Kartushina and Martin 2019). For example, Kartushina and paper, including the different processes involved in crosslinguistic in­ Martin (2019) found that after participating in a two-week study abroad fluences and the acquisition of voiced stops. program in the Netherlands – without explicit L2 instruction – using mainly the L2 as a lingua franca, Spanish-Basque learners of English 1.1. Effects of L2 experience on L1 and L2 improved their pronunciation of English vowels. In addition, there was evidence of influence from their L2 on their native languages. However, The manner in which the first and the second language interact has these effects diminished after four weeks back in their home country, been addressed by the most influential L2 speech models. For example, highlighting the role of language use in crosslinguistic influence. the Speech Learning Model (SLM, Flege, 1995, 2002, 2007), and its Extensive L2 instruction may also result in more target-like productions revised version (SLM-r, Flege and Bohn, 2021), posit that crosslinguistic in the L2 and phonetic drift of the L1 towards L2 values even in an L1 influence is motivated by the fact that the L1 and the L2 coexist in a setting (e.g., Herd et al. 2015; Nagle 2019). For instance, Herd et al. common phonetic space, which allows bidirectional influence – i.e., (2015) found a greater use of negative VOT – i.e., Spanish-like values – from the L1 on the L2 as well as from the L2 on the L1. At initial stages of in the L1 of advanced L1-English learners of Spanish than in the case of L2 acquisition, an L2 phone may be assimilated to a close L1 phone – i.e., learners who had received a smaller amount of L2 instruction. Another perceived in terms of a similar L1 phone – hindering the creation of a possibility is that L2 experience in the form of L2 instruction may result new L2 category. Given sufficient exposure, an L2 learner may be able to in changes in the L1 categories in the opposite direction from the L2 eventually create a separate category for the L2 phone. Early versions of counterparts – i.e., category dissimilation. For instance, Huffman and the SLM considered L2 experience, quantified mainly as length of resi­ Schuhmann (2016) found that some American learners of Spanish pro­ dence, as the main factor affecting crosslinguistic influence. The extent duced fewer instances of prevoiced stops in L1 English after 6 weeks of and direction of crosslinguistic influence may in fact be modulated by a L2 instruction. number of factors, including, for instance, starting age of learning, L1 Moreover, combining more than one type of L2 experience may and L2 use, formal instruction and motivation (Piske et al., 2001). A enhance L2 acquisition to a greater extent. Lord (2010) compared the wide range of studies have investigated the effect of length of residence effect of an 8-week immersion program on English learners of Spanish in an L2 setting (e.g., Bohn and Flege 1990; Flege 1987; Flege et al., who had previously received L2 instruction and on a similar group with 1997; Gorba 2019; Gorba and Cebrian 2021; Jun and Cowie 1994; Levy no prior L2 instruction. Results showed that, even though both groups and Law 2010; Stevens 2001). Overall, living in an L2 setting has been improved, participants with prior instruction outperformed the group found to improve the production of L2 sounds. For example, Stevens with no instruction on the production of L2 obstruents both before and (2001) found that American learners of Spanish studying in Spain with after the immersion setting. This finding indicated that, even though no previous pronunciation instruction improved their production of L2 immersion has a beneficial effect on L2 production, it can be boosted 94 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 with prior L2 instruction. In short, length of residence in an L2 setting stops in three different sessions by L1-English students enrolled in a modulates the amount and direction of crosslinguistic influence (e.g., 15-week Spanish phonetics course. Results showed that, by the last Flege 1987; Mora 2008), but other factors, such as L2 use (e.g., Kar­ session, the L2 learners produced Spanish /p/ with VOT values that tushina and Martin 2019) and L2 instruction (e.g., Herd et al. 2015; resembled those of Spanish native speakers and that differed from their Nagle 2019) also play an important role. Next, research conducted on own productions of English /p/. Similarly, Casillas (2019) investigated the acquisition of L2 stops, which are the focus of the present paper, will the progress of English learners of Spanish in their production of /p/ be reviewed. throughout a seven-week immersion course where the use of their L1 was prohibited and found that learners performed in a target-like 1.2. The acquisition of L2 stops manner at the end of the program, as their VOT values did not differ from those of Spanish-English bilinguals. Chang (2012) studied the The acquisition of L2 stops may pose a difficulty for L2 learners given production of L1 and L2 stops and vowels of L1-English learners of potential phonetic differences between the L1 and L2 phones and dis­ Korean who were both enrolled in a language course and living in an similarities regarding the use of the voicing cues. In fact, the L2 and L1 immersion setting. Results showed evidence of L1 phonetic drift towards categories of L2 learners may differ from those of monolingual speakers. L2 values as early as two weeks into the L2 course. It should also be Particularly, the production of VOT, which has been found to be the noted that the influence of the L2 on the L1 was greater for beginner main cue for voicing in stops in several languages (Lisker and Abram­ learners than for learners who already had some knowledge of the L2 son, 1964), has been investigated in a number of previous studies (Chang, 2013). This difference was attributed to a novelty effect – i.e., assessing L1 and L2 production (e.g., Castañeda 1986; Docherty 1990; the fact that novel perceptual experiences may result in a greater Lisker and Abramson 1964). This section reviews relevant studies on the salience and heightened encoding of L2 stimuli. In short, L2 learners are acquisition of L2 stops and ends with a description of the VOT system of capable of improving their production of L2 voiceless stops by both both Spanish and in English – i.e., the two languages examined in this learning to use long-lag VOT (e.g., Spanish and French learners of En­ study. glish) and reducing VOT duration towards short-lag values (e.g., English learners of French/Spanish). Moreover, the L1 may also drift towards L2 1.2.1. Previous studies on the acquisition of L2 stop voicing values given sufficient exposure to the L2 – e.g., a long residence in the The acquisition of stops has received considerable attention in the L2 setting – or due to a novelty effect. literature, especially when it comes to voiceless stops (e.g., Caramazza So far, the studies reported above have focused on voiceless stops. et al. 1973; Casillas 2019; Flege 1990; Fowler et al. 2008; Harada 2003; Research investigating their voiced counterparts seems to suggest Hazan and Boulakia 1993; Major 1992; Schuhmann and Huffman 2015; overall that voiced stops may pose a greater difficulty for L2 speakers Williams 1977; Zampini 1998). The results generally suggest that L2 than voiceless stops and, in fact, studies investigating both types of production of voiceless stops evolves towards more target-like values categories show a more target-like production regarding the latter with increased L2 experience (e.g., Flege 1987; Gorba and Cebrian 2021; (Gorba, 2020). For example, a number of studies have investigated the Major 1992; Schuchmann and Huffman 2019; Stevens 2001; Yang et al. acquisition of short-lag voiced stops, especially English stops, by 2022), and that the L1 category may also experience changes in the speakers of languages that contrast prevoicing (voiced stops) and direction of the L2, particularly given sufficient L2 input and use (e.g., short-lag VOT (voiceless stops), such as Spanish, Italian or French (e.g., Flege 1987; Major 1992; Sancier and Fowler 1997; Herd et al. 2015). For Caramazza et al. 1973; Flege and Eefting 1987; Hazan and Boulakia instance, Flege (1987) examined the production of L1 and L2 /t/ by 1993; MacKay et al. 2001; Nathan 1987; Williams 1977). These studies French learners of English – as well as L1-English learners of French – have generally found that L2 learners did not produce L2 stops with differing in setting and length of residence and found evidence of bidi­ target-like values. Flege and Eefting (1987) found that L1-Spanish rectional influence: French learners of English living in the US produced L2-English speakers – of different ages and with different degrees of /t/ in both languages with VOT values that differed from monolingual L2 exposure – produced English voiced stops with short-lag VOT speakers’ values and were intermediate to the values expected for each significantly less often than L1-English speakers. Even though they made language. Flege (1987) also found that two groups of American learners a small numerical difference in the use of prevoicing in Spanish as of French in an L1 setting produced significantly longer VOT values for opposed to English (6% more frequent in Spanish), this difference did French /t/ than French monolinguals, whereas the Americans living in not reach significance. Similarly, MacKay et al. (2001) found that an immersion setting did not, illustrating a positive effect of residence in L1-Italian L2-English speakers differing in age of learning and L2 use an L2 setting on L2 production. In a study that examined the production prevoiced English stops significantly more often than English native of voiceless stops by the same group of L1-English speakers as in the speakers. Even though all groups differed from a group of English con­ current study, Gorba and Cebrian (2021) found that English learners of trols, the early learners used prevoicing numerically less frequently than Spanish with experience living in the L2 setting outperformed those with the late learners. Regarding their L1, the Italian speakers were found to no experience in an L2 setting in their production of Spanish /p/ and produce prevoiced L1 stops less often than Italian monolinguals did, /k/, indicating a positive effect of L2 experience on L2 accuracy. The particularly in the case of early bilinguals with a low L1 use – but it same result was replicated by Spanish learners of English (Gorba, 2020). should be noted that the learners and the monolinguals completed However, contrary to Flege (1987), no influence of the L2 on the L1 was different production tasks. Moreover, results showed that the percent found regardless of the amount of L2 experience. The different outcome use of prevoicing in the L1 and in the L2 were positively correlated for all regarding the effect of the L2 on the L1 in the studies reported above groups – i.e., the more they used prevoicing in the L1 the more they used could be explained by the fact that the length of residence of the most it in the L2. Nathan (1987) elicited the production of English stops by experienced group in Gorba and Cebrian (2021) and Gorba (2020) was L1-Spanish speakers at two different times 18 months apart and found considerably shorter than that of the group tested in Flege (1987) – different results for /b/ and /d/ as opposed to /g/. Speakers’ L1 and L2 about 4 years vs. 12, respectively –, as well as other factors, including a production for /b/ and /d/ did not differ significantly at any testing limited use of the L2 – even in the L2 setting – and amount of L2 time, as the L2 learners used prevoicing in both languages. However, in instruction. the case of /g/, the learners were found to use prevoicing significantly As reported above, other factors, such as formal instruction, have more frequently in Spanish (100% both times) than in English (80% and also been found to have a positive effect on the acquisition of voiceless 50%, the first and second time tested, respectively). This result could be stops (e.g., Casillas 2019; Chang 2012, 2013; Nagle 2019; Schuhmann linked to the fact that maintaining the rate of glottal airflow needed to and Huffmann 2015; 2019; Yang et al. 2022; Zampini 1998). For produce voicing involves a greater articulatory effort for a velar occlu­ instance, Zampini (1998) investigated the production of initial bilabial sion than for more anterior articulations (e.g., Ohala 1983). Still, this 95 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 difficulty exists for Spanish velars too. Thus, the results for /g/ may category in the L1 (e.g., short-lag VOT for Spanish and Italian speakers). show that learners were already making a difference between the L1 and Thus, when it comes to the acquisition of L2 prevoiced stops on the part the L2, as an articulatorily marked gesture is maintained in the L1 but a of English learners, improvement has been observed with sufficient more unmarked production emerges in the L2, where prevoicing is not exposure or instruction, (e.g., Casillas 2019; Nagle 2019; Schuhmann crucial.1 and Huffman 2015). However, Spanish and Italian learners of English The studies described above explored the acquisition of the short-lag appear to change the use of these cues in the direction of native speakers vs. long-lag VOT distinction. The reverse pattern, that is, the acquisition to a lesser extent (e.g., Flege and Eefting 1987; Mackay et al. 2001; of prevoicing as a contrastive feature has received comparatively less Nathan 1987). Still, in both cases, learners rarely achieve target-like attention in the literature, although it has been addressed by some values and their productions tend to differ from those of native speakers. recent studies (Baese-Berk, 2019; Casillas, 2019; Dmitrieva et al., 2020; Herd et al., 2015; Huffman and Schuhmann, 2016; Hutchinson and 1.2.2. Comparison of the English and Spanish VOT systems Dmitrieva, 2021; Maye and Gerken, 2000; Nagle, 2019; Schuhmann and In English, the voicing distinction in stops is generally described as a Huffman, 2015, 2019; Yang et al., 2022; Zampini, 1998). Maye and contrast between short-lag and long-lag VOT, particularly in syllable Gerken (2000) and Baese-Berk (2019) studied the perception of the initial position. For example, Lisker and Abramson (1964) measured the contrast between short-lag /t/ and prevoiced /d/ by English speakers. VOT productions of American English speakers and reported that /p/ is Both studies found that after a short perception training – of one session produced with 58 ms of VOT, /t/ with 70 ms and /k/ with 80 ms (see in the case of Maye and Gerken (2000) and of two and three sessions Table 1 for the values reported in several studies). As for voiced stops, regarding Baese-Berk (2019) – the English speakers started to discrim­ most participants were found to use short-lag VOT, although some used inate the two phones, particularly if exposed to a bimodal distribution of voice-lead VOT (average short-lag and voice-lead VOT for /b/ were 1 ms the VOT continuum – i.e., if learners were mainly presented with the and -101 ms for /b/, respectively: 5 ms, and -102 for /d/ and 21 ms and stimuli at the two ends of the continuum. Zampini (1998); Casillas -88 ms for /g/). That is, short-lag VOT and voice-lead VOT do not (2019); Nagle (2019) and Schuhmann and Huffmann (2015) examined contrast, as both are used in the production of voiced stops. the production of Spanish voiced stops by English learners. Zampini The Spanish VOT range is shorter than that of English in terms of the (1998) found that English speaking learners did not produce Spanish /b/ possible realizations of voiced and voiceless stops, which range from accurately, as they presented mean positive VOT values across all testing short-lag VOT for voiceless stops to voice-lead VOT for voiced stops (see sessions, although, as reported above, they produced /p/ with Table 1). For example, Castañeda (1986) found that speakers of Castilian target-like values. Even though the learners appeared to produce English Spanish produced /p/ with a mean VOT of 7 ms, /b/ with -70 ms, /t/ and Spanish /b/ with somewhat numerically different values – being the with 10 ms, /d/ with -78, /k/ with 29 ms and /g/ with -174 ms. It should English VOT values greater – this difference did not reach significance. also be noted that a greater variability in terms of VOT duration has been Conversely, Casillas (2019) found that adult English learners of Spanish found for voiced stops than for voiceless stops, as the duration of pre­ did improve their production of Spanish /b/ throughout the seven-week voicing in voiced stops can vary from relatively very long to relatively immersion course described above. In fact, their production of /b/ short (Lisker and Abramson, 1964; Castañeda, 1986, see Table 1 for significantly shifted towards more target-like values – i.e., longer pre­ ranges). voicing – after the third week of the program and continued to improve Considering these differences between the two languages under during the following weeks. However, their VOT use differed from that study, it can be assumed that the acquisition of L2 voiced stops for En­ of English-Spanish bilinguals, as they presented significantly less pre­ glish learners of Spanish and for Spanish learners of English will involve voicing. It should also be noted that, even though there was some different processes. In the case of the former, learners will need to learn improvement in their production of /b/ – i.e., learners presented more that the use of prevoicing is distinctive in Spanish, and that short-lag target-like VOT at the end of the course –, their improvement with /p/ VOT signals a different category. As for the Spanish learners, the was greater and occurred earlier in the course. Similarly, Nagle (2019) opposite process needs to take place: they will have to learn that short- found that beginning L1-English learners of Spanish improved their pronunciation of voiced stops over the course of two semesters, and that Table 1 the greatest improvements took place during the first half of the study. VOT means in ms for initial stops in English and Spanish reported in some Schuhmann and Huffmann (2015) observed that after receiving pho­ previous studies. Standard deviations (single values) and ranges (range of netic instruction, English learners of Spanish produced Spanish stops values) are given within parentheses, as reported by each study. with more prevoicing. However, the difference between the learners’ L1 Language/Study /p/ /b/ /t/ /d/ /k/ /g/ and L2 stops approximated – but did not reach – statistical significance. English As for the L1, although the English learners of Spanish presented British English ( 42 15 (8) 65 21 (6) 62 27 (11) numerically lower VOT values at the end of the program than before Docherty, (10) (13) (14) receiving training, this difference was also not significant. Interestingly, 1990) both before and after phonetic training, participants who prevoiced the American 58 1/-101 70 (30 5/-102 80 21/-88 most in Spanish also did so in English. Baese-Berk (2019) also found English ( (20 – (0 – 5 / – 105) (0 – 25 / (50 – (0 – 35/ Lisker and 120) -130 – -155 – 135) -150 – that, when trained on both perception and production, the participants Abramson, -20) -40) -60) improved their production of voice-lead VOT, although their perception 1964) improved to a lesser extent than when trained on perception only. Spanish In short, L2 learners may be able to improve their production of Castilian 7 (0 – -70 10 (5) -78 (26) 26 -58 Spanish ( 24) (-166 – (11) (-132 – voiced stops, although generally to a lesser extent than voiceless stops Castañeda, -24) -16) (Nagle, 2019; Dmitrieva et al., 2020; Gorba, 2020; Schuhmann and 1986) Huffman, 2019, but cf. Herd et al., 2015; Hutchinson and Dmitrieva, Castilian 13 (4 -92 14 -92 29 -74 2021). Moreover, the acquisition of the voiced categories may be Spanish – 49) (-173 – (6.2 – (-188 – (10 (-165 – affected by whether the main cue to voicing is available in the L1 as an (Rosner et al., -23) 30.5) -28) –63) -14 2000) additional albeit non-predominant cue to the same category (e.g., Puerto Rican 4 (0 – -138 9 (0 - -110 29 -108 voice-lead VOT for English speakers), or if the L2 cue signals a different Spanish ( 15) (-235 – 15) (-170 – (15 – (-165 – Lisker, and -60) -75) 55) 45) Abramson, 1 1964) We are grateful to an anonymous reviewer for pointing out this fact. 96 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 lag and voice-lead do not contrast and that short-lag VOT, which in stay in an L2 setting. Percent weekly L2 use was drawn from the par­ Spanish signals voiceless stops, is typically characteristic of voiced stops ticipants’ self-reported amount of L2 use in different contexts (i.e., at in English, particularly in stressed initial position, in opposition to long- work/school, with family and with friends). The inexperienced English lag VOT. Note, however, that using prevoicing to produce English voiced group (EnInexpUK) consisted of 11 L1-English students of Spanish who stops would not result in intelligibility problems, since, as mentioned were enrolled in the first or second year of Hispanic Studies or in the above, that is a possible albeit less frequent way to produce them. highest levels of Spanish language courses at a language center at QMUL. They had never lived in a Spanish-speaking country, although some reported short holiday visits, and were living in the UK at the time 1.3. The present study of testing. EnInexpUK reported using Spanish 12.3% on a weekly basis and had been learning Spanish for a mean of 3.8 years. The moderately The current study presents two experiments which investigate the experienced English group (EnExpUK) were fourth-year students of acquisition of L2 voiced stops by two populations that are mirror-image Hispanic Studies at QMUL and had spent between a term and a year (M in terms of L1 and L2. Specifically, the acquisition of /b/ and /g/ on the = 9.4 months) in a Spanish country the previous academic year. They part of English learners of Spanish (experiment 1) and Spanish learners were back in the United Kingdom at the time of testing. EnExpUK had of English (experiment 2) varying in L2 experience will be examined. studied Spanish for 7.5 (SD: 5.3) years and reported a weekly use of The coronal stop /d/ is not investigated, as it is produced in a different Spanish of 15%. The most experienced English group in terms of time place of articulation in the two languages under study – i.e., dental in spent in an L2-speaking setting (EnExpSp) had been living in Spain – Spanish and alveolar in English (Lisker and Abramson, 1964). The where they resided at the time of testing – for a mean of 50.7 months – i. learners’ production of both L1 and L2 stops will be examined in order to e., 4.2 years. In spite of having resided in an immersion setting for the assess if learners produce L2 stops with different values from L1 stops, if longest period of time, EnExpSp had received the least amount of L2 the learning of the L2 stops has any effect on the L1 categories, and if L2 instruction (M = 2.4). Moreover, their percent weekly L2 use was very experience has an effect on these issues. The main research questions are similar to that of the English learners in an instructional setting (16.3 thus the following: %). The Spanish control group was made up of 10 students of History and Spanish Studies at Universitat Autònoma de Barcelona (Spain) who (1) What is the effect of L2 experience on the production of L2 voiced were functional monolingual speakers of Spanish. They all had studied stops? some English at school, as English is a mandatory course in the Spanish (2) Does the learning of L2 voiced stops have an effect on the pro­ primary and secondary education curricula, but all reported not using it. duction of L1 stops? If so, is this effect modulated/affected by L2 experience? 2.1.2. Task and procedure Participants had to complete a sentence reading task in both lan­ Furthermore, given that two different populations are tested, namely guages (see Appendix A for the complete list)3. Control groups per­ L1-English L2-Spanish speakers and L1-Spanish L2-English speakers, the formed the task only in their L1. The reading task consisted of a list of 32 design of the study allows for comparisons between the two. This carrier sentences – for each language – that presented bilabial and velar comparison is of particular theoretical interest since, even though the stops in word-initial position, as well as fillers. All words were disyllabic two languages under study share a distinction between voiced and and carried stress on the penultimate syllable and all target stops were voiceless stops, each language makes use of a different distribution of followed by vowel /i/. Each sentence was repeated twice and a total of VOT duration to signal the contrast, as illustrated above (see Section 20 productions of voiced stops – 10 per segment – were elicited for each 1.2.2) participant and language. It was decided to use a carrier sentence reading task in order to be able to control the position in which the 2. Experiment 1. English learners of Spanish target phones appeared, since Spanish stops in medial position – except after nasal – are realized as approximants (Colina 2020; Hualde 2005). 2.1. Methodology The carrier sentence in the English task was ‘X is the next word’ and the Spanish equivalent – i.e., ‘X es la siguiente palabra’ – was used in the 2.1.1. Participants Spanish condition. Order of completion of the English and Spanish A total of 50 participants completed this study, including a control reading tasks was counterbalanced. Furthermore, participants group of functional monolinguals for English and Spanish, respectively, completed an online linguistic background questionnaire based on the and two groups of English learners of Spanish. The English control group Bilingual Language Profile (Birdsong et al., 2012) before the testing (EnMono) were nine functional English monolinguals who were session. enrolled in undergraduate or graduate studies at Queen Mary University The recordings made in an English setting took place in a soundproof of London (QMUL) and were mostly from the London metropolitan area room at QMUL and a Neumann TLM 103 microphone, a booth mixer (EnMono)2 and 10 Spanish functional monolinguals constituted the Steinberg UR22 MKII and the software Audacity Team, 2016 were used. Spanish control group (SpMono). Table 2 summarizes the main char­ Most of the participants tested in Spain were recorded at the speech acteristics of each group, including sex, L1, L2, months spent in an L2 laboratory at Universitat Autònoma de Barcelona (UAB) in an acousti­ setting, location, percent weekly L2 use and years of L2 instruction. The cally treated room using an Audio-Technica AT 2050 microphone and an information was retrieved using a linguistic background questionnaire Alesis Multimix 8 mixer. For participants’ convenience, a few recordings based on the Bilingual Language Profile (BLP, Birdsong et al., 2012), were made in a different institution in Barcelona, namely Universitat de which also included additional questions regarding the recency of the 2 3 There were some dialectal differences between the English speakers, The data presented in this paper are part of a larger study that tested the particularly in the group tested in Spain due to the difficulty recruiting vol­ perception and production of stops by Spanish learners of English and English unteers. Four participants were from Southern England, one from Northern learners of Spanish. In the production task, which was a carrier sentence England, one from Scotland, two from Ireland, one from Canada, one from the reading task, the production of both voiced and voiceless stops was collected East Coast of the United States and one from the South of the United States. (see Appendix A for the complete list of sentences). Participants also had to However, these dialectal differences were not expected to affect the L1 pro­ complete two identification tests (one for the /p/-/b/ contrast and one for the duction of voiced stops, and, in fact, no significant differences were observed /k/-/g/ contrast) in both languages (see Gorba 2019 and Gorba and Cebrian between groups (see results section). 2021). 97 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 Table 2 Characteristics of the five groups that participated in Experiment 1: number of participants (N), gender, L1, L2, months in L2 setting, location at the time of testing, % weekly L2 use and years of L2 instruction (inst.). Standard deviations are provided in parentheses. Group N Sex L1 L2 Months in L2 setting Location % weekly L2 use Years of L2 inst. EnMono 9 4 F; 5 M English NA NA UK - - EnInexpUK 11 6 F; 5 M English Spanish none / minor UK 12.3% (8) 3.8 (3.3) EnExpUK 9 4 F; 6 M English Spanish 9.4 (4.2) UK 15% (6.5) 7.5 (5.3) EnExpSp 11 6 F; 5 M English Spanish 50.7 (27.3) Spain 16.3% (10.5) 2.4 (2.9) SpMono 10 4 F; 6 M Spanish NA NA Spain - - Barcelona, in a soundproof booth using a Shure-SM58 microphone and a different models were very consistent and the levels of significance for Marantz PMD-660 digital recorder. All sound files were recorded in each factor and pairwise comparison were very similar across models. WAV format. SPSS software (IBM Corp, 2017) was used. Finally, correlation tests were conducted to determine whether the use of prevoicing in the two lan­ 2.1.3. Data analysis guages was related. A total of 2780 target words were analyzed in Praat (Boersma and Weenik, 2016). Some productions were disregarded because they were 2.2. Results produced with a sound other than a stop – i.e., an approximant or a fricative – or because VOT was difficult to measure due to the quality of 2.2.1. Production of English voiced stops the recording at the point where the stop was produced – e.g., due to First, the production of the L1-English groups in English was exam­ overlapping noise. The VOT of the remaining words, a total of 2756 ined. Tables 3 and 4 present the mean VOT and the proportion and (English /b/: 689 words; English /g/: 682 words; Spanish /b/: 696 percentage of prevoiced productions for /b/ and /g/, respectively, for words; Spanish /g/: 689 words), was measured manually. Short-lag VOT each group. Given the bimodal distribution of VOT durations described productions were measured from the onset of the burst to the beginning for English voiced stops (Lisker and Abramson, 1964), the tables also of voicing as indicated by the periodicity of the signal, considering both show the mean VOT for short-lag productions and for voice-lead pro­ the spectrogram and the oscillogram. Measurements of voice-lead pro­ ductions separately. As expected, most voiced stops were produced with ductions were made from the beginning of the prevoicing until the onset short-lag VOT, but instances of prevoicing were found in all groups. of the burst. Regarding /b/, EnInexpUK had the highest average VOT (8 ms), fol­ Two different measures were obtained, namely a continuous mea­ lowed by EnMono and EnExpSp (close to 0) and EnExpUK a mean sure (VOT duration) and a categorical measure (the presence or the negative VOT (-9 ms). The two groups with the longest length of resi­ absence of prevoicing). Most previous studies investigating voiced stops dence also showed the largest ranges of VOT, and the mean durations of have used VOT duration as the measure of analysis – with the exception voice-lead VOT were also the greatest (EnExpUK: -90; EnExpSp: -76). A of MacKay et al. (2001), which used percentage of prevoiced pro­ similar pattern of results was found with respect to the percentage of ductions, and, to our knowledge, none of them used a categorical vari­ prevoicing, with EnInexpUK having the lowest amount (1%) and able (presence of prevoicing: yes/no). Given the great variability often EnExpUK the highest (18%). found in the duration of prevoicing (Castañeda, 1986; Lisker and Similar results were found for /g/ (see Table 4). EnInexpUK pre­ Abramson, 1964), a decision was made to use the categorical measure in sented the greatest VOT value (20 ms), followed by EnMono (14 ms), the statistical analysis4. Therefore, although descriptive results report whereas EnExpSp and EnExpUK had the lowest VOT means (6 and 1 ms, both VOT means and percentages of prevoiced productions, the latter is respectively). As for the percentage of prevoiced /g/, EnExpUK pre­ the one used in the statistical analysis throughout the paper. Thus, for sented the greatest amount of prevoiced tokens (26%), whereas EnI­ each stop produced by each speaker in each language a negative VOT nexpUK prevoiced the least (5%). value was considered an instance of prevoicing, whereas a VOT of 0 ms or greater was coded as an instance of short-lag VOT (i.e., as lack of Table 3 prevoicing). In order to examine differences between groups in their L1 Mean VOT and range of all productions, mean VOT of short-lag and voice-lead and L2 productions, the results were submitted to a series of generalized productions, and proportion and percentage (in parentheses) of prevoicing in linear mixed models (GLMM) with presence of prevoicing (yes/no) as the production of English /b/ by English learners of Spanish and English the dependent variable, and group (three learner groups and the monolinguals. Standard deviations are given in parentheses. monolingual control group) and segment (/b/ and /g/) as independent Group /b/ VOT /b/ /b/ mean /b/ mean Proportion variables. Separate analyses were carried out for each language – En­ mean VOT short-lag voice-lead prevoiced glish and Spanish. In addition, between-language differences for each L1 range VOT VOT population (i.e., differences in the production of voiced stops in the L1 EnMono 1 (25) -94 – 20 7 (4) -69 (26) 9/90 (10%) and in the L2) were also explored by means of a series of GLMMs with EnInexpUK 8 (6) -47 – 24 8 (4) -47 1/110 (1%) language (English and Spanish), group (the three learner groups for each EnExpUK -9 (40) -146 – 9 (4) -90 (31) 16/90 (18%) L1 population) and segment (/b/ and /g/), and their interactions, as 23 EnExpSp 0 (35) -160 – 7 (10) -76 (36) 17/110 (15%) fixed factors. Different combinations of random intercepts and slopes for 19 participant and word were also examined in the models. The selection of the best model was based on the lowest Akaike Information Criterion (AIC) and on model convergence. In fact, in every case, the results for the The results of a GLMM with presence (yes/no) of prevoicing as the dependent variable, group, segment and the interaction as fixed factors and a random intercept for participant revealed a significant effect of 4 In fact, a parallel statistical analysis involving a series of linear mixed group [F(3, 799) = 3.521; p = .015], but no effect of segment [F(1, 799) models with VOT duration as the dependent variable was also conducted. The = 2.437; p = .119], and no interaction [F(3, 799) = 1.065; p = .363]. two sets of analyses – on VOT duration or on presence/absence of prevoicing – yielded very similar results, the only difference being that more cases of sig­ nificant group differences – particularly between groups sharing the same L1 – were revealed in the prevoicing analysis. 98 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 Table 4 Table 6 Mean VOT and range of all productions, mean VOT of short-lag and voice-lead Mean VOT and range of all productions, mean VOT of short-lag and voice-lead productions, and proportion and percentage (in parentheses) of prevoicing in productions, and proportion and percentage (in parentheses) of prevoicing in the production of English /g/ by English learners of Spanish and English the production of Spanish /g/ by English learners of Spanish and Spanish monolinguals. Standard deviations are given in parentheses. monolinguals. Standard deviations are given in parentheses. Group /g/ /g/ /g/ mean /g/ mean Proportion Group /g/ /g/ /g/ mean /g/ mean Proportion VOT VOT short-lag voice-lead prevoiced VOT VOT short-lag voice-lead prevoiced mean range VOT VOT mean range VOT VOT EnMono 14 (38) -255 – 23 (7) -93 (77) 7/90 (8%) SpMono -77 (29) -138 – 20 (3) -79 (26) 97/99 (98%) 43 22 EnInexpUK 20 (25) -109 – 23 (8) -84 (28) 5/110 (5%) EnExpSp -29 (48) -123 – 25 (9) -61 (30) 68/110 (62%) 45 51 EnExpUK 1 (48) -129 – 27 (9) -77 (26) 23/90 (26%) EnExpUK -55 (51) -175 – 22 (10) -75 (35) 67/85 (79%) 58 43 EnExpSp 6 (41) -137 – 24 (8) -75 (28) 20/110 (18%) EnInexpUK -6 (52) -198 – 25 (9) -72 (43) 34/107 (32%) 49 48 Fisher’s LSD5 pairwise comparisons exploring group differences showed intercept for participant also revealed a significant effect of group [F(3, a significant difference between the groups with experience and the least 799) = 10.916, p = .000], no effect of segment [F(1, 799) = 3.008, p = experienced group, that is, between EnExpSp (mean = 17%, SD = 37) .083] and no interaction [F(3, 799) = 0.449, p = .718]. Pairwise com­ and EnInexpUK (mean = 3%, SD = 16), p = .039, and between EnExpUK parisons with a Bonferroni correction indicated that EnInexpUK (mean (mean = 22%, SD = 41) and EnInexpUK, p = 031. EnInexpUK was the = 29%, SD = 45) differed significantly from SpMono (mean = 98%, SD group that prevoiced the least often, but none of the groups differed = 14), p < .001, and EnExpSp (mean = 58%, SD = 49) also differed significantly from EnMono. significantly from SpMono (p = .011). By contrast, the difference be­ tween SpMono and EnExpUK (mean = 71%, SD = 46) approximated but 2.2.2. Production of Spanish voiced stops did not reach significance (p = .077). Moreover, EnExpUK were found to As for the L2 production of Spanish stops, as expected, the English use prevoicing in Spanish significantly more often than EnInexpUK (p = groups prevoiced less often than the Spanish monolinguals and, conse­ .001). quently, presented greater VOT means (see Tables 5 and 6). Regarding /b/, EnExpUK, with a mean VOT of -48 ms, was the group that presented 2.2.3. L1 vs. L2 production of voiced stops the closest mean VOT duration to SpMono’s (-78 ms), and prevoiced /b/ The production of voiced stops in each of the languages under study the most often (64%). EnExpSp presented a mean VOT for /b/ of -33 ms was also compared within each group. Figs. 1 and 2 show the distribu­ and prevoiced /b/ 56% of time, while EnInexpUK presented the closest tion of VOT durations for each learner group in each language. All En­ mean VOT to 0 ms, with -13 ms, and prevoiced it the least (26%). A glish groups presented either longer positive VOT values or shorter similar pattern was followed in the case of /g/. EnExpUK performed the voice- lead VOT values in English than in Spanish for the production of most similarly to SpMono – with a mean VOT of -55 ms and 79% of /b/ and /g/, although the difference was greater in some cases than in prevoiced tokens – followed by EnExpSp – with a mean VOT of -29 ms others. EnInexpUK used a smaller range of values and produced gener­ and 62% of prevoiced /g/ –, whereas EnInexpUK presented the closest ally shorter voice-lead VOT durations, whereas EnExpUK displays the mean VOT to 0 ms (-6 ms) and prevoiced the least (32% of time). greater differences between L1 and L2 and shows greater variability in In this case, the results of a GLMM on amount of prevoicing with L1. group, segment and their interaction as fixed factors and a random As described above (see Section 2.1.3), a series of generalized linear mixed models were conducted with presence or absence of prevoicing as Table 5 the dependent variable. Fig. 3 below presents the mean percentages of Mean VOT and range of all productions, mean VOT of short-lag and voice-lead use of prevoicing per group and language. The results of a GLMM that productions, and proportion and percentage (in parentheses) of prevoicing in included group, language, segment, group x language and segment x the production of Spanish /b/ by English learners of Spanish and Spanish language as fixed factors and a random intercept for participant revealed monolinguals. Standard deviations are given in parentheses. Group /b/ /b/ /b/ mean /b/ mean Proportion VOT VOT short-lag voice-lead prevoiced mean range VOT VOT SpMono -78 (29) -137 – 7 7 (0) -80 (26) 95/97 (98%) EnExpSp -33 (47) -118 – 7 (19) -70 (26) 61/110 (56%) 119 EnExpUK -48 (52) -203 – 6 (10) -82 (34) 58/90 (64%) 49 EnInexpUK -13 (45) -207 – 10 (4) -78 (47) 28/109 (26%) 22 5 Fisher’s Least Significant Difference (LSD) test was chosen because, despite the significant effect of group, Bonferroni-adjusted pairwise comparisons did not reach significance and it was relevant to explore what group differences drove the effect of group. Although the LSD does not correct for multiple comparisons, it is deemed appropriate for comparisons involving less than five levels Derek and DeMars (2019). In the remaining of the analyses, Fig. 1. Boxplot of VOT production (in ms) for English and Spanish /b/ by the Bonferroni-adjusted pairwise comparisons were applied. L1-English speakers. 99 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 no group differences were observed in the production of L1-English stops, which was comparable to that of monolingual English speakers, whereas group differences emerged in the production of L2-Spanish stops. Regarding the production of L2 voiced stops, both EnInexpUK – the least experienced group – and EnExpSp – the group with the longest length of residence in the target-language setting – were found to differ from SpMono, whereas EnExpUK did not differ from the Spanish con­ trols. In other words, the groups with the least and the most experience living in an L2 setting produced Spanish voiced stops differently from Spanish controls, whereas the group with intermediate experience in an L2 setting – but a greater amount of L2 instruction – presented the most target-like VOT values. Thus, it appears that length of residence in an L2 setting alone does not account for the differences observed between the English groups. Despite residing in an immersion setting, EnExpSp did not report a notably greater amount of L2 use than the other L1-English groups (12% and 15% for EnInexpUK and EnExpUK, respectively, vs. 16% for EnExpSp). EnExpSp’s relatively small L2 use may be related to the fact that they used English at work, as they were English teachers, Fig. 2. Boxplot of VOT production (in ms) for English and Spanish /g/ by the and that most of them had English-speaking partners. This finding is in L1-English speakers. line with the fact that Flege and Bohn’s (2021) SLM-r underscores the importance of language use and proposes measuring L2 experience in terms of a composite measure combining length of residence and amount of L2 use (Full Time Equivalent or FTE), which is calculated by multiplying the number of years spent in an L2 setting by the percentage of weekly L2 uses. Although the current study was not designed to evaluate specifically the effect of FTE, the data in Table 1 above allow us to compare the groups in terms of this variable. EnExpSp and EnExpUK did not differ greatly in FTE of target language input (0.12 vs 0.69, respectively)6 despite the former’s much longer stay in the L2 setting. Moreover, the fact that EnExpUK outperformed EnExpSp in Spanish voiced stop production may be related to amount of L2 instruction and student status. Recall that EnExpUK were the English group that had received the most L2 instruction (7.5 years, vs. 2.4/3.8, see Table 2 above). Furthermore, EnExpUK were in fact enrolled in Spanish classes at the time of testing and at the institution where the testing took place. This may have triggered more careful productions, as students may have Fig. 3. Mean use of prevoicing in the production of English and Spanish stops interpreted the experiment task as an academic test. By contrast, in spite by the L1-English speakers. The dashed line and the dotted line show the mean of having lived in an L2 setting for the longest period of time, EnExpSp values for SpMono and EnMono, respectively. Percentage use of prevoicing and had received scarce – sometimes none – formal instruction in Spanish standard error of the mean are given. (mean = 2.4 years) and were not studying Spanish at the time of testing. Thus, the current results add to the findings from previous studies an effect of group [F (2, 1223) = 7.4; p = .001], language [F(1, 1223) = showing a beneficial effect of formal instruction on voiced stop pro­ 173.9; p < .001] and segment [F(1, 1223) =10.032; p = .002], but no duction (e.g., Casillas 2019; Lord 2010; Nagle 2019; Schuhmann and significant interaction (language x group: ([F(2, 1223) = 0.483; p = Huffmann 2015). It is possible, thus, that a greater amount of L2 in­ .617]; language x segment: [F(1, 1223) = 0.08; p = .777]). Thus, all struction had resulted in more target-like productions of Spanish stops groups produced voiced stops with more prevoicing in Spanish (mean = on the part of EnExpUK compared to EnExpSp. 51, SD = 50) than in English (mean = 13, SD = 34), and prevoicing was With respect to the production of English voiced stops, none of the somewhat more frequent with /g/ (mean = 35, SD = 48) than with /b/ groups differed from EnMono statistically. This finding is in line with (mean = 29, SD = 45). Regarding the effect of group, Bonferroni- Gorba and Cebrian (2021), a study involving the same population which adjusted pairwise comparisons indicated that both EnExpSp (mean = did not find an effect of the L2 on the L1 production of voiceless stops. 38%, SD = 48) and EnExpUK (mean = 46%, SD = 50) used prevoicing The authors argued that a greater amount of L2 experience and L2 use more frequently (across languages) than EnInexpUK (mean = = 16%, may be necessary for an L1 phonetic drift towards the L2 to occur, an SD = 36), p = .027 and p = .004, respectively) explanation that may also apply to the present study. A novelty effect Finally, in order to explore whether the learners’ production in the similar to that found in Chang (2012, 2013) would not apply to the L1 and the L2 were related, the relationship between the use of pre­ present study, as none of the groups were absolute beginners in an im­ voicing in L1 and L2 voiced stops was examined by conducting a one- mersion setting. Still, although EnExpUK and EnExpSp did not differ tailed Pearson correlation for each stop. The use of prevoiced stops in significantly from the English controls in their use of prevoicing in the the L1 and in the L2 was significantly, albeit moderately, correlated for L1, the two more experienced groups made a greater use of prevoicing both /b/ (r = .359, N = 31, p = .024) and /g/ (r = .479, N = = 31; p = than EnInexpUK (see Fig. 3) and this difference reached significance. .002). This may indicate that learning to use more prevoicing in the L2 may carry with it a greater use of that feature in the L1, providing some 2.3. Interim discussion Experiment 1 investigated the production of L2 and L1 /b/ and /g/ 6 Flege and Bohn’s (2021) exemplifications of differences in FTE between by English learners of Spanish. Results showed that, as expected, the experienced vs inexperienced groups are notably greater, e.g., 4.3 vs 0.3 (Flege learners presented greater VOT values in English than in Spanish. In fact, et al., 1997) or 17.2 vs 9.2 (Flege, 1991), respectively. 100 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 indication of a possible effect of the L2 on the L1. As for the inexperi­ parallel to the English learners of Spanish in Experiment 1 in terms of L2 enced group, even though they used prevoicing numerically less than experience, and the same two control groups as in Experiment 1 the English native speakers, there is no evidence for category dissimi­ (SpMono and EnMono).7 The inexperienced Spanish learners lation, as this difference was numerically very small (EnMono: /b/ = (SpInexpSp) were 10 undergraduate students in their first or second year 10%, /g/ = 8% prevoiced productions; EnInexpUK: /b/ = 1%, /g/ = 5% of English Studies at UAB who had never lived in an immersion setting. prevoiced productions) and did not reach significance (c.f., Dmitrieva They used English 22.7% of time on a weekly basis and had learnt it for et al., 2020). The difference between the L1 and L2 production of voiced 13.2 years. The moderately experienced Spanish learners of English stops was also assessed. All groups were found to use prevoicing less (SpExpSp) were 10 students in their fourth year of English studies in the often in English than in Spanish, although this difference was greatest in same institution who had spent 7.4 months in an immersion setting the the case of EnExpUK and smallest regarding EnInexpUK. That is, all previous academic year as part of a study-abroad program. SpExpSp English learners of Spanish were able to distinguish L1 and L2 voiced reported to use English 26.3% of the time weekly and had been learning stops in production – to a greater or lesser extent. A previous study it for 13.9 years. Finally, the highly experienced Spanish group (SpEx­ looking at English learners of Spanish, Zampini (1998), also found that pUK) was made up of nine L1-Spanish speakers who had been living – L2 learners made a numerical difference between their L1 and L2 pro­ and still resided – in the UK for 47.9 months – i.e., four years. SpExpUK duction of voiced stops, in this case regarding VOT values, but this were the group that used the L2 the most (54.6% weekly) and had learnt difference was not significant. It is possible that methodological differ­ it for 13.4 years. Even though the L1-English and the L1-Spanish groups ences between the two studies may explain the different result, as were comparable in terms of length of residence and setting, there were Zampini’s study examined a smaller amount of production data and considerable differences regarding amount of L2 instruction and L2 use involved participants who had no experience living in the L2 setting. that stem from the greater presence of English in the Spanish education Finally, it was investigated whether there was a relationship between system than of Spanish in the UK and in the media and entertainment the use of prevoicing in the L1 and in the L2. Given that the use of industry. prevoicing is subject to individual and free variation in English, it was hypothesized that the use voice-lead VOT in their L2 may be related to 3.1.2. Tasks, procedure and data analysis its use in the L1. The correlations conducted on the use of prevoicing in Participants had to complete the same tasks that were presented in the L1 and L2 indicated that those English learners that used prevoicing Experiment 1 and the data was analyzed following the same procedure. the most frequently in English also did so in Spanish. As a matter of fact, at a group level, EnExpUK used prevoicing the most in both languages. 3.2. Results This outcome is in agreement with some previous studies that showed that the use of negative VOT in the L1 and the L2 were correlated, and, 3.2.1. Production of Spanish voiced stops thus, that the use of prevoicing in the L1 was overall transferred to the L2 The L1 speakers of Spanish, as expected, produced most voiced stops (MacKay et al., 2001; Schuhmann and Huffmann, 2015). That is, even with voice-lead VOT, although a few instances of short-lag productions though the English learners of Spanish were able to distinguish L1 and were found in all groups (see Tables 8 and 9). All groups showed a L2 stops by producing shorter voice-lead VOT and longer short-lag VOT similar mean duration of voice-lead VOT, ranging from a mean of -64 in English than in Spanish, the use of prevoicing might have been – and -68 ms for SpInexpSp and SpExpUK, respectively, to a mean of -77 positively – transferred from the L1 to the L2 to a certain extent, and, and -80 ms for SpMono and SpExpSp, respectively. Regarding the per­ thus, facilitated the production of target-like – prevoiced – voiced stops. centage of prevoiced productions, results were also relatively similar In sum, a positive effect of L2 experience was found on the produc­ across groups. SpExpSp produced almost exclusively prevoiced stops tion of L2 stops, as those learners with some experience in the L2 setting (99%), whereas SpExpUK had the lowest percent of prevoicing (88%). outperformed those with none. However, other factors, including longer Regarding /g/, SpExpSp presented the longest mean voice-lead VOT L2 instruction on the part of EnExpUK and insufficient amount of L2 use (-80 ms), followed by SpMono and SpExpUK (-77 ms), whereas in the case of EnExpSp, may have played a role, as the former out­ SpInexpSp presented a somewhat shorter mean (-71 ms). As for the performed the latter. No clear evidence of L1 phonetic drift towards the percentage of prevoiced tokens, SpInexpSp prevoiced the most (100%), L2 were found regardless of amount of L2 experience, as no group whereas SpExpUK prevoiced the least (93%). These results were sub­ differed from the monolingual speakers in their L1 production. Still, the mitted to a series of generalized linear mixed models with presence of greater use of prevoicing on the part of the two more experienced prevoicing as the dependent variable, following the same procedure as groups, both in L1 and L2, points to a possible influence of the L2 on the described for Experiment 1 (see section 2.1.3). A GLMM with group, L1. In addition, all groups seemed to make a difference between L1 and segment and their interaction as fixed factors and a random intercept for L2 stops by using prevoicing more often in the L2. It remains to be seen if participant revealed no significant main effects and no interaction the current results are specific to the population under study (L1-English (group: [F(3, 766) = 1.737, p = .158]; segment: F(1, 766) = 0.001, p = learners of L2- Spanish) or generalizable to other populations. This issue .98]; group x segment: F(3, 766) = = 0.165, p = .92]). Thus, on the is explored in Experiment 2, which examines the production of L1 and whole, experience with English did not seem to affect the use of pre­ L2 voiced stops by Spanish learners of English. voicing in the L1, although SpExpUK was the group to display the numerically lowest percent use of prevoicing, particularly in the case of 3. Experiment 2. Spanish learners of English /b/. 3.1. Methodology 3.2.2. Production of English voiced stops The same analysis as in the case of Spanish was conducted for the 3.1.1. Participants production of English voiced stops. All groups tended to produce voiced This study involved 49 participants (see Table 7), including 29 L1- stops with prevoicing more often than with short-lag VOT (see Tables 10 Spanish L2-English speakers, who were distributed in groups that were and 11). Regarding /b/, SpExpUK presented the shortest mean amount 7 All L1-Spanish participants were speakers of Castilian Spanish. Most of them were also speakers of Catalan, but the knowledge of this language was not expected to have any effect on the production of VOT, as Spanish and Catalan present identical VOT values (Julià i Muné, 1981). 101 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 Table 7 Characteristics of the five groups that participated in Experiment 2, including number of participants (N), gender, L1, L2, months in L2 setting, location at the time of testing, % weekly L2 use and years of L2 instruction (inst.). Standard deviations are provided in parentheses. Group N Sex L1 L2 Months in L2 setting Location % weekly L2 use Years of L2 inst. EnMono 9 4 F; 5 M English NA NA UK - - SpExpUK 9 5 F; 4 M Spanish English 47.9 (23.3) UK 54.6% (24.6) 13.4 (2.3) SpExpSp 10 7 F; 3 M Spanish English 7.4 (6) Spain 26.3% (10) 13.9 (1.9) SpInexpSp 10 6 F; 4 M Spanish English none / minor Spain 22.7% (9.7) 13.2 (2.2) SpMono 10 4 F; 6 M Spanish NA NA Spain - - Table 8 Table 11 Mean VOT and range of all productions, mean VOT of short-lag and voice-lead Mean VOT and range of all productions, mean VOT of short-lag and voice-lead productions, and proportion and percentage (in parentheses) of prevoicing in productions, and proportion and percentage (in parentheses) of prevoicing in the production of Spanish /b/ by Spanish learners of English and Spanish the production of English /g/ by Spanish learners of English and English monolinguals. Standard deviations are given in parentheses. monolinguals. Standard deviations are given in parentheses. Group /b/ /b/ /b/ mean /b/ mean Proportion Group /g/ /g/ /g/ mean /g/ mean Proportion VOT VOT short-lag voice-lead prevoiced VOT VOT short-lag voice-lead prevoiced mean range VOT VOT mean range VOT VOT SpMono -78 (29) -137 –7 7 (0) -80 (26) 95/97 (98%) EnMono 14 (38) -255 – 23 (7) -93 (77) 7/90 (8%) SpInexpSp -64 (36) -137 – 10 (2) -72 (25) 94/100 (94%) 43 69 SpExpUK -24 (53) -132 – 27 (11) -68 (24) 50/90 (55%) SpExpSp -80 (26) -135 –3 3 -81 (25) 99/100 (99%) 65 SpExpUK -68 (37) -126 – 8 (4) -79 (19) 79/90 (88%) SpExpSp -49 (41) -127 – 15 (6) -62 (30) 80/100 (80%) 19 59 SpInexpSp -50(46) -166 – 26 (11) -66 (31) 83/100 (83%) 46 Table 9 Mean VOT and range of all productions, mean VOT of short-lag and voice-lead /g/ 80% of time and presented a mean voice-lead VOT of -49 ms, and productions, and proportion and percentage (in parentheses) of prevoicing in SpInexpSp prevoiced /g/ 83% of time and produced it with the longest the production of Spanish /g/ by Spanish learners of English and Spanish mean negative VOT (-50 ms). monolinguals. Standard deviations are given in parentheses. In this case, the results of a GLMM on presence (yes/no) of prevoicing Group /g/ /g/ /g/ mean /g/ mean Proportion as the dependent variable, with group, segment and their interaction as VOT VOT short-lag voice-lead prevoiced fixed factors and a random intercept for participant revealed a signifi­ mean range VOT VOT cant effect of group [F(3, 742) = 20.95, p < .001], but no effect of SpMono -77 (29) -138 – 20 (3) -79 (26) 97/99 (98%) segment [F(1, 742) = 0.868, p = .352] and no interaction [F(3, 742) = 22 0.008, p = .967]. All three Spanish groups (SpExpUK: mean = 58%, SD SpInexpSp -71 (23) -128 – - -71 (23) 100/100 -28 (100%) = 50, SpExpSp: mean = 85%, SD = 35, SpInexpSp: mean = 86%, SD = SpExpSp -80 (51) -175 – 11 -81 (32) 97/98 (99%) 34) used prevoicing significantly more frequently than EnMono (mean 11 = 9%, SD = 29), as confirmed by Bonferroni adjusted pairwise com­ SpExpUK -77 (40) -239 – 18 (5) -84 (31) 84/90 (93%) parisons (p < .001 in all three cases). In addition, SpExpUK were also 25 found to use prevoicing less frequently than the other two Spanish groups (p = .021 in both cases). Table 10 3.3. L1 vs. L2 production of voiced stops Mean VOT and range of all productions, mean VOT of short-lag and voice-lead The results obtained for the production of Spanish and English stops productions, and proportion and percentage (in parentheses) of prevoicing in by the Spanish learners of English were compared. All Spanish groups the production of English /b/ by Spanish learners of English and English presented mean negative VOT values for both English and Spanish monolinguals. Standard deviations are given in parentheses. voiced stops, although the voice-lead VOT duration tended to be Group /b/ /b/ /b/ mean /b/ mean Proportion numerically shorter in English than in Spanish for all groups, particu­ VOT VOT short-lag voice-lead prevoiced mean range VOT VOT larly in the case of the most experienced learners (see Figs. 4 and 5). As done in experiment 1, the results involving the presence (yes/no) EnMono 1 (25) -94 – 20 7 (4) -69 (26) 9/90 (10%) of prevoicing (presence or absence of prevoicing for each production) SpExpUK -43 (49) -182 – 8 (4) -79 (31) 53/90 (59%) 29 were submitted to a series of GLMMs (see Fig. 6 for the mean percent use SpExpSp -71 (45) -217 – 7 (3) -83 (35) 85/99 (86%) of prevoicing per group and language). In this case, the best fitting 16 model was a GLMM with group, language, segment, group x language SpInexpSp -66 (44) -182 – 13 (6) -79 (30) 88/100 (88%) and segment x language as fixed effects and random intercepts for 27 participant and word. The statistical model showed a significant effect of group [F(2, 1140) = 5.928, p = .003] and language [F(1, 1140) = of voice-lead VOT (-43 ms) – i.e., the closest to EnMono (1 ms) – and 37.958, p = .000], but no effect of segment [F (1, 1140) = 2.142, p = used prevoicing the least (59%). SpExpSp and SpInexpSp used pre­ .144]. Thus, prevoicing was more frequently used in Spanish (mean = voicing more often (86% and 88%, respectively) and produced the 96%, SD = 21) than in English (mean = 77%, SD = 42). Bonferroni longest mean negative VOT (-71 ms and -66 ms, respectively). With pairwise comparisons indicated that SpExpUK used prevoicing (across respect to velar stops, SpExpUK presented the shortest mean voice-lead languages) significantly less often (mean = = 74%, SD = 44) than VOT duration (-24 ms) and used prevoicing less frequently than the SpExpSp (mean = 92%, SD = 27) and SpInexpSp (mean = 91%, SD = other Spanish groups (55% of time, see Table 11). SpExpSp prevoiced 28), p = .029, in both cases. The language x group interaction did not reach significance [F(2, 1140) = 1.527 p = .218], but the segment x 102 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 prevoicing with Spanish /b/ than with /g/ (88% vs. 93%, respectively). Finally, the Pearson correlation test revealed a significant – though moderate – correlation between L1 and L2 /b/ and /g/ (/b/: r =.316; N = 29, p = .047; /g/: r = .466; N = 29; p = .005). 3.4. Interim discussion Experiment 2 focused on the production of L2 and L1 /b/ and /g/ by Spanish learners of English varying in amount of L2 experience. Results showed that L2 experience was found to modulate the accuracy of production of L2 voiced stops, since SpExpUK presented the most target- like productions and SpInexpSp were the Spanish group that performed the least accurately. That is, the greater the amount of L2 experience learners had, the more target-like their productions of English voiced stops were. In this case, differences in experience are to be explained by differences in length of residence, as, unlike with the English L1 groups, the Spanish L1 groups did not differ in years of L2 instruction (averaging Fig. 4. Boxplot of VOT production (in ms) for English and Spanish /b/ by the from 13.2 to 13.9 years, see Table 5 above). In addition, and unlike the L1-Spanish speakers. L1-English groups, the L1-Spanish groups did differ in amount of L2 use, as the group living in the L2 setting used the L2 the most (54.6%, vs. 26.3% and 22.7% for SpExpSp and SpInexpSp, respectively). Conse­ quently, in this case, differences in FTE years of L2 input (Flege and Bohn, 2021) between the two most experienced groups were also greater (2.18 vs. 0.16) than with the L1-English groups. It should be noted, though, that none of the L1-Spanish groups produced English voiced stops in a target-like manner, as all used prevoicing significantly more than EnMono. Still, SpExpUK, who presented the closest values to EnMono’s and prevoiced less often, differed significantly from SpInexpSp and SpExpSp. This finding is in line with previous studies that have found that L2 experience results in a more accurate production in the L2 (e.g., Flege 1987; Flege et al. 1997; Gorba and Cebrian 2021; Lev-Ari and Peperkamp 2013; Levy and Law 2010) and indicates that the positive effect of L2 length of residence on voiceless stop production can be extended to voiced stops. As for the results obtained for L1 production, the Spanish learners’ L1 production of voiced stops was still found to fall within native-like values, as none of the groups differed from Spanish monolinguals. Thus, it seems that living in an L2 setting did not result in phonetic drift towards L2 values. A similar explanation as the one given for the English Fig. 5. Boxplot of VOT production (in ms) for English and Spanish /g/ by the L1-Spanish speakers. learners above could apply to the Spanish learners: specifically, it is possible that their length of residence was not long enough for the L1 to change towards L2 values and novelty effects would not apply given their amount of L2 experience. However, SpExpinUK used prevoicing numerically less frequently than the other L1 Spanish-groups, particu­ larly in the case of /b/. This finding may indicate that the production of /b/ in Spanish may be beginning to shift towards more L2-like values in the case of the most experienced group. Regarding the differences between L1 and L2 production in the Spanish learners, all groups were found to produce voiced stops differ­ ently in Spanish and in English – i.e., used prevoicing significantly more often in Spanish than in English. Nevertheless, this difference was numerically small, and the production of short-lag VOT voiced stops (14-15% for SpInexpSp and SpExpSp, 42% for SpExpUK) was still far from native English values (91%). This shows a notable difficulty in using an L2 cue that is associated with a voiceless category in their L1. Still, an effect of experience was observed, since the most experienced group used prevoicing the least. In short, L2 experience, related to Fig. 6. Mean use of prevoicing in the production of for English and Spanish longer residence in an L2 setting and greater amount of L2 use, affected stops by the L1-Spanish speakers. The dashed line and the dotted line show the the L2 production of Spanish learners of English. A greater amount of mean values for SpMono and EnMono, respectively. Percentage use of pre­ experience with English seemed to result in a greater use of short-lag voicing and standard error of the mean are given. VOT, and a reduction in the use of prevoicing, the predominant L1 cue. No clear effects of the L2 on the L1 were found, possibly due to a language did [F(1, 1140) = 4.073, p = .044]. Bonferroni pairwise relatively small amount of L2 experience. Overall, all groups used pre­ comparisons indicated that /b/ and /g/ differed in Spanish (/b/: 93%, voicing significantly more frequently in the L1 than in the L2, even SD = 25; /g/: 98%, SD = 15; p = .038), but not in English (/b/: mean = though the difference between the two languages was very small, 78%, SD =41; /g/ = 76%, SD = 43), possibly because SpExpUK used less particularly for the less experienced groups. 103 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 4. General discussion production of L1 voiced stops and the role of L2 experience. Similar results were obtained for the two populations, as a clear effect of the L2 This study has examined the production of L1 and L2 voiced stops on the L1 was not observed – none of the groups differed significantly (specifically /b/ and /g/) by English learners of Spanish (experiment 1) from their L1 controls, that is, no clear influence of the L2 was found on and Spanish learners of English (experiment 2), whose L1s differ in the the L1. A greater variation was observed in the case of the English use and weight of the main voicing cues, namely short-lag VOT and learners of Spanish, both within group and between groups, possibly due prevoicing. The production of the two populations in the two languages to the fact that prevoicing is used natively in English, in free variation has been compared in terms of the presence or absence of prevoicing (i. with short-lag VOT. The L1-English group that prevoiced the most often e., voice-lead VOT), which is the main cue to stop voicing (in initial post- in English was EnExpUK (22%), the one with some previous experience pausal position) in Spanish, and a possible, but not main, cue in English, in the L2 setting and the greatest amount of L2 instruction, as mentioned where voiced stops are often produced with short-lag VOT. The effect of above, but this greater use of prevoicing was not significantly different L2 experience on L2, and L1, production has been assessed by comparing from the English monolingual pattern (9%). Still, EnExpUK and EnExpSp three groups for each population differing in experience (in terms of (17%) made a significantly greater use of prevoicing in English than length of residence in the L2 setting, but also amount of L2 use and L2 EnInexpUK (3%), showing that an increase in prevoicing in the L2 may instruction). Thus, L1 and L2 production were examined by comparing carry with it an increment of prevoicing in L1, and illustrating a possible the production of the L2 learners sharing the same L1 and the corre­ effect of L2 learning on L1 patterns. In the case of the Spanish learners, sponding control monolingual group in each experiment. The perfor­ the only indication that L2 experience may have influenced L1 pro­ mance of the learner groups in the two languages was also compared so duction is the fact that the most experienced Spanish group (SpExpUK) as to determine whether learners produced L1 and L2 voiced stops yielded a lower percent use of prevoicing in Spanish compared to the differently and to evaluate the role that L2 experience played on the other L1-Spanish groups (/b/: 88% vs. 94-99%; /g/: 93% vs. 99-100%). ability to make this difference. So far, the outcomes of each experiment In other words, SpExpUK produced the greatest amount of short-lag VOT have been discussed separately. This section will bring together the re­ voiced stops in Spanish (12%), thus showing some influence from the L2 sults obtained for both experiments and will attempt to answer the on the L1. Still, given that no significant differences were observed with research questions presented in Section 1.3. the Spanish control group, there is not enough evidence to claim that The first research question addressed the effect of L2 experience on there has been a drift towards L2 values in the production of the L1. The the production of L2 voiced stops. Results indicated that, overall, none of absence of a clear effect of the L2 on the L1 may be explained by the fact the groups performed in a target-like manner, as, with the sole exception that the experienced learners in the present study had been living in an of EnExpUK, all groups from both populations differed significantly immersion setting for a shorter period than participants in previous from the target language controls in the use of prevoicing in the pro­ studies (e.g., about four years in the present study as opposed to 12 in duction of L2 stops. In the case of the Spanish learners, all groups studies such as Flege (1987)). Therefore, just as in the case of Gorba and differed from English monolinguals, but the most experienced group Cebrian (2021), it is possible that phonetic drift was not evident because (SpExpUK) performed in a more target like-manner, as they prevoiced a longer period of L2 experience and a more intensive use of the L2 – in the least often. In fact, SpExpUK used prevoicing in the production of detriment of the L1 – is required for it to occur. Moreover, none of the English voiced stops significantly less frequently than the other L1- groups in the present study were absolute beginners, as they had either Spanish groups (58% vs. 85%, 86%; EnMono: 9%). Differences in resided in an L2 setting or received L2 formal instruction – or both –, years of residence in the L2 setting, together with the amount of L2 use, thus canceling a novelty effect (Chang, 2012, 2013). In this line, other can explain SpExpUK’s better performance (recall that the L1-Spanish previous studies that have reported changes in the L1 involved very groups hardly differed in years of L2 instruction). In the case of the short periods of time using the L2 intensively, and where changes in the English learners of Spanish, as mentioned above, EnExpUK – the group L1 diminished a few weeks after returning to the L1 setting, for Spanish with some previous experience living in the L2 setting – presented the learners of English (Kartushina and Martin, 2019). As for English most target-like use of prevoicing for Spanish voiced stops (71%, vs. learners, amount of L2 instruction has also been found to be key for L1 EnExpSp: 58% and EnInexpUK: 29%; SpMono: 98%). As discussed in the drift to occur, both in the direction of the L2 (Herd et al., 2015; Nagle, interim discussion (see Section 2.3), other factors may have resulted in 2019) and in terms of dissimilation between L1 and L2 categories EnExpUK’s more target-like performance than EnExpSp, the group with (Dmitrieva et al., 2020). the longest length of residence. The fact that EnExpUK had received the By testing both the L1 and the L2, it is possible to determine whether greatest amount of L2 instruction together with their student status at the L2 learners produced the L1 and the L2 with different values and to the time of testing may have resulted in a more accurate production. In assess which of the possible scenarios discussed in the introduction may fact, previous studies report positive effects of L2 instruction on L2 have taken place. The overall results show that all groups produced productions in L1 settings (e.g., Dmitrieva et al. 2020; Nagle 2019). voiced stops with different values in English and Spanish, and, thus, that Moreover, the group with longer experience in an L2 setting did not the production of L2 stops is distinguished from the production of L1 report using Spanish notably more often than the learners in the UK stops – i.e., all groups produced more prevoiced stops in Spanish than in (16% vs 15% and 13% for EnExpSp and ExpInexpUK, respectively), English. This may indicate that learners are not simply transferring their contrary to expectation, which resulted in a relatively small difference L1 patterns, particularly the English learners of Spanish, who increased between EnExpSp and EnExpUK in terms of years of FTE. Hence, in the considerably the amount of prevoiced productions in the L2 with respect absence of actual differences in L2 use or FTE, EnExpUK’s better per­ to the L1. Still, producing L1 and L2 stops differently does not mean that formance in the L2 can be attributed to the role of L2 instruction. L2 stops were produced in a target-like manner. Only EnExpUK’s L2 Another explanation, as will be discussed below, is related to the values were comparable to – that is, not significantly different from – possible transfer of the use of prevoicing from the L1 to the L2. In brief, native values, and even in this case the frequency of use was notably L2 experience has been found to improve L2 performance, but the fac­ lower (71%, vs. 98% for SpMono). This result is similar to the outcome tors that contribute to experience vary for the two populations. Years of reported by Gorba and Cebrian (2021), who examined the production of L2 instruction seems to play the main role in the case of the English voiceless stops by the same groups of English participants as the current learners of Spanish, given the absence of real differences in L2 use or study. The results showed that all learners, regardless of their amount of FTE, while length of residence and L2 use are the crucial factors in the L2 experience, produced English voiceless stops with significantly case of the Spanish learners of English. These differences in experience longer VOT values than Spanish stops. In that case, however, the two between the two populations are further discussed below. experienced groups performed similarly to the Spanish control group, The second research question investigated the effect of the L2 on the and only the inexperienced group differed significantly from the 104 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 controls. The more target-like performance observed by Gorba and learners of English and the Spanish monolinguals. Finally, no indication Cebrian (2021) suggests a somewhat greater difficulty to acquire L2 of the remaining possibility, i.e., L1-L2 category dissimilation as a result voiced stops than voiceless stops, in line with what the outcomes of of L2 learning, was found (EnInexpSp’s use of prevoicing was numeri­ previous studies suggest (see Section 1.2.2. for a review of studies cally lower than EnMono’s, 9% vs. 3%, respectively, but as mentioned involving voiceless and voiced stops). Regarding the Spanish learners of above this difference was non-significant). English, despite significant differences between L1 and L2, the use of The main research questions have been discussed. Given that two prevoicing was predominant in both languages, and close to L1-Spanish populations were investigated, and different outcomes were observed in values (see Figs. 3 and 6 above), pointing to a clearer evidence of some cases, the possible explanations for these disparities will be dis­ transfer of the L1 pattern onto the L2. Transfer in this case resulted in cussed next. The first difference between the two populations examined acceptable productions in English given that, as stated above, prevoicing has to do with the factors that contributed to L2 experience in each case. is also found in native English even if less frequently than short-lag VOT. As mentioned above, the L1-Spanish group that performed in a more Only SpExpUK illustrates a greater percent use of short-lag VOT (42%), target-like manner was SpExpSp, that is, the group with the longest but still far from the English norm (EnMono: 91% short-lag length of residence in the L2-speaking setting. By contrast, its L1-English productions). counterpart, EnExpSp, was outperformed by EnExpUK, a group with The relationship between the use of prevoicing in the L1 and in the only a previous shorter stay in the L2 country. This illustrates that length L2 was also investigated. Moderate though significant correlations were of residence did not have the same effect for both populations, despite found for both populations and both phones, in line with some previous the fact that the groups were comparable across populations in terms of results (Hutchinson and Dmitrieva, 2021; MacKay et al., 2001). When it the number of years spent in the L2 country. This difference between the comes to the English learners of Spanish, given that the use of prevoicing two populations is related to a related variable, amount of L2 use. In the is subject to individual and free variation in English, it was hypothesized case of the L1-Spanish groups, SpExpUK and SpExpSp differed both in that the use of voice-lead VOT in their L2 may be related to its use in the length of residence and in amount of L2 use. By contrast, the English L1 L1 due to phonetic transfer of an available L1 cue. The correlations groups did not differ much in L2 use despite differences in place and conducted on the use of prevoicing in the L1 and L2 indicated that those length of residence. As a matter of fact, L2 use has been attached great English learners that used prevoicing the most frequently in English also importance in the revised version of the SLM (SLM-r, Flege and Bohn, did so in Spanish. As a matter of fact, at a group level, EnExpUK used 2021). Recall that, according to the SLM-r, a determining factor in L2 prevoicing the most in both languages. This outcome is in agreement speech learning is FTE (Full Time Equivalent of years of L2 input), a with some previous studies that showed that the use of negative VOT in composite variable that considers both length of residence and L2 use. the L1 and the L2 were correlated, and, thus, that the use of prevoicing in As discussed above, the English groups differed little in FTE despite the the L1 was transferred to the L2 (MacKay et al., 2001; Schuhmann and difference in length of residence, while the Spanish groups had more of a Huffmann, 2015). Still, the fact that use of prevoicing in the L1 and the difference. L2 is correlated could also be interpreted as an increase in prevoicing in On the other hand, the L1-English groups differed in amount of L2 the L1 resulting from its use in the L2. In any case, the fact that pre­ instruction, as EnExpUK had received on average more years of Spanish voicing is a possible native cue to voiced stops in English may have instruction than EnExpSp (7.5 vs. 2.4, respectively). L2 formal instruc­ facilitated the production of target-like – prevoiced – voiced stops in tion is a factor that has been reported to improve the production of L2 Spanish; hence the overall better L2 performance of the L1-English stops (e.g., Casillas 2019; Chang 2012; Lord 2010; Nagle 2019; Schuh­ groups than of the L1-Spanish groups. Still, instances of negative mann and Huffmann, 2015; Zampini 1998). Interestingly, the amount of transfer are also evident in the fact that the L1-English speakers pro­ L2 instruction received by the English learners of Spanish was smaller duced Spanish stops with short-lag VOT stops 29-71% of the time. As we than that received by the Spanish groups (13.5 years on average across have seen, this type of negative transfer diminished as a function of L2 groups) and yet the English learners of Spanish produced L2 voiced stops experience. With respect to the Spanish learners of English, the common more similarly to native speakers of the target language than the Spanish use of prevoicing in both languages shows L1 transfer onto the L2, as learners of English. It seems plausible, thus, that greater differences discussed above. between the two populations – namely, a greater advantage of the En­ On the other hand, little evidence has been found regarding the other glish population over the Spanish population – would have been two scenarios discussed in the introduction, i.e., involving an effect of observed if the L1-English speakers had received a comparable amount the L2 on the L1 either as the result of the L2 pulling the L1 towards more of L2 instruction to that of the L1-Spanish speakers. L2-like forms or as the result of category dissimilation between L1 and This difference in what contributes to L2 experience between the two L2 sounds. With respect to the former, recall that the two most experi­ populations (FTE vs. L2 instruction), however, may be an artifact of the enced L1-English L2-Spanish groups (EnExpUK and EnExpSp) were samples examined in the current study, and not generalizable to the two found to use prevoicing in English significantly more often than the least populations. For instance, length of residence may have played a greater experienced group (EnInexSp). This may indicate that learning to use role with L1-English speakers if the group residing in the L2 setting had prevoicing predominantly as a cue to stop voicing in Spanish may have consisted of English speakers who made a greater use of the L2, e.g., not carried with it a greater use of prevoicing in the L1. Still, none of the English teachers as it was mostly the case in the current study. It should three L1-English groups differed in use of prevoicing from the English be noted, however, that these differences in L2 use and amount of L2 controls. Regarding the L1-Spanish groups, some influence of the L2 on instruction are in fact a reflection of the linguistic reality of the two the L1 can be detected particularly in the case of the most experienced populations under study. Regarding language use, English is more pre­ group, who prevoiced the least in both languages. In the L1, SpExpUK sent than Spanish in the international media and entertainment in­ used prevoicing numerically less than the Spanish controls, particularly dustry, resulting in a relatively greater presence of English in an L1- in the case of /b/ (SpMono: 98% of prevoiced tokens; SpExpUK: 88%). Spanish setting than the reverse. Moreover, whereas Spanish speakers The use of short-lag VOT on the part of SpExpUK (12% of productions) is in the UK generally use English at work and in everyday life interactions, mainly attributable to English influence, given that short-lag VOT is the the reverse is not always the case for English speakers in Spain, since most common cue for voiced stops in English, while it is not used to many use English at work and in their social interactions. In fact, that is signal voicing in Spanish. Thus, the greater use of short-lag VOT in L1 mostly the case in this study, as most of the participants in the EnExpSp production, as well as the fact that the use of prevoicing in the two group were teaching English as a foreign language. In addition, languages was correlated, may also show some influence of the L2 on the regarding L2 instruction, English is more widely taught as a foreign L1. Recall, however, that this effect on the L1 was not significant, as no language in Spain than the other way around. Finding truly comparable differences emerged in L1 production between any of the Spanish groups in terms of L2 use and L2 instruction may be challenging, at least 105 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 for these specific populations (English learners of Spanish in Spain and 1987). As a result, it is possible that certain crosslinguistic processes, Spanish learners of English in the UK). such as it is the case of phonetic/phonological changes of the L1 towards A second difference between the two populations has to do with the the L2, were not clearly observed (Gorba and Cebrian, 2021). Another allophonic distribution of the feature under study in the two languages limitation regarding participants is related to sample size: although investigated. Recall that in English the initial stop voicing contrast is around 50 participants were tested in each experiment, each group was implemented as a short-lag vs. long-lag VOT contrast, and voice-lead made up of between 9 and 11 participants. It should also be noted that unequivocally signals a voiced stop, although it is used less frequently the tasks used to retrieve the production of voiced stops – i.e., a carrier than short-lag VOT. In Spanish, on the other hand, the contrast is be­ sentence reading task in each language – was a formal and controlled tween voice-lead and short-lag VOT. Thus, an English learner of Spanish task. While this was motivated by allophonic patterns of Spanish and has to learn to make a greater use of an existing L1 cue, voice-lead, when was done to prevent spirantization of the stops (/b/ and /g/ are pro­ producing L2-Spanish voiced stops, and to reduce the amount of VOT duced as [β] and [ɣ], respectively, between continuant sounds (e.g., used in the case of L2 voiceless stops. By contrast, a Spanish learner of Colina 2020; Hualde 2005)), the data examined in the current paper English needs to learn that what constitutes a voiceless category in the may not be representative of spontaneous speech. Further research L1, i.e., short-lag VOT, in fact typically signals the contrasting category should also collect spontaneous data in order to corroborate that the in the L2, a voiced stop. Increasing the use of an existing albeit not results of this study reflect the real speech of L2 learners. The acquisition predominant L1 cue (voice-lead for English L1 speakers) may be easier – on the part of English learners of Spanish – and transfer to the L2 – in to achieve than to use an existing L1 cue that signals a different category the case of Spanish learners of English – of the process of spirantization in the L1 (short-lag VOT for Spanish L1 speakers). This may be due to the needs also to be investigated in order to get a full picture of the acqui­ fact that while, in the case of the English L1 speakers, learning the L2 sition of voiced stops in these populations in Spanish. system does not affect the phonological value associated with voice-lead VOT (a voiced stop both in the L1 and the L2), in the case of the Spanish 5. Summary and conclusions L1 speakers, learning the English contrast implies learning to dissociate short-lag VOT from the L1 (voiceless) category and associating it with This study has investigated the acquisition of L2 voiced stops by two the contrasting (voiced) category. Hence the greater difficulty on the populations, English learners of Spanish and Spanish learners of English, part of the Spanish learners of English to produce English voiced stops whose L1s base the initial stop voicing contrast on a different use of the with short-lag VOT than on the part of English learners of Spanish to same phonetic cues. Moreover, the effect of L2 experience on L1 and L2 produce Spanish voiced stops with prevoicing. Furthermore, simply production of /b/ and /g/ was analyzed by comparing groups differing transferring the use of the main L1 cue onto the L2 has different con­ in amount and type of L2 learning experience. Differences in the pro­ sequences for the two populations. For L1-Spanish L2-English speakers duction of the two languages was also assessed by comparing L1 and L2 using prevoicing in English does not impede intelligibility, since pre­ production. Overall, L2 experience has been found to have a positive voiced stops can only be interpreted as voiced stops. By contrast, for effect on L2 voiced stop production, as the inexperienced learners of English L1 Spanish L2 speakers, producing Spanish stops with short-lag both populations consistently performed in a less target-like manner VOT can result in lack of intelligibility. Consequently, it is more crucial than the groups with some L2 experience. Still, L2 speakers’ production for the L1-English L1 speakers than for the L1-Spanish speakers to avoid generally differed from native speech, particularly in the case of the transfer of the L1 main cue. Spanish learners of English, who were found to produce prevoiced stops Nevertheless, stating that the Spanish learners of English did not to a much greater extent than monolingual English speakers. Different produce L2 voiced stops accurately on the basis that they used pre­ factors have been found to contribute to L2 experience for each popu­ voicing more often than English monolingual speakers is arguable. This lation of L2 learners. Length of residence together with the amount of is because prevoicing is used in English voiced stops in initial position in language use accounted for between group differences among the free variation – that is, voice-lead and short-lag VOT do not contrast – Spanish learners of English. In the case of the English learners of although short-lag VOT is more commonly used (Lisker and Abramson, Spanish, differences in the amount of time spent in the L2-speaking 1964; Docherty, 1990). Perhaps, it would be more appropriate to state country did not influence L2 performance, presumably because groups that the Spanish learners generally did not produce L2 voiced stops in did not differ in the amount of L2 use. In this context, the amount of L2 the most common manner for native English speakers and were not able instruction and the condition of being a student may have been a more to make a clear distinction between the L1 and the L2. On the other determining factor in the case of the English learners of Spanish. No hand, the English learners of Spanish possibly made a clearer distinction notable effect of L2 experience on the L1 was found, as none of the between the two languages, thanks to the subphonemic status of pre­ learner groups differed from their L1 controls. The only indication of an voicing in English, which coincides with that of prevoicing in Spanish – influence of the L2 on the L1 can be found in the English learners of i.e., is used to produce voiced stops. In short, it appears that learning to Spanish with some experience in the L2 setting, as they used prevoicing produce on an existing L1 cue may be easier than learning to use a cue in English significantly more often than the least experienced group and associated with a different L1 category. numerically more than the English controls. Overall, longer length of This study is not without limitations. As mentioned above, the two residence in the L2 setting, together with a greater amount of L2 use, populations under study were mirror-image in terms of L1 and L2 and may be necessary for L2 effects of the L1 to be more generally observed. comparable in terms of length of residence in an L2 setting but differed Regarding the comparisons between L1 and L2 production, all groups in some respects – i.e., in amount of L2 use and L2 instruction. It is were found to use significantly more prevoicing in Spanish than in En­ possible that different outcomes would have been observed given more glish. Despite the significant difference, most Spanish groups were found similar groups. Therefore, future research with more comparable L1- to use prevoicing predominantly in both languages, pointing to the English and L1-Spanish speakers in terms of length of residence as transfer of the L1 pattern onto the L2. L1-English speakers made a well as language use is necessary to fully assess the roles of these vari­ greater difference between the two languages than L1-Spanish L1 ables for these two populations. In fact, future studies may benefit from speakers did and were overall more successful in that they approximated using compound independent variables – i.e., that combine more than target language use of prevoicing more closely. The differences between just one factor –, such as it is the case of the SLM-r’s FTE, as that may the English and the Spanish learners can be explained by the differences help to control for individual differences between participants. in the use of the voicing cues in the languages under study. The fact that Furthermore, the amount of L2 experience of the participants tested – i. prevoicing is a possible cue, even if not the main one, to initial stop e., the number of years spent in an L2 setting – was considerably smaller voicing in English may have facilitated a more target-like production of in the present study than it was in some previous research (e.g., Flege voiced stops on the part of the English learners of Spanish. On the other 106 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 hand, Spanish learners may have initially assimilated L2 /b/ and /g/ to 20 Flawless is the next word their corresponding L1 categories, that is, with voice-lead VOT, as short- 21 Peanut is the next word lag VOT is associated with phonologically voiceless stops in their L1. 22 Geekfest is the next word Recall, however, that prevoicing is possible, although less common, in 23 Flipflop is the next word English voiced stops and, thus, it can be argued that using Spanish values 24 Beetroot is the next word in English is acceptable. Further research is necessary to evaluate the 25 Beating is the next word contribution of the experience variables examined here, with more 26 Headphones is the next word balanced designs. In addition, it remains to be seen if the results of this 27 Keener is the next word study would be replicated in the production of more spontaneous 28 Geeky is the next word speech, as opposed to read speech. These issues are left for future 29 Ancient is the next word research. 30 Peacock is the next word 31 Gearshift is the next word CRediT authorship contribution statement 32 Keyhole is the next word Celia Gorba: Conceptualization, Methodology, Software, Valida­ tion, Formal analysis, Investigation, Resources, Data curation, Writing – A.2. Spanish production task original draft, Writing – review & editing, Visualization, Project administration, Funding acquisition. Juli Cebrian: Conceptualization, Lee cada una de las siguientes frases dos veces: Methodology, Software, Validation, Formal analysis, Investigation, Re­ sources, Data curation, Writing – original draft, Writing – review & 1 Monte es la siguiente palabra editing, Visualization, Supervision, Project administration, Funding 2 Pila es la siguiente palabra acquisition. 3 Hombre es la siguiente palabra 4 Quicio es la siguiente palabra Declaration of Competing Interest 5 Bizco es la siguiente palabra 6 Guinda es la siguiente palabra The authors declare the following financial interests/personal re­ 7 Piso es la siguiente palabra lationships which may be considered as potential competing interests: 8 Mono es la siguiente palabra Juli Cebrian reports financial support was provided by Spain’s Ministry 9 Bicho es la siguiente palabra of Science and Innovation. 10 Guiso es la siguiente palabra 11 Rata es la siguiente palabra Data availability 12 Birla es la siguiente palabra 13 Ante es la siguiente palabra Data will be made available on request. 14 Quinta es la siguiente palabra 15 Orden es la siguiente palabra 16 Guía es la siguiente palabra Acknowledgments 17 Pista es la siguiente palabra 18 Perro es la siguiente palabra This work was supported by Research Grants No. FFI2017-88016-P 19 Biblia es la siguiente palabra and PID2021-122396NB-I00 from the Spanish Ministries of Economy 20 Guiño es la siguiente palabra and Competitiveness and Science and Innovation. 21 Uña es la siguiente palabra 22 Quita es la siguiente palabra Appendix A. List of sentences used in the production tasks 23 Uso es la siguiente palabra 24 Bici es la siguiente palabra A.1. English production task 25 Móvil es la siguiente palabra 26 Quince es la siguiente palabra Read each of the following sentences twice: 27 Pico es la siguiente palabra 28 Cama es la siguiente palabra 1 Kitten is the next word 29 Quise es la siguiente palabra 2 Peeler is the next word 30 Susto es la siguiente palabra 3 Beaches is the next word 31 Guita es la siguiente palabra 4 Watches is the next word 32 Pino es la siguiente palabra 5 Gearbox is the next word 6 Cellphone is the next word References 7 Keychain is the next word Audacity Team., 2016. Audacity(R): free audio editor and recorder [computer 8 Peaceful is the next word application]. Version 2.1.1. Retrieved in February 2016 from https://audacityteam. 9 Tiger is the next word org/. 10 Pieces is the next word Baese-Berk, M.M., 2019. Interactions between speech perception and production during learning of novel phonemic categories. Atten. Percept. 81 (4), 981. http://mendeley. 11 Houses is the next word csuc.cat/fitxers/40f99bf6fdfa610abb22465d179a7f43. 12 Beefcake is the next word Bergmann, C., Nota, A., Sprenger, S.A., Schmid, M.S., 2016. L2 immersion causes non- 13 Keenly is the next word native-like L1 pronunciation in German attriters. J. Phon. 58, 71–86. https://doi. org/10.1016/j.wocn.2016.07.001. 14 Girlfriend is the next word Best, C. T., Michael D. T., 2007. Nonnative and second-language speech perception: 15 Keeper is the next word commonalities and complementarities. In Ocke-Schwen Bohn & Murray J. Munro 16 Music is the next word (Eds.), Language Experience in Second Language Speech Learning: In Honor of 17 Geezer is the next word James Emil Flege, 10389 (2): 13–34. doi:10.1121/1.1332378. Birdsong, D., Gertken, L.M., Amengual, M., 2012. Bilingual Language Profile: An Easy-to- 18 Razor is the next word Use Instrument to Assess Bilingualism. COERLL, University of Texas at Austin. https 19 Beetle is the next word ://sites.la.utexas.edu/bilingual/. 107 C. Gorba and J. Cebrian Speech Communication 146 (2023) 93–108 Boersma, P., Weenink, D., 2016. Praat: doing phonetics by computer [computer Herd, W.J., R.L. Walden, Whitney L. Knight, and Savana N. Alexander. 2015. Phonetic software]. Version 6.0.19. Retrieved February 2016 from http://www.praat.org/. drift in a first language dominant environment. Proceedings of the Paper Presented Bohn, O.S., Flege, J.E., 1990. Interlingual identification and the role of foreign language at the Meetings on Acoustics, Jacksonville, FL, USA, November 2-6; Volume 23, pp. experience in L2 vowel perception. Appl. Psycholinguist. 11 (3), 303–328. https:// 1-9. doi.org/10.1017/S0142716400008912. Hualde, J.I., 2005. The Sounds of Spanish with Audio CD. Cambridge University Press. Caramazza, A., Yeni-Komshian, G.H., Zurif, E.B., Carbone, E., 1973. The acquisition of a Huffman, M.K., Schuhmann, K.S., 2016. Effect of Early L2 Learning on L1 Stop Voicing. new phonological contrast: the case of stop consonants in French-English bilinguals. In: Proceedings of the Paper presented at the Meetings on Acoustics, 23. Acoustical J. Acoust. Soc. Am. 54 (2), 421. http://mendeley.csuc.cat/fitxers/503cef848b96 Society of America, Pittsburgh, PA, USAPittsburgh, pp. 1–10. September 9. b7b7a540101eabff812. Hutchinson, A.E., Dmitrieva, O., 2021. Individual variation in acquisition of voicing Casillas, J., 2019. Phonetic category formation is perceptually driven during the early contrasts by classroom learners of French. Journal of Second Language stages of adult L2 development. Lang. Speech 28. https://doi.org/10.1177/ Pronunciation 7 (3), 408–434. 0023830919866225. Julià i Muné, J., 1981. Estudi Contrastiu del Oclusius de L’anglès i del Català : Un Castañeda, M.L., 1986. El VOT de las oclusivas sordas y sonoras españolas. Estud. Fon. Experiment Acústic, 1. Estudi General: Revista de La Facultat de Lletres de La Exp. 2, 92–110. Universitat de Girona, p. 75, 7. Chang, C.B., 2012. Rapid and multifaceted effects of second-language learning on first Jun, S., Cowie, I., 1994. Interference for ‘New’ and ‘Similar’ Vowels in Korean Speakers language speech production. J. Phon. 40 (2), 249–268. https://doi.org/10.1016/j. of English, 43. Ohio State University, pp. 117–130. Working Papers. wocn.2011.10.007. Kartushina, N., Frauenfelder, U.H., Golestani, N., 2016. How and when does the second Chang, C.B., 2013. A novelty effect in phonetic drift of the native language. J. Phon. 41 language influence the production of native speech sounds: a literature review: L2 (6), 520–533. https://doi.org/10.1016/j.wocn.2013.09.006. influences on L1: a literature review. Lang. Learn. 66 (S2), 155–186. https://doi.org/ Colina, S., 2020. Spirantization in Spanish: the role of the underlying representation. 10.1111/lang.12187. Linguistics 58 (1), 1–35. Kartushina, N., Martin, C.D., 2019. Third-language learning affects bilinguals’ Derek, C., DeMars, C., 2019. An updated recommendation for multiple comparisons. production in both their native languages: a longitudinal study of dynamic changes Adv. Methods Pract. Psychol. Sci. 2 (1), 26–44. in L1, L2 and L3 vowel production. J. Phon. 77, 100920 https://doi.org/10.1016/j. Docherty, G.J., 1990. An experimental phonetic study of the timing of voicing in English wocn.2019.100920. obstruents. Diss. Abstr. Int. 51 (3). http://mendeley.csuc.cat/fitxers/41e02d2e Kuhl, P.K., 1991. Human adults and human infants show a “perceptual magnet effect” for 0aa875774b1d4557e8b2ded4. the prototypes of speech categories, monkeys do not. Perception 50 (2), 93. Dmitrieva, O., Jongman, A., Sereno, J.A., 2020. The effect of instructed second language Lev-Ari, S., Peperkamp, S., 2013. Low inhibitory skill leads to non-native perception and learning on the acoustic properties of first language speech. Languages 5 (4), 44. production in bilinguals’ native language. J. Phon. 41 (5), 320–331. https://doi.org/ Flege, J.E., 1987. The production of “new” and “similar” phones in a foreign language: 10.1016/j.wocn.2013.06.002. evidence for the effect of equivalence classification. J. Phon. 15 (1), 47–65. https:// Levy, E.S., Law, F.F., 2010. Production of french vowels by American-English learners of doi.org/10.1016/S0095-4470(19)30537-6. French: language experience, consonantal context, and the perception-production Flege, J.E., 1990. Production of the English word-final /t/-/d/contrast by native speakers relationship. J. Acoust. Soc. Am. 128 (3), 1290. https://doi.org/10.1121/ of Mandarin and Spanish. Acoust. Soc. Am. J. 88, 81. https://doi.org/10.1121/ 1.3466879. 1.2029176. Lisker, L., Abramson, A., 1964. A cross-language study of voicing in initial stops: Flege, J.E., 1991. Age of learning affects the authenticity of voice-onset time (VOT) in acoustical measurements. Word 20, 384–422. https://doi.org/10.1080/ stop consonants produced in a second language. J. Acoust. Soc. Am. 89 (1), 395–411. 00437956.1964.11659830. https://doi.org/10.1121/1.400473. Lord, G., 2010. The combined effects of immersion and instruction on second language Flege, J.E., Winifred, S., 1995. Second language speech learning: theory, findings, and pronunciation. Foreign language annals 43 (3), 488–503. problems. Speech Perception and Linguistic Experience: Issues in Cross-Language MacKay, I., Flege, J., Piske, T., Schirru, C., 2001. Category restructuring during second- Research. York Press, Timonium, MD, pp. 233–277. https://doi.org/10.1111/j.1600- language (L2) speech acquisition. J. Acoust. Soc. Am. 110, 516–528. https://doi.org/ 0404.1995.tb01710.x. 10.1121/1.1377287. Flege, J.E., Bunneister, P., Piske, T., Rohde, A., 2002. Interactions between the native and Major, R.C., 1992. Losing English as a first language. Mod. Lang. J. 76 (2), 190. http:// second-language phonetic systems. An Integrated View of Language Development: mendeley.csuc.cat/fitxers/0af2fb3eae3a42c82adde937d9b0701f. Papers in Honor of Henning Wode. Wissenschaftlicher Verlag, Trier, pp. 217–244. Maye, J., Gerken, L., 2000. Learning phonemes without minimal pairs. In: Proceedings of Flege, J.E., Schiller, Niels O., Meyer, Antje S., 2003. Assessing constraints on second- the 24th Annual Boston University Conference on Language Development, 2, language segmental production and perception. Phonetics and Phonology in pp. 522–533. Language Comprehension and Production: Differences and Similarities. de Gruyter Mora, J. C., 2008. Learning context effects on the acquisition of a second language Mouton, Berlin, pp. 319–355. phonology. In C. Pérez-Vidal, M. Juan-Garau, A. Bel, (eds.), A Portrait of the Young Flege, J.E., Cole, J., Hualde, J.I., 2007. Language contact in bilingualism: phonetic in the New Multilingual Spain. Clevedon: Multilingual Matters. 241-263. system interactions. Laboratory Phonology 9. De Gruyter Mouton, New York, Nagle, C., 2019. A longitudinal study of voice onset time development in L2 Spanish pp. 353–381. stops. Appl. Linguist. 40 (1), 86–107. https://doi.org/10.1093/applin/amx011. Flege, J.E., Bohn, O.S., 2021. The Revised Speech Learning Model (SLM-r). In: Nathan, G.S., 1987. On second-language acquisition of voiced stops. J. Phon. 15 (4), Wayland, R. (Ed.), Second language speech learning: theoretical and empirical 313–322. progress. Cambridge University Press, New York, NY, pp. 3–83. Ohala, J.J., MacNeilage, P.F., 1983. The origin of sound patterns in vocal tract Flege, J.E., Bohn, O.S., Jang, S., 1997. Effects of experience on non-native speakers’ constraints. The Production of Speech. Springer-Verlag, New York, pp. 189–216. production and perception of English vowels. J. Phon. 25 (4), 437–470. https://doi. Piske, T., MacKay, I.R.A., Flege, J.E., 2001. Factors affecting degree of foreign accent in org/10.1006/JPHO.1997.0052. an L2: A review. J. Phon. 29 (2), 191–215. https://doi.org/10.1006/ Flege, J.E., Eefting, W., 1987. Production and perception of English stops by native jpho.2001.0134. Spanish speakers. J. Phon. 15 (1), 67–83. https://doi.org/10.1016/S0095-4470(19) Sancier, M.L., Fowler, C.A., 1997. Gestural drift in a bilingual speaker of Brazilian 30538-8. Portuguese and English. J. Phon. 25 (4), 421–436. https://doi.org/10.1006/ Fowler, C.A., Sramko, V., Ostry, D.J., Rowland, S.A., Hallé, P., 2008. Cross language jpho.1997.0051. phonetic influences on the speech of French-English bilinguals. J. Phon. 36 (4), Schuhmann, K., Huffman, M. K., 2015. L1 Drift and L2 Category Formation in Second 649–663. https://doi.org/10.1016/j.wocn.2008.04.001. Language Learning. Gorba, C., 2019. Bidirectional influence on L1 Spanish and L2 English stop perception: Schuhmann, K.S., Huffman, M.K., 2019. Development of L2 Spanish VOT before and the role of L2 experience. J. Acoust. Soc. Am. 145 (6), EL587–EL592. https://doi. after a brief pronunciation training session. J. Second Lang. Pronunciation 5 (3), org/10.1121/1.5113808. 402–434. https://doi.org/10.1075/jslp.18018.sch. Gorba, C., 2020. Crosslinguistic influence in L2 speakers: the effect of language Stevens, J., 2001. Study abroad learners’ acquisition of the Spanish voiceless stops. experience on L1 and L2 VOT production and perception [Unpublished PhD thesis MIFLC Rev. 10, 137–151. Universitat Autònoma de Barcelona]. Williams, L., 1977. The perception of stop consonant voicing by Spanish-English Gorba, C., Cebrian, J., 2021. The role of L2 experience in L1 and L2 perception and bilinguals. Percept. Psychophys. 21 (4), 289–297. https://doi.org/10.3758/ production of voiceless stops by English learners of Spanish. J. Phon. 88. BF03199477. Harada, T., 2003. L2 influence on L1 speech in the production of VOT. In: Proceedings of Yang, Y., Chen, X., Xiao, Q., 2022. Cross-linguistic similarity in L2 speech learning: the 15th International Congress of Phonetic Sciences, pp. 1085–1088. evidence from the acquisition of Russian stop contrasts by Mandarin speakers. Hazan, V.L., Boulakia, G., 1993. Perception and production of a voicing contrast by Second Lang. Res. 38 (1), 3–29. https://doi.org/10.1177/0267658319900919. french-english bilinguals. Lang. Speech 36 (1), 17–38. https://doi.org/10.1177/ Zampini, M.L., 1998. The relationship between the production and perception of L2 002383099303600102. Spanish Stops. Tex. Pap. Foreign Lang. Educ. 3 (3), 85. https://mendeley.csuc.cat/f itxers/77294836f95fbc4b90c465f296937509. 108