NOVEMBER 07 2022 Perception of English and Catalan vowels by English and Catalan listeners. Part II. Perceptual vs ecphoric similarity Juli Cebrian J Acoust Soc Am 152, 2781–2793 (2022) https://doi.org/10.1121/10.0014902 CrossMark   View Export Online Citation Related Content Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf Perceptual assimilation of British English vowels to Spanish monophthongs and diphthongs J Acoust Soc Am (January 2019) “If a perceptual problem exists, what are its roots?” Types of cross-language similarity, hypercorrection, and unexplored root canals J Acoust Soc Am (March 2018) Examining the success of three kinds of cross-language similarity in predicting English listeners’ discrimination of Danish vowels J Acoust Soc Am (October 2021) ARTICLE ................................... Perception of English and Catalan vowels by English and Catalan listeners. Part II. Perceptual vs ecphoric similarity Juli Cebriana) Departament de Filologia Anglesa i Germanıstica, Universitat Aut onoma de Barcelona, Facultat de Filosofia i Lletres, Campus Universitat Aut onoma de Barcelona (UAB), Bellaterra 08193, Spain ABSTRACT: Although crosslinguistic similarity is a crucial concept for many disciplines in the speech sciences, there is no clear consensus as to the most appropriate method to measure it. This paper assessed the perceived similarity between English and Catalan vowels by means of an overt direct task evaluating perceptual similarity. The extent to which perceptual similarity is reciprocal is also explored by comparing perceptual judgements obtained by speakers of the two languages involved. Twenty-seven native Catalan speakers and 27 native English speakers rated the perceived dissimilarity between two aurally presented vowel stimuli. Trials included native–non-native pairs as well as native-native pairs to serve as baseline data. Some native–non-native pairs were perceived to be as similar as same-category native pairs, illustrating Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf cases of very high crosslinguistic perceptual similarity. Further, in terms of reciprocity, the results showed a bidirectional- ity in similarity relationships that point to some cases of near-identical or shared categories and also illustrate the role of language-specific cue weighting in determining perceptual similarity. Finally, a comparison with the outcome of a previ- ous study [Cebrian (2021). J. Acoust. Soc. Am. 149(4), 2671–2685], involving the same participants and languages but exploring ecphoric similarity, shows a generally high degree of agreement and a close relationship between the two types C 2022 Acoustical Society of America. https://doi.org/10.1121/10.0014902 of similarity. V (Received 10 April 2022; revised 27 September 2022; accepted 29 September 2022; published online 7 November 2022) [Editor: Ewa Jacewicz] Pages: 2781–2793 I. INTRODUCTION ecphoric similarity (Cebrian, 2021). Sections I A–I D review the previous findings that motivate these goals. One of the main questions that phonetics addresses is the extent to which sounds from different languages resem- A. Importance of crosslinguistic similarity in second ble each other (e.g., Ladefoged and Ferrari Disner, 2012; language speech research Ladefoged and Johnson, 2014). Bohn (2002) stresses how this question, which we can refer to as the issue of crosslin- Most second language (L2) speech models aim to guistic similarity, is crucial for different disciplines within explain the processes that take place in adult L2 speech the speech sciences. Still, there is no general consensus as to learning and make predictions about the difficulty of target the most appropriate method of measuring crosslinguistic language perception and production (see Bohn, 2002, 2017; similarity (Bohn, 2002, 2017; Strange, 2007). The aim of Chang, 2019; Strange, 2007). For example, the speech learn- this paper is to evaluate one method of assessing crosslin- ing model by Flege (1995) and its revised version, the SLM- guistic similarity, namely, a paired comparison technique in r (Flege and Bohn, 2021), claims that adult L2 speech learn- which listeners rate the dissimilarity between a native and a ers may be able to establish accurate or target-like catego- non-native sound (i.e., a rated dissimilarity task or RDT). In ries for L2 sounds if they are able to detect differences addition, the extent to which the degree of perceived (dis)- between L1 and L2 phones. Given enough experience with similarity between pairs of sounds from two different lan- the target language, this is more likely to happen in the case guages is determined by the specific characteristics of each of dissimilar L1 and L2 phones than in the case of similar language is also explored. This is performed by examining L1 and L2 phones (Flege, 1995). Further, according to the the perceived dissimilarity between the same pairs of perceptual assimilation model (PAM) from Best (1995) on Catalan and English vowels by native speakers of each of non-native perception and its adaptation to L2 perception these two languages who performed the same RDT. Finally, (PAM-L2; Best and Tyler, 2007), target language phones the outcome of the current study is compared with the will be perceptually assimilated to (i.e., identified with) the results of a previous study that involved the same languages closest native category and categorized as better or worse and speakers as in this study but used an interlingual identi- exemplars of that category. On the other hand, a target fication approach aimed at analyzing crosslinguistic phone with no counterpart in the L1 may be uncategorized or even heard as nonspeech. Based on the non-native/L2 to L1 perceptual mappings, PAM/PAM-L2 makes a series of a) Electronic mail: juli.cebrian@uab.cat predictions for accuracy of discrimination of non-native or J. Acoust. Soc. Am. 152 (5), November 2022 0001-4966/2022/152(5)/2781/13/$30.00 C 2022 Acoustical Society of America V 2781 https://doi.org/10.1121/10.0014902 L2 segmental contrasts (see Tyler, 2021, for an overview of identify a given stimulus as one from a set of possible PAM’s assumptions and predictions). Similarly, other responses. In addition, while RDT is an overt task in which speech perception and L2 speech learning models draw on the two stimuli to be compared are physically present, PAT the construct of crosslinguistic distance to explain how L2/ is a covert task as the listener is indirectly required to com- non-native sounds are perceptually categorized. The native pare a physical stimulus to a mental category or memory language magnet theory (NLM; e.g., Kuhl et al., 2008) representation. explains that the smaller the phonetic distance between a Bohn (citing Tulving, 1981) suggests that overt and non-native sound and an L1 category prototype, the more covert tasks may be assessing different types of similarity, likely it is that the non-native sound is perceived in terms of “what Tulving refers to as ‘perceptual similarity’ (in overt the native category. Finally, the second language linguistic tasks) as opposed to ‘ecphoric similarity’ (in covert tasks)” perception model (L2LP; e.g., Elvin and Escudero, 2019) (Bohn, 2002, p. 201). Despite the prevalence of PATs in establishes different learning scenarios for non-native/L2 crosslinguistic perception studies [see work by Strange and sounds that depend on the degree of acoustic and/or percep- colleagues (e.g., Strange et al., 1998) and those by Best and tual overlap between native and non-native sounds. colleagues (e.g., Faris et al., 2016), among many others], Therefore, establishing a reliable and principled method of there is no clear consensus as to whether this is the most measuring crosslinguistic phonetic similarity is crucial for suitable technique. This paper examines the perceptual simi- the evaluation of current theories of speech perception and larity between English and Catalan vowels using an overt L2 speech (Bohn, 2002; Strange, 2007). Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf direct approach (Bohn, 2002) in which speakers of both lan- guages involved provide perceptual dissimilarity ratings B. Measuring perceived similarity: Ecphoric vs (DRs) on a number of native–non-native vowel pairs. In the perceptual approaches last part of the paper, these results will be compared to L2 speech researchers have advocated for the use of ecphoric similarity judgements obtained in Cebrian (2021) perceptual approaches over other types of methods of evalu- to assess the relationship between these two types of ating crosslinguistic phonetic similarity, such as articulatory approaches (perceptual similarity vs ecphoric similarity), descriptions and acoustic comparisons (Bohn, 2002; something which remains largely untested, at least in rela- Strange, 2007). This preference is motivated by the fact that tion to L2 speech.1 acoustic analyses are more likely to be affected by factors such as speech rate, phonetically conditioned variability, C. Previous findings involving perceptual similarity and individual variation. In addition, some studies have Few previous studies have examined perceptual similar- reported discrepancies between acoustic and perceptual ity between native and non-native vowels using a RDT or measures (Strange et al., 2004; Strange, 2007). Bohn (2002) similar task (Cebrian et al., 2011; Flege et al., 1994). Flege and Strange (2007) conclude that perceptual tasks constitute et al. (1994) tested two groups of Spanish learners of a more direct method of evaluating crosslinguistic similar- ity, especially if the goal is to evaluate the way L1 percep- English, differing in length of residence in the United States tion influences the categorization of non-native or L2 (1.8 vs 7 years), as well as native English-speaking controls. sounds (Tyler, 2021; Tyler et al., 2014; but cf. Thomson Using a nine-point scale, listeners rated the degree of dis- et al., 2009 for a statistical approach based on acoustic similarity between Spanish-Spanish, English-English, and measures). Spanish-English vowel pairs. The results showed that DRs The two most common perceptual approaches are were influenced by auditory distance as the greater the dis- paired comparisons and interlingual identification (Bohn, tance between vowels in an F1-F2 bark space, the higher 2017). A paired comparison technique, such as a graded dis- were the DRs obtained. However, L2 experience did not crimination or RDT, involves using a Likert scale to indicate seem to play a role as the two groups of Spanish speakers the degree of (dis)similarity between two stimuli presented did not differ in their DRs. Some native–non-native pairs aurally, typically a native and a non-native phone (e.g., received relatively low DRs by all of the groups, such as Flege et al., 1994; Fox et al., 1995). RDT is a direct method English /i/-Spanish /i/ and English /ˆ/-Spanish /a/, although of assessing crosslinguistic similarity as listeners are specifi- the absence of same-category same-language pairs in the cally asked to compare the two stimuli presented and indi- experiment (e.g., E /i/-E /i/) makes it difficult to assess how cate how close they are. An interlingual identification native-like the non-native vowels were perceived to be. approach is exemplified by perceptual tasks known as a Specifically involving Catalan (C) and English (E) vow- forced choice categorization task or perceptual assimilation els, Cebrian et al. (2011) collected perceptual similarity data task (PAT). In a PAT, listeners are presented with exemplars from a group of Catalan learners of English by means of a of non-native segments and asked to identify them in terms RDT that involved pairs of L1-L2 vowels (e.g., CE /a/-æ/) of L1 categories, typically using labels representing the L1 as well as pairs of same-language same-category (e.g., EE sounds, and provide a goodness of fit rating (GR; e.g., /i+/-/i+/, CC /u/-/u/) and different-category (CC /i/-/e/, EE Cebrian, 2021). PAT illustrates an indirect method of /i+/-/I/) vowels. The task used a 7-point scale and was com- assessing similarity as listeners are not directly instructed to pleted by 47 undergraduate students majoring in English. judge the similarity between two sounds but rather to The results showed that the two pairs of CC same-category 2782 J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian https://doi.org/10.1121/10.0014902 vowels obtained the lowest DRs (mean ¼ 1.7 out of 7), fol- time (GR, 5.6). Two more pairs obtained reciprocal but lowed closely by the EE same-category pairs (mean ¼ 2, unequal mappings: E /i+/ and /`/ were more consistently range ¼ 1.7–2.3). CE pairs yielded DRs that ranged from 2.2 assimilated as C /i/ (87%) and /O/ (82%) than the reverse (CE /E/-/E/) to 6.4 (CE /E/-/æ/) out of 7. Thus, some CE pairs (62% and 53%, respectively), and C /o/ and E /O+/ did not yielded DRs within the range of those obtained for receive reciprocal mappings. These discrepancies were EE same-category pairs (CE /E/-/E/, mean ¼ 2.2; /i+/-/i/ and linked to role of context-specific cues such as vowel dura- /O/-/`/, mean ¼ 2.3; followed by /a/-/æ/, mean ¼ 2.5). This tion in L1 perception. Catalan vowels are, on average, simi- shows evidence of a high degree of perceived crosslinguistic lar in duration to English lax vowels. Hence, C /i/ and /o/ similarity. Still, the study had some shortcomings. First, the are closer in duration to E /I/ and /`/ than to E /i+/ and /O+/. stimuli used in the experiments were unmodified /b/ + vowel Thus, the English tense vowels were not consistently + /t/ words (henceforth, /bVt/ words). It is possible that dif- selected as the L1 match for the Catalan stimuli despite ferences in the prevoicing of the /b/ and the release (and being spectrally closer (Cebrian, 2021). Similarly, in the aspiration) of the /t/ may have influenced the listeners’ simi- study by Schmidt (1996, 2007) on crosslinguistic perception larity judgements (see Sec. II A 2). In addition, the RDT task of consonants, the English and Korean listeners generally in Cebrian et al. (2011) contained only a subset of the possi- provided reciprocal patterns of non-native to native map- ble vowel combinations, and the target vowels were not pings. Still, Schmidt also found that crosslinguistic similarity equally distributed across trials. Finally, the listeners in that varied depending on context-specific cues such as variations Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf study were experienced L2 English speakers (undergraduate in stop voice onset time (VOT), the presence of burst-like English majors), and no native English speakers were tested. transitions in nasals, and the degree of labialization of the This study differs from Cebrian et al. (2011) in three main vowels following the target consonants. Schmidt concluded ways: (a) methodological limitations regarding the stimuli that “reciprocal studies … can delineate the relationship and task design have been addressed; (b) participants have between sounds in two languages, as well as the cues used no or minimal experience with the non-native language; and by speakers of each language so that more accurate predic- (c) the perceived similarity between the same set of Catalan tions about cross-linguistic speech perception can be for- and English vowels is elicited from native speakers of the mulated” (Schmidt, 2007, pp. 199–200). This paper extends two languages involved. This reciprocal or bidirectional our previous study by testing the same group of participants approach to perceptual similarity is discussed next. and assessing perceptual (as opposed to ecphoric) similarity. D. Reciprocal assessment of similarity E. The current study Most studies examining crosslinguistic perceived simi- The focus of this study is the perceptual similarity larity obtain similarity judgements from speakers of one of between Catalan and SSBE vowels. SSBE has an inventory the two languages involved. To our knowledge, only a study of 12 vowels (/i+ I E ˘+ æ ˆ A+ ` O+ U u+ @/) and 8 diphthongs, on crosslinguistic perception of consonants by Korean e.g., /aI aU eI @U/ (Cruttenden, 2014). The Central or (Schmidt, 1996) and English (Schmidt, 2007) speakers and Eastern variety of Catalan has an inventory of seven vowels our previous study (Cebrian, 2021) have elicited similarity (/i, e, E, a, O, o, u/), a reduced vowel ([@]) in unstressed posi- from two parallel populations. Cebrian (2021) investigated tion, and a number of diphthongs resulting from the combi- the extent to which speakers of the two languages examined nation of vowels and high glides, e.g., /ai , ei , au, ou /, etc., would yield comparable similarity judgements for the same (Recasens, 1993; see Cebrian, 2021, for an& acoustic & & compar- & pairs of native and non-native vowels. This was done to ison of the two vowel systems). The results of two experi- obtain a more complete picture of the similarity relation- ments examining the perceptual similarity between Catalan ships between two sound systems, examine if sounds from and English vowels are presented. In the first experiment, a two different languages could be considered perceptually group of native Catalan speakers rated the perceived dissim- the same, and explore whether perceptual distance is a ilarity between Catalan and English vowels by means of a shared construct by speakers of the two languages involved. RDT. In the second experiment, the same task was adminis- To this end, native speakers of Catalan and Standard tered to a group of SSBE speakers. Southern British English (SSBE) evaluated the ecphoric similarity of the same set of 11 Catalan and 13 English vow- II. EXPERIMENT 1. PERCEPTUAL SIMILARITY els and diphthongs. Out of the 11 potential parallel matches, BETWEEN SSBE AND CATALAN VOWELS: CATALAN LISTENERS 8 Catalan-English pairs received reciprocal modal responses with assimilation scores equal to or higher than a 70% A. Methodology categorization threshold (Tyler et al., 2014; see Tyler, 2021, 1. Participants for a discussion of categorization thresholds). These were CE /au /-/aU/, /E/-/E/, /a/-/æ/, /ou /-/@U/, /ei /-/eI/, Twenty-seven Catalan-dominant Catalan-Spanish bilin- /ai /-/aI/, /e/-/I/,& and /u/-/u+/. For example, C & /E/ was assimi- & guals (15 females; average age, 24 years old; age range, & lated to E /E/ 92% of the time with an average GR of 4.2 out 17–48 years old) participated in the experiment and were of 7, whereas E /E/ was identified as C /E/ 91% of the mostly undergraduate students. Many had studied some J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian 2783 https://doi.org/10.1121/10.0014902 English at school, where the focus of English language combinations across the total number of same-language and instruction is mostly grammar and vocabulary. Participants different-language pairs. For example, E /æ/ was paired up reported minimal or no immersive experience in English. with C /a/ and /E/ (found to be perceptually close to the Hence, they were considered naive listeners (Best and Tyler, English vowel) and C /ai / to ensure the same number of 2007). All of the participants reported normal hearing and instances of these vowels& throughout the experiment. This were compensated for their participation. design resulted in 42 Catalan-English pairs (36 of these were used in experiments 1 and 2; see Sec. III A 3), and a set 2. Stimuli of same-language (L1-L1) pairs that were specific to each The stimuli consisted of 13 SSBE and 11 Catalan vow- experiment. The L1-L1 pairs included some same-category els and diphthongs. The Catalan stimuli were the seven and different-category pairs, which provided baseline rat- monophthongs /i e E a O o u/2 and four rising diphthongs, ings to compare to the native–non-native pairs. Each vowel namely, /ei ou ai au /. Diphthongs, i.e., combinations of a combination appeared six times, that is, in three different monophthong & &and& a& semivowel (/i / or /u /), were included talker combinations and two orders (e.g., E /æ/ - C /a/ and C & given their reported role in crosslinguistic& perception /a/ - E /æ/). Given the resulting large number of stimuli, two (Cebrian, 2011, 2019; Escudero and Williams, 2011). Still, versions of the task were created, grouping mostly front and to limit the complexity and duration of the task, only those mostly back vowels separately. Set A involved the English diphthongs that had been found to be perceptually close to vowels /aI, æ, eI, E, I, i+, ˘+/ and the Catalan vowels /i e E ei ai a/, whereas Set B included E /u+ @U O+ ` A+ aU ˆ/ and C /u& Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf English vowels in previous studies involving Catalan and & o O au a/. C /a/ appeared in both sets. Table I presents the English (Cebrian et al., 2011; Cebrian, 2021) were selected. ou & of the list & vowel pairs included in each of the subtasks (sets The English stimuli contained the vowels and diphthongs /i+I E ˘+ æ ˆ A+ ` O u+ eI aI aU @U/.3 The vowels were pro- A and B) in experiment 1. The interstimulus interval was duced in a /b/ þ vowel þ /t/ context, a possible sequence in 1.2 ms to encourage the use of phonetic information stored both languages, embedded in a carrier phrase. Phrases were in long term memory (Flege et al., 1994; Pisoni, 1973; produced by three male Catalan and three male English among others). The total number of trials per set was 180 native speakers (average age, 32.5 years old, range 21–44 (30 vowel pairs  3 talker combinations  2 orders). years old; see Cebrian, 2021, for further details). The pro- 4. Procedure duction of stops in Catalan and English may differ in terms of the VOT of initial voiced stops (voice-lead vs short lag The listeners performed two types of perceptual tasks, a VOT) and the intensity of the release and possible aspiration perceptual assimilation task (reported in Cebrian 2021) and in final voiceless stops (Docherty, 1992; Recasens 1993). the RDT. The order of the two sets (A and B) was counter- Thus, stimuli were edited to prevent such crosslinguistic balanced across all of the listeners. Instructions were given differences in consonant production from affecting percep- at the beginning of each task. The listeners were told to pay tual judgements of vowel similarity. Consequently, each attention to the vowel sounds and disregard traces of conso- stimulus started from the release of the /b/ and ended with nants in the stimuli as well as possible voice quality or pitch the beginning of the /t/ closure, therefore, maintaining intact differences. They were also encouraged to use the whole the relevant cues to the vowel (i.e., vowel transitions; range of the scale. The listeners performed the tasks individ- Strange, 1987). ually in a soundproof room. They listened to the stimuli over headphones [Sennheiser (Wedemark, Germany) HD- 3. RDT 25] and selected a rating on a nine-point scale, where one In this task, listeners were presented with a pair of meant “very similar” and nine meant “very dissimilar” by sounds and had to indicate the degree of perceived (dis)simi- clicking on a computer screen. Praat software (Boersma and larity on a continuous nine-point scale, where one meant Weenink, 2018) was used to conduct the experiments. Each very similar and nine meant very dissimilar (Flege et al., task was preceded by a short practice phase to familiarize 1994). Pairing all of the English vowels with all of the participants with the procedure. The practice session con- Catalan vowels with all of the possible talker and order sisted of nine trials (four native–non-native and five native- combinations would have resulted in an extremely large and native pairs). On average, participants took about 10 min to impractical number of stimuli. Hence, a subset of vowel complete each task. combinations were selected, ensuring that all of the crucial B. Results L1-L2 vowel pairs that are adjacent in a perceptual space (following previous findings, e.g., Cebrian et al., 2011; The average DRs for each pair of vowels for each partici- Cebrian, 2021; Rallo Fabra, 2005) were included. The task pant was calculated.4 The degree of inter-rater agreement was designed so that all of the individual vowels appeared among the 27 participants was evaluated first by obtaining the the same number of times throughout the task. Thus, each reliability coefficients (Cronbach’s alpha). A very high degree non-native vowel appeared in three vowel combinations in of agreement was observed (0.990; note that 0.700 is consid- the Catalan-English pairs and every single Catalan and ered the lower limit for evidence of reliability; Cortina, 1993). English vowel appeared in the same number of This result indicates that the listeners were very consistent in 2784 J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian https://doi.org/10.1121/10.0014902 TABLE I. The vowel pairs included and number of trials in each RDT in experiment 1. Number of vowel pairs per subtask RDT A (mostly front vowels) RDT B (back and rounded vowels) 5 L1 Catalan same-category vowel pairs /a/-/a/, /ai /-/ai /, /ei /-/ei /, /E/-/E/, /i/-/i/ /a/-/a/, /au /-/au /, /ou /-/ou /, /O/-/O/, /u/-/u/ & & & & & & & & 4 L1 Catalan different-category vowel pairs /ai /-/a/, /ei /-/e/, /E/-/e/, /i/-/e/ /a/-/O/, /o/-/O/, /o/-/ou /, /u/-/au / & & & & 21 Catalan-English vowel pairs /a/-/æ/, /E/-/æ/, /ai /-/æ/, /a/-/A+/, /au /-/A+/, /O/-/A+/, & /a/-/E/, /E/-/E/, /e/-/E/, & /o/-/O+/, /O/-/O+/, /ou /-/O+/, /ai /-/eI/, /e/-/eI/, /ei /-/eI/, /a/-/aU/, /au /-/aU/, /ou & /-/aU/, &/e/-/I/, /ei /-/I/, /i/-/I/, & & & /O/-/`/, /o/-/`/, /a/-/`/, & /ei /-/i+/, /i/-/i+/, /ai-/i+/, /u/-/u+/, /ou-/u+/, /o/-/u+/, /ai&/-/aI/, /i/-/aI/, /a/-/aI/, /ou /-/@U/, /o/-/@U/, /u/-/@U/, & /a/-/˘+/, /E/-/˘+/, /e/-/˘+/ &/a/-/ˆ/, /u/-/ˆ/, /au /-/ˆ/ & their assessments of perceived dissimilarity. Further, the p < 0.001). The cases of discrepancy involved pairings with potential effect of task order, that is, front vowels (set A) first vowel /I/. Participants in both studies found E /I/ to be per- or back vowels (set B) first, was examined by comparing the ceptually more similar to C /e/ than to C /i/, but the differ- average DRs for each vowel pair obtained by each order sub- ence was considerably larger in the current study (/e/-/I/, 2.3 Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf group (AB, BA) and for each subset. The results of a general- vs /i/-/I/, 5.6) than in the previous study ized linear mixed model with vowel set (A or B), task order (/e/-/I/, 3.7 vs /i/-/I/, 4.3).5 (AB or BA), their interaction as fixed factors, and a random intercept for participant indicated that the order in which the C. Discussion tasks were completed did not affect the average DRs for either subset of vowels [task order, F(1,9554) ¼ 0.039, Experiment 1 examined the perceptual similarity p ¼ 0.844; vowel set, F(1,9554) ¼ 0.059, p ¼ 0.809; task between pairs of Catalan and SSBE vowels from the per- order  vowel set, F(1,9554) ¼ 3.002, p ¼ 0.083)]. In spective of Catalan-speaking listeners. The results show addition, the results for the two order subgroups were that, generally, for each of the English vowels, there is a sin- very highly correlated (r ¼ 0.974, p < 0.001, N ¼ 60). SPSS gle native vowel that is perceived as least dissimilar, e.g., software (IBM Corp, 2021) was used for the statistical CE /E/-/E/, /e/-/I/, /i/-/i+/, /au /-/aU/, /o/-/O+/, /ei /-/eI/, /ai /-/aI/, & /ou /-/@U/, and /u/-/u+/, indicating & a systematic & of pattern analyses. & non-native to native category mapping. However, on two Figure 1 displays the average DRs per vowel pair, grouped by language combination, i.e., CC and CE, and occasions, the same native vowel was chosen as the most vowel set, i.e., front vowels (set A) and back vowels (set B). similar to two English vowels with varying DR values: C /a/ The average DR for CC same-category pairs was two, rang- for E /æ/ (DR, 2.4) and /ˆ/ (3.2), and C /O/ for E /`/ (3.3) ing from 1.4 to 2.5, with a standard deviation of 1.2. The CC and /A+/ (4.3). This outcome points to a potential difficulty different-category pairs obtained ratings ranging from 4.4 on the part of Catalan speakers to distinguish these pairs of (/O/-/o/) to 8 (/u/-/au /). Regarding the CE pairs, the DRs English vowels (/æ/-/ˆ/, /`/-/A+/). Finally, E /˘+/ appears to ranged from 2.3 to &8, showing great variation among CE have no clear match in the L1 as it was consistently rated as combinations in the degree of perceptual similarity. Three rather dissimilar from all of the L1 vowels it was paired of the CE pairs received ratings within the range of those with (DRs, 6.3–6.9). The more dissimilar that the non-native obtained for CC same-category pairs, namely, CE /E/-/E/, vowels are from the closest L1 vowel (e.g., E /˘+/), the more /e/-/I/, and /a/-/æ/, with ratings of 2.3, 2.3, and 2.4, respec- likely it is that Catalan learners of English detect differences tively. Three more pairs obtained ratings that were equal or between them and ultimately establish a separate more lower than 3.2, that is, within one standard deviation (SD) of target-like category for the non-native vowel given sufficient the mean rating for CC same-category pairs (i.e., mean experience with the target language (Flege, 1995; Flege and 2 þ SD 1.2 ¼ 3.2): CE /i/-/i+/, /au /-/aU/, and /a/-/ˆ/, with rat- Bohn, 2021). By contrast, the fact that some CE pairs & ings of 2.8, 2.9, and 3.2, respectively, whereas /O/-/`/ obtained DRs within the range of those received by same- received the following lowest rating (3.3). category L1 vowels (e.g., CE /E/-/E/, /e/-/I/, /a/-/æ/, /i/- These results follow the general trends reported by /i+/, /au /-/aU/) suggests that those English vowels may be Cebrian et al. (2011), a previous study that also looked at readily&assimilated to existing native categories. In cases of SSBE and Catalan vowels but with some methodological very close similarity (near-identity), the non-native vowels differences (see Sec. I C). This study examined a total of 60 may be perceived and produced in terms of the correspond- vowel combinations (18 CC pairs and 42 CE pairs), which ing native categories by Catalan learners of English, and the included the 7 CC and 16 CE pairs tested in the previous use of Catalan categories may go undetected by English lis- study. The results of a Pearson correlation conducted on the teners. Testing all of these possibilities lies beyond the scope DRs obtained for the common 23 pairs yielded a high level of this paper. Still, analyzing the perception of Catalan vow- of agreement between the two studies (r ¼ 0.901, N ¼ 23, els by native English speakers (experiment 2) can shed light J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian 2785 https://doi.org/10.1121/10.0014902 Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf • FIG. 1. The mean DRs for Catalan-Catalan ( ) and Catalan-English (䊏) vowel pairs elicited from Catalan L1 speakers. Set A vowels are presented on the left, and set B vowels are presented on the right. on the degree to which some Catalan and English vowels are of Spanish learners of English differing in amount of L2 perceptually similar or even near-identical. experience (years of residence) did not differ in their DRs Regarding the comparison to previous studies, the for pairs of Spanish and American English vowels. Rallo results of the current experiment are strongly correlated with Fabra and Romero (2012), another study looking at Catalan the results reported by Cebrian et al. (2011), also involving and English vowels, did not specifically elicit perceptual SSBE and Catalan vowels. The two studies only showed dissimilarity but included seven Catalan-English pairs in a some differences with pairs involving E /I/. Whereas in this categorical oddity discrimination task that also included study, E /I/ was heard as notably closer to C /e/ than to C /i/ English-English pairs. Although the English stimuli repre- (DRs of 2.3 vs 5.6), in Cebrian et al. (2011), the difference sented American English, the results are, on the whole, in between the DRs for CE /e/-/I/ and CE /i/-/I/ was smaller line with the current RDT results (DRs). The CE pair /i/-/I/ (3.7 vs 4.3, respectively). One possible explanation for this (current DR, 5.6) was generally better discriminated than may stem from the fact that unlike naive listeners, the L2 /E/-/E/ (2.3), /a/-/æ/ (2.4), /i/-/i+/ (2.8), /a/-/ˆ/ (3.2), and speakers in Cebrian et al. (2011) may have had orthographic /a/-/A+/ (5), and the latter three pairs tended to be better dis- and metalinguistic knowledge about English vowels. This criminated than /E/-/E/ and /a/-/æ/. knowledge may have resulted in crosslinguistic perceptual In sum, the results of experiment 1 have illustrated dif- mappings that are more phonologically than phonetically ferent patterns of perceptual similarity between English and motivated (Chang, 2019; Holliday, 2015). Specifically, the Catalan vowels, from very high degrees of similarity to E /I/, which is acoustically and perceptually closest to C /e/ vowels with no clear match in the L1. The extent to which (Cebrian, 2021), may be perceived by Catalan L2 English these patterns will be replicated in the parallel language situ- learners as less dissimilar to C /i/ given the learners’ meta- ation, native English speakers’ perception of Catalan, is linguistic knowledge [e.g., /I/ is commonly spelled with let- explored next. ter hii, which corresponds to C /i/, and is often learned in opposition to E /i+/, being often (incorrectly) described as a III. EXPERIMENT 2. PERCEPTUAL SIMILARITY “short /i/”]. In any case, the remaining pairs obtained very BETWEEN SSBE AND CATALAN VOWELS: ENGLISH LISTENERS consistent results across studies. This consistency was not unexpected as the stimuli in both studies had the same origin A. Methodology and both were elicited in bVt words produced by the same 1. Participants talkers even if in the current study, neighbouring consonants were edited to minimize consonant differences (see Sec. The participants were 27 native speakers of SSBE (23 II A 2). The fact that the current participants were naive lis- females), who were undergraduate (19) and graduate (8) teners while those in Cebrian et al. (2011) were L2 learners students in the Division of Psychology and Language does not seem to have affected the overall results. This is in Sciences at University College London (average age, line with the finding by Flege et al. (1994) that two groups 22 years old; range, 18–30 years old). They had been raised 2786 J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian https://doi.org/10.1121/10.0014902 in Southern England or born to parents from Southern interaction as fixed factors and a random intercept for partic- England. Most spoke no other language fluently except for ipant indicated no effect of task order on the average DRs one subject who was fluent in German and another with for either subset of vowels [task order, F(1,9566) ¼ 2.7, passive knowledge of Spanish. All of the participants p ¼ 0.1; vowel set, F(1,9566) ¼ 0.705, p ¼ 0.401; task order reported normal hearing and were paid for their  vowel set, F(1,9566) ¼ 0.787, p ¼ 0.375]. Finally, the participation. results for the two order subgroups were very highly corre- lated (r ¼ 0.988, N ¼ 58, p < 0.001). 2. Stimuli Figure 2 presents the average DRs obtained for each pair of vowels, including the L1-L1 English pairs and The stimuli were the same as in experiment 1. native–non-native CE pairs, organized by vowel set. The English same-category pairs obtained an average DR of 1.5, 3. RDT ranging from 1.2 to 1.9, with a standard deviation of one, The design of the RDT used in experiment 2 followed whereas the L1 different-category pairs obtained ratings the same criteria as in experiment 1. A total of 36 Catalan- ranging from 5.3 (/i+/-/I/) to 8.1 (/˘+/-/I/, /˘+/-/aI/, and /˘+/-/i+/). English pairs were examined, which were also included in The ratings obtained for the CE pairs ranged from 1.8 to 7.7, experiment 1 (the RDTs in experiment 1 had 3 more CE showing that some native–non-native vowel pairs were per- pairs given the greater number of vowels in English). Each ceived to be as close as some L1-L1 pairs. Specifically, four pairs were within the range of the values for same-category Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf Catalan vowel appeared in combination with an English vowel three times (e.g., C /e/-E /I/, C /e/-E /eI/, C /e/-E /E/). English pairs: CE /a/-/æ/, /E/-/E/, /au /-/aU/ and /e/-/I/, all four Also included were 22 English-English pairs, 10 same- with a rating of 1.8. In addition, the&DR for CE /ei /-/eI/ (2.4) was also within one standard deviation of the average & rating category pairs, and 6 different-category pairs, yielding a total of 58 vowel pairs, which is almost equal to the number for same-category L1 pairs (i.e., mean 1.5 þ SD 1 ¼ 2.5), fol- of pairs used with the Catalan listeners (60). Similar to lowed closely by CE /ou /-/@U/ and /i/-/i+/ with 2.6 and 2.7, respectively. The remaining& CE pairs received ratings that experiment 1, two subtasks were created, one for (mostly) front vowels and one for back and rounded vowels. Table II were above five. lists all of the vowel pairs tested in experiment 2. The total number of trials per subtask was 174 (29 vowel pairs  3 C. Discussion talker combinations  2 orders). Unlike experiment 1, experiment 2 is the first study to report dissimilarity judgements between English and 4. Procedure Catalan vowels elicited from L1 English speakers, hence, The procedure for experiment 2 was the same as the there are no previous results to compare the current results one described for experiment 1 in Sec. II A 4. to. As was observed with the Catalan listeners, most non- native (Catalan) vowels were perceived as closest to a dif- B. Results ferent L1 (English) vowel, namely, CE /a/-/æ/, /E/-/E/, /au /-/aU/, /e/-/I/, /ei /-/eI/, /ou /-/@U/, /i/-/i+/, /ai /-/aI/, and As was done with the results for the Catalan listeners, & /u/-/u+/, some of which& received & DRs comparable & to those the degree of agreement among the 27 listeners in their DRs given to same-category EE pairs. The exceptions were the was examined first. The reliability coefficient (Cronbach’s Catalan high mid and low mid back rounded vowels (/o/ and alpha) revealed a very high degree of inter-rater agreement /O/, respectively). Given the height difference between the (0.992), confirming the very high level of consistency in lis- Catalan back rounded vowels, it could be expected that teners’ responses, as was also observed in experiment 1. English listeners would find C /o/ to be least dissimilar to Next, the potential effect of task order (subtask A followed the English mid back rounded vowel /O+/ and C /O/ to the by subtask B or the reverse) was examined. As with experi- English low vowel /`/.6 However, both Catalan vowels were ment 1, the results of a generalized linear mixed model with found to be least dissimilar to E /`/, followed by E /O+/. In vowel set (A or B), task order (AB or BA), and their fact, this finding replicates the results in Cebrian (2021) as TABLE II. The vowel pairs included and number of trials in each RDT in experiment 2. Number of vowel pairs per subtask RDT A (mostly front vowels) RDT B (back and rounded vowels) 5 L1 English same-category vowel pairs /æ/-/æ/, /aI/-/aI/, /eI/-/eI/, /E/-/E/, /i+/-/i+/ /A+/-/A+/, /ˆ/-/ˆ/, /@U/-/@U/, /aU/-/aU/, /u+/-/u+/ 6 L1 English different-category vowel pairs /aI/-/˘+/, /eI/-/aI/, /E/-/æ/, /i+/-/˘+/, /I/-/˘+/, /i+/-/I/ /@U/-/O+/, /aU/-/O+/, /aU/-/u+/, /A+/-/`/, /ˆ/-/`/, /u+/-/`/ 18 Catalan-English vowel pairs /a/-/æ/, /a/-/˘+/, /a/-/E/, /a/-/A+/, /a/-/ˆ/, /a/-/aU/, /E/-/æ/, /E/-/E/, /E/-/˘+/, /O/-/A+/, /O/-/O+/, /O/-/`/, /ai /-/æ/, /ai /-/eI/, /ai /-/aI/, /o/-/O+/, /o/-/`/, /o/-/@U/, &/e/-/E/, /e/-/I/, & & /e/-/eI/, /ou /-/O+/, /ou-/u+/, /ou /-/@U/, /ei /-/eI/, /ei /-/I/, /ei /-/i+/, /au& /-/A+/, /au /-/aU/, /au & /-/ˆ/ &/i/-/I/, /i/-/i+/, & & /i/-/aI/ & & & /u/-/u+/, /u/-/@U/, /u/-/ˆ/, J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian 2787 https://doi.org/10.1121/10.0014902 Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf • FIG. 2. The mean DRs for English-English ( ) and Catalan-English (䊏) vowel pairs elicited from English L1 speakers. Set A vowels are presented on the left, and set B vowels are presented on the right. discussed below. A possible explanation may lie in the role overall, very consistent. This was confirmed by the results that duration may play as a phonetic cue to vowel identity of a Pearson correlation (r ¼ 0.847, N ¼ 36, p < 0.001) and for native English speakers. SSBE /O+/ has a notably longer is illustrated in Fig. 3. Despite the overall consistency, duration than either C /o/ or /O/, both of which are close in English listeners tended to provide lower DR (mean ¼ 4.7, duration to SSBE /`/. English speakers’ sensitivity to dura- SD ¼ 1.75) than Catalan listeners (mean ¼ 5.1, SD ¼ 1.57). tion as a cue to vowel identity has been reported in previous This difference between the two groups reached significance studies (e.g., Fox et al., 1995). Thus, despite C /o/’s greater on a paired samples t-test [t(35) ¼ 2.98, p ¼ 0.005]. spectral proximity to E /O+/, the mismatch in duration may have rendered it perceptually closer to E /`/ (Cebrian, TABLE III. The DRs obtained by the Catalan listeners in experiment 1 (Cat 2021). The possible predictions for English learners of L) and English listeners in experiment 2 (Eng L) for each of the 36 Catalan- Catalan are along the lines of what was described above for English vowel pairs common to both experiments (ordered from lowest to Catalan learners of English. Some Catalan vowels, particularly highest for the Cat L column). those in CE pairs that received DRs within the range of EE CE pair Cat L’s DR Eng L’s DR CE pair Cat L’s DR Eng L’s DR same-category pairs, are likely to be readily assimilated to their English counterparts. The higher the DR between a native and /E/-/E/a 2.3 1.8 /ai /-/æ/ 5.5 6.3 & a non-native vowel, the more likely it is that differences /e/-/I/a 2.3 1.8 /o/-/oU/ 5.5 4.9 between them are detected and the non-native vowel is eventu- /a/-/æ/a 2.4 1.8 /ei /-/I/ 5.5 5.8 & ally categorized in a more target-like manner. Again, making /i/-/i+/b 2.8 2.6 /a/-/aU/ 5.6 6.3 finer predictions and evaluating them lies beyond the scope of /au /-/aU/b 2.9 1.8 /i/-/I/ 5.6 3.9 & /a/-/ˆ/b 3.2 4.0 /ai /-/eI/ 5.7 4.2 this paper (but see Cebrian, 2021 for some general predictions). & /O/-/`/ 3.3 3.0 /O/-/O+/ 5.7 4.2 This study is concerned with the degree to which the findings /o/-/O+/ 3.9 4.1 /ei /-/i+/ 5.9 4.7 for the two language situations are reciprocal. This is analyzed /ei /-/eI/c 4.0 2.4 & /au /-/A+/ 6.0 6.7 in more detail in Sec. IV. & & /ai /-/aI/ 4.1 3.0 /a/-/˘+/ 6.3 7.7 & /ou /-/oU/ 4.3 2.7 /e/-/eI/ 6.4 5.6 IV. BIDIRECTIONAL PERCEPTUAL SIMILARITY & /O/-/A+/ 4.3 4.9 /E/-/˘+/ 6.9 7.4 /u/-/u+/ 4.5 3.3 /u/-/oU/ 7.0 5.5 A. Comparison of the results of experiments 1 and 2 /ou /-/O+/ 4.7 5.4 /ou /-/u+/ 7.0 6.6 & & Experiments 1 and 2 shared the majority of the CE /o/-/`/ 4.9 3.3 /a/-/E/ 7.1 6.1 vowel pair stimuli (36 pairs; see Secs. II A and III A for /a/-/A+/ 5.0 5.7 /E/-/æ/ 7.4 6.1 details). Thus, the perceived (dis)similarity between Catalan /au /-/ˆ/ 5.2 6.4 /u/-/ˆ/ 7.6 6.5 & /e/-/E/ 5.3 4.3 /i/-/aI/ 8.0 6.9 and English vowels obtained from native speakers of each language can be compared. Table III presents the average a DR within the range of EE and CC ratings. DR per vowel and group. As can be observed, despite some b DR within one SD of the mean for CC pairs. c isolated differences, the ratings across listener groups are, DR within 1 SD of the mean for EE pairs. 2788 J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian https://doi.org/10.1121/10.0014902 lines, as discussed above, the fact that Catalan vowels are similar in duration to English lax vowels may render pairs of a lax English vowel and a Catalan vowel (e.g., CE /i/-/I/, /o/-/`/) less dissimilar to English listeners, who have been found to be sensitive to duration as a phonetic cue to vowel identity (e.g., Fox et al., 1995), than to Catalan listen- ers. The latter, on the other hand, may be focusing on spec- tral differences only because duration is not a relevant cue in Catalan. These facts may underscore the role of L1 pho- netic cue weighting in non-native vowel perception (Cebrian, 2021; Schmidt, 2007). For instance, Schmidt (1996, 2007) reported that Korean and English speakers tended to agree in their perception of similarity between Korean and English consonants with some exceptions. The exceptions involved cases where the two languages differ in the way some phonetic cues are weighted (e.g., variations in the amount of labialization of the following vowel, VOT, or Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf presence of burst-like transitions in nasals). In brief, while the strong correlation of the results of experiments 1 and 2 FIG. 3. The scatter plot of the average DRs obtained by the Catalan listen- illustrates that overall cross-language perceptual similarity ers (experiment 1) and English listeners (experiment 2) on 36 common Catalan-English vowel pairs. may be independent of the listeners’ L1, there is also evi- dence that the characteristics of the L1 may influence the degree of perceived similarity. Obtaining perceived similar- B. Discussion ity judgements from the two populations involved allows for a better understanding of the similarity between two sounds The results of experiments 1 and 2 show a high degree systems and helps evaluate the factors that affect the percep- of consistency, as evidenced by the strong correlation. The tion of similarity for each group of speakers. five CE pairs that obtained the lowest DRs were the same for both groups, namely, CE /a/-/æ/, /E/-/E/, /e/-/I/, /au /-/aU/, V. PERCEPTUAL VS ECPHORIC SIMILARITY and /i/-/i+/, whose values were within the range or &within A. Comparison of current results with Cebrian (2021) one SD of the mean of same-category L1-L1 pairs. This indicates that some non-native vowels are perceived as rep- The participants in this study also participated in a pre- resentative exemplars of native categories by both popula- vious study examining ecphoric similarity between Catalan tions, pointing to the possibility of shared or near-identical and SSBE vowels (Cebrian, 2021). Thus, the DRs obtained categories across languages. The two groups also tended to in this study can be compared to the patterns of crosslinguis- agree about the remaining vowel pairs, which were per- tic mappings that were revealed in the previous study. ceived to be gradually more dissimilar. Despite the overall Tables IV (Catalan listeners) and V (English listeners) pre- consistency, English listeners tended to provide lower DR. sent for each non-native vowel, the most similar native For example, in the case of CE diphthongs (e.g., /ou /-/@U/, vowel in terms of the lowest DR in the current study, along- /ai /-/aI/, /ei /-/eI/), English speakers’ DRs were over & one side the percent assimilation to a native category and the & & point lower than those of Catalan listeners. CE /ei /-/eI/, in corresponding GRs observed in the PAT. The comparison is fact, received a DR which was as low as same-category & EE limited to assimilations above chance level in the PAT study values. It is possible that greater exposure to within- (16.7% for Catalan listeners and 12.5% for English listen- category variation in English (due to phonetic context or ers).7 In fact, in all of the cases, the lowest DRs obtained in dialectal differences) may make English listeners more the RDT involve a pair made up of a native vowel and the accepting of less canonical variants and, thus, provide lower non-native vowel that received the highest percent assimila- ratings for non-native–native pairs. For example, more L1 tion in the PAT (i.e., the modal response). For example, variability in off-glide duration (e.g., diphthongs in closed E /i+/ was perceived by Catalan listeners as the most similar vs open syllables) may explain the lower DR given to CE to C /i/ in the RDT (with the lowest DR among all of the diphthong combinations by English listeners in general. pairs involving C /i/ and an English vowel, namely, 2.8), Further, a less central more rounded beginning for E /@U/ and was very consistently assimilated to C /i/ in the PAT (hence, closer to C /ou /) can be found in some L1 phonetic (87% of the time with a GR of 5.7 out of 7). Similarly, C /a/ contexts (e.g., preceding & /l/) and in some varieties of was rated by English listeners as the least dissimilar to E /æ/ English. This may render C /ou / perceptually more native- (DR ¼ 1.8) and had E /æ/ as the modal response in the PAT like to English ears than E /@U/&is to Catalan ears (Cebrian, (86% assimilation, GR ¼ 5.3). In cases when a non-native 2021) and, therefore, CE /ou /-/@U/ may be judged to be vowel was assimilated to more than one native vowel above closer by SSBE than by Catalan & listeners. Along similar chance in the PAT, the difference in percent assimilation is J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian 2789 https://doi.org/10.1121/10.0014902 TABLE IV. A comparison of RDT (CE vowel pairs) and PAT (E to C TABLE V. A comparison of RDT (CE vowel pairs) and PAT (C to E assimilation) results obtained by the same 27 Catalan speakers in the cur- assimilation) results obtained from the same 27 English speakers in the cur- rent study and Cebrian (2021). For the latter, “>70%” indicates assimila- rent study and Cebrian (2021). For the latter, “>70%” indicates assimila- tions above a categorization threshold of 70% (Tyler et al., 2014), and tions above a categorization threshold of 70% (Tyler et al., 2014), and “<70%” and “<50%” indicate assimilations above chance level (16.7%) “<70%” and “<50%” indicate assimilations above chance level (12.5%) but below 70%/50%; GR ¼ goodness of fit ratings. but below 70%/50%; GR ¼ goodness of fit ratings. Non-L1 PAT results (Cebrian, 2021) Non-L1 PAT results (Cebrian, 2021) vowel Native vowel RDT results vowel Native vowel RDT results (English) (Catalan) DR Percent assimilation GR (out of 7) (Catalan) (English) DR Percent assimilation GR (out of 7) /E/ /E/ 2.3 >70% >5 /E/ /E/ 1.8 >70% >5 /I/ /e/ 2.3 >70% >5 /au / /aU/ 1.8 >70% >4, <5 & /æ/ /a/ 2.4 >70% >5 /a/ /æ/ 1.8 >70% >5 /i+/ /i/ 2.8 >70% >5 /e/ /I/ 1.8 >70% ¼5 /aU/ /au / 2.9 >70% >4, <5 /E/ 4.3 <50% <4 & /ˆ/ /a/ 3.2 >70% >4, <5 /ei / /eI/ 2.4 >70% ¼5 & /`/ /O/ 3.3 >70% ¼5 /i/ /i+/ 2.6 <70% >4, <5 /O+/ /o/ 3.9 >70% >4, <5 /I/ 3.9 <50% >4, <5 /eI/ /ei / 4 >70% >4, <5 /ou / /@U/ 2.7 >70% ¼5 & & Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf /ai / 5.7 <50% >4, <5 /ai / /aI/ 3 >70% >5 & & /aI/ /ai / 4.1 >70% <4 /O/ /`/ 3 <70% ¼5 & /@U/ /ou / 4.3 >70% <4 /O+/ 4.2 <50% >4, <5 & /u+/ /u/ 4.5 >70% <4 /u/ /u+/ 3.3 ¼70% >4, <5 /A+/ /O/ 4.3 <70% >4, <5 /o/ /`/ 3.3 <70% >4, <5 /a/ 5 <50% <4 /O+/ 4.1 <50% <4 reflected in the difference in DRs in the RDT. For instance, Furthermore, the reciprocity between English and Catalan Catalan listeners rated CE /A+/-/O/ as less dissimilar than CE listeners’ perceptual judgements reported in Sec. IV was /A+/-/a/ (DRs of 4.3 and 5, respectively); in the same fashion, also observed in Cebrian (2021). Regarding E /˘+/, this E /A+/ was assimilated to C /O/ and C /a/ 54% and 31% of vowel was perceived to be highly dissimilar from any L1 the time, respectively, in the PAT. Similarly, English listen- Catalan vowel in the current study (/a/-/˘+/, 6.3; /e/-/˘+/, 6.7; ers perceived C /O/ to be more similar to E /`/ than to E /O+/ /E/-/˘+/, 6.9). This vowel was not tested in Cebrian (2021), in the RDT (DRs, 3 and 4.2, respectively) and PAT (53% vs but the current result is consistent with the results of the 33% assimilation, respectively). PAT in Cebrian et al. (2011), which revealed no clear pat- In the case of the English listeners (Table V), the mag- tern of assimilation for E /˘+/, with responses being split nitude of the similarity was not always comparable across between C /e/ (30%), /E/ (24%), and /a/ (23%) and low GRs tasks and vowel pairs. For instance, CE /e/-/I/ was among (1.4–2 out of 7). the five CE pairs that obtained the lowest DR (1.8), yet, its assimilation scores in the PAT were comparatively lower B. Discussion than those in the other four (73% vs 86%–94%), possibly because C /e/ was also perceived to be similar to another L1 The comparison of the results of this study and the out- vowel (E /E/, 24%). Similarly, CE /i/-/i+/ obtained a rela- come of Cebrian (2021) shows a close link between the two tively low DR (2.6) but the assimilation of C /i/ to E /i+/ did types of similarities, ecphoric and perceptual. This is evident not reach the categorization threshold of 70% as it was split in the strong negative correlations between their respective between E /i+/ and /I/ (62% and 37%, respectively). Still, the measures: the higher the degree of assimilation of a non- patterns observed in the PAT are reflected by the DRs native vowel to a native category and the higher the GR, the obtained in the RDT (CE /e/-/I/, 1.8 vs /e/-/E/, 4.3, and CE less dissimilar the two vowels are perceived to be. These /i/-/i+/, 2.6 vs /i/-/I/, 3.9). Therefore, there is a general pattern results are in line with the outcome of Cebrian et al. (2011), of consistency across the two tasks. This consistency was who also found a high level of agreement between percep- confirmed by the results of a series of Spearman’s correla- tual and ecphoric similarity judgements obtained from a tion tests involving the DR obtained for each CE pair in the group a Catalan learners of English (20 participants com- RDT and the assimilation percentages and GRs yielded for pleted a PAT, and 47—the previous 20 plus 27 more— the corresponding crosslinguistic mapping obtained in the completed a RDT). Although the RDT in that study did not PAT (Catalan listeners, DR, and percent assimilation, q include the same amount of vowel combinations as in the ¼ –0.860, N ¼ 43, p < 0.001; and DR and GR, q ¼ –0.733, current study, the results did show that the English vowels N ¼ 43, p < 0.001; English listeners, DR, and percent assim- that were perceived to have a more consistent match in the ilation, q ¼ –0.940, N ¼ 42, p < 0.001; and DR and GR, q L1 (e.g., E /æ/ to C /a/, E /i+/ to C /i/, E /E/ to C /E/, and E /`/ ¼ -0.787, N ¼ 42, p < 0.001). The lower the DR in the RDT, to C /O/) corresponded to the L2-L1 pairs that obtained the the higher the assimilation rate and GR in the PAT. lowest DRs. 2790 J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian https://doi.org/10.1121/10.0014902 Given the high level of cross-task consistency observed, previous study indicate that both approaches are highly the comparison of the current and previous results sheds lit- comparable and strongly related. However, it remains to be tle light on the question of which of the two types of meth- examined which type makes better predictions for L2 learn- ods constitutes a better approach to measuring ing. Previous studies have shown that PATs and RDTs can crosslinguistic similarity, at least on the grounds of a poten- be used to make predictions about L2/non-native sound cat- tial difference in their results. L2 speech researchers have, egorization (e.g., Fox et al., 1995; Strange et al., 2011; however, argued for one or the other approach. Flege and Tyler et al., 2014). However, to our knowledge, hardly any Bohn (2021) support a perceptual similarity measure argu- previous studies have actually contrasted the predictions of ing that speakers may differ in the nature of their L1 catego- both methods. One such study is Bohn and Garibaldi (2021), ries and, thus, ecphoric comparisons may not be equivalent who examined which of three methods (ecphoric similarity across individuals. They add that PATs may also accentuate using a PAT, perceptual similarity from a RDT, or an acous- differences among speakers because they involve two dis- tic comparison) best predicted non-native discrimination. tinct actions (identifying the stimulus in terms of a native Ten SSBE native speakers performed a PAT and RDT category and passing a judgement on how good an exemplar involving four SSBE and four Danish front unrounded of that category the stimulus is). Flege and Bohn (2021) vowels and were tested on their ability to discriminate the claim that these potential inter-speaker differences may be Danish vowels. The results indicated that ecphoric and per- better controlled using a single-action task like RDT. Inter- ceptual similarity made different predictions for some vow- speaker differences are also mentioned by Strange (2007) Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf els and none of the three methods fully predicted accuracy but as potentially problematic for a perceptual similarity of non-native vowel discrimination. Still, that study was measure. Strange points out that in a RDT, listeners directly limited to four native and non-native vowels. Perhaps, a bet- compare non-native categories to L1 categories produced by ter approximation to the relationship between the two mea- L1 speakers other than themselves, and these L1 stimuli sures could be obtained if a greater representation of the may differ from the listeners’ own L1 categories. Our results native and non-native systems is assessed. do not show a greater likelihood of inter-speaker variability in one or the other task. As reported above, there was a very VI. GENERAL DISCUSSION AND CONCLUSIONS high degree of inter-rater agreement in experiments 1 and 2 (Cronbach’s alpha values of 0.990 and 0.992, respectively). This study has presented the results of two experiments This was, in fact, also true of the PAT results in Cebrian specifically designed to assess the perceptual similarity (as (2021) with inter-rater reliability coefficients for assimila- opposed to ecphoric similarity) between Catalan and SSBE tion percentages and GRs ranging from 0.920 to 0.956. This vowels elicited from two parallel populations (SSBE and high degree of inter-rater agreement lends little support to a Catalan native speakers) that performed a RDT. This task potential disadvantage of PATs (or RDTs) due to a possible illustrates a direct overt method (Bohn, 2002) in which lis- greater inter-speaker variability. teners are presented with a pair of sounds involving a native Tyler (2021), on the other hand, argues that a PAT and non-native stimulus (overt method) and specifically allows researchers to assess precisely how consistent percep- asked (direct method) to assess the similarity between the tual mappings are across speakers and stimuli. In addition, two. Crucially, the RDT also included pairs of same- Tyler favours a PAT approach as it provides more informa- category and different-category native stimuli for baseline tion about crosslinguistic perception than a RDT does (e.g., purposes. The results of the two experiments show that, gen- whether non-native sounds are perceived as speech or not or erally, every non-native vowel has been found to be per- if they are consistently mapped onto a single native category ceived as most similar to a given native vowel. One or more). Moreover, PATs allow listeners to make their own exception is E /˘+/, which seems to be dissimilar from all choices about what non-native phones are closer to native Catalan vowels. The other exceptions are pairs of non- phones. By contrast, in a RDT, it is the researcher who pro- native vowels perceived as most similar to the same native vides the listeners with the paired stimuli; thus, to represent vowel, namely, E /æ/ and /ˆ/ as C /a/, E /A+/ and /`/ as C /O/, all possible “choices,” the experiment needs to include the and C /o/ and /O/ as E /`/. In the first two cases, this is whole list of possible combinations of stimuli. Strange (2007) related to the smaller size of the Catalan inventory and con- refers to this as a disadvantage of RDTs, highlighting the pro- sequent unavailability of L1 categories to assimilate all of hibitive number of trials that a RDT would require to consider the non-native vowels to. By contrast, C /o/ and /O/ were all of the possible combinations. Still, a practical approach is found to be most similar to E /`/ despite the availability of the one followed in this paper. where all native and non- additional L1 categories in the inventory, for instance, E /O+/. native vowels were equally represented in the RDT while As discussed above, the role of duration as a phonetic cue to only including a subset of all of the possible vowel combina- the identity of E /O+/ may have rendered C /o/ as a poor match tions. The selection of combinations can be based on the despite being spectrally closer (Cebrian, 2021). This shows results of exploratory investigations using a PAT approach that specific characteristics of the native language determine (as was performed in this study). the way perceptual similarity relations are established. In brief, despite the advantages and disadvantages of Finally, the results also show that a number of each of the methods, the combined results of this and our non-native vowels are perceived to be as similar to native J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian 2791 https://doi.org/10.1121/10.0014902 categories as two exemplars from the same L1 category, degree to which high levels of perceptual similarity between pointing to the possibility of near-identical or shared catego- native and non-native vowels may be an indication of near- ries between the two languages. identity or shared categories across the two languages. The results of this study on perceptual similarity have Further, the general reciprocity of the results of the two been compared to earlier results on ecphoric similarity (indi- experiments confirms the language-independent nature and rect covert method) involving the same stimuli and partici- the bidirectionality of perceived similarity. However, the pants (Cebrian, 2021). This constitutes the first direct characteristics of each language, such as the degree of comparison of ecphoric and perceptual similarity between a within-category variability or weighting of specific acoustic large number of native and non-native stimuli. The compari- cues (e.g., vowel duration) also influence crosslinguistic son indicates a very close relationship between the two types perception of similarity and may account for differences of measures: for both experiments, the RDT results (DRs) between the two populations in the perceptual similarity of were highly correlated with the PAT measures (percent the same pairs of sounds. Future studies will need to evalu- assimilation, GRs) obtained by Cebrian (2021). This sug- ate the predictions that can be derived from these measures gests that individuals’ judgements of perceptual similarity of perceptual and ecphoric similarity for L2 speech learn- and ecphoric similarity were highly comparable, at least ing, e.g., in terms of the ease of discrimination of regarding the similarity between native and non-native vow- non-native/L2 pairs or production of L2 sounds found to be els. The use of the two tasks may help to obtain a more reli- perceptually very close or near-identical to native sounds able measure of crosslinguistic similarity. For instance, (see Cebrian, 2021, for some predictions for Catalan learn- Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf researchers can design a RDT that includes all of the theo- ers of English and English learners of Catalan). Still, given retically relevant native–non-native combinations based on the relevance of the notion of similarity for a number of the modal responses previously obtained in a PAT, thus, disciplines and L2 speech research in particular, this paper avoiding the need to include a prohibitive number of possi- hopes to have contributed to a better understanding of the ble native–non-native vowel combinations. Further, it is nature of (crosslinguistic) perceived similarity, underscor- possible that each type of task can serve a different purpose. ing the importance of obtaining a reliable, yet practical RDTs may be more suitable in cases where the use of cate- method of assessment. gory labels should be discouraged to avoid problems of inconsistency and interpretability of responses (Flege and ACKNOWLEDGMENTS Bohn, 2021; Strange, 2007), e.g., if listeners’ literacy or This work was supported by Research Grant Nos. level of L2 proficiency is a concern. On a similar note, FFI2017-88016-P and PID2021-122396NB-I00 from the Cebrian et al. (2011) argued that PATs may be a good Spanish Ministries of Economy and Competitiveness and method to evaluate the perceived similarity between native Science and Innovation and Research Group Grant No. and non-native sounds at a given point in time (e.g., naive 2017SGR34 from the Catalan Government. vs inexperienced L2 learners) while RDT can be suitable for exploring changes in perceived similarity over time. This 1 Although it seems incoherent to refer to a task that measures “ecphoric” issue is left for future research. similarity as a PAT—it should probably be renamed “ecphoric assimila- This study is not without limitations. First, the RDTs in tion task”—the term PAT will continue to be used here for the sake of comparison with previous studies that use this term. Thus, “perceptual” in this study did not include all of the possible combinations of PAT has a general meaning, i.e., “perceived.” native and non-native vowels that would ideally be included 2 The Catalan vowel inventory includes a reduced vowel ([@]). This vowel to assess the full range of similarity relationships between was not included as in most varieties, including Eastern Catalan, it only native and non-native vowels (Strange, 2007). Still, the appears in unstressed position and is subject to large dialectal variability (Recasens, 1993). selection of vowel pairs was guided by the results of previ- 3 The English vowel /U/ was excluded as previous studies (Cebrian, 2009; ous research using a PAT, allowing the inclusion of all of Rallo Fabra, 2005) have shown that this vowel is perceived to be most dis- the crucial pairs. In addition, the phonetic context in which similar and not consistently assimilated to any single Catalan vowel, and often poorly identified by native English speakers [77% accuracy by vowels were presented was limited to /bVt/. A more com- native Canadian English speakers in Cebrian (2009) as opposed to plete assessment of crosslinguistic perceptual similarity 88%–100% for the rest of the vowels]. 4 would need to include a greater number of contexts, given Median values were also calculated as some studies report median rather that consonantal context may affect vowel perception (e.g., than mean values for rating data (e.g., Strange et al., 2001). The two mea- sures did not differ notably and thus mean values are reported following Strange et al., 2001). Finally, this study has focused on a other studies (e.g., Faris et al., 2016). specific variety of each language (Eastern Catalan and 5 The DR scale used in Cebrian et al. (2011) was a seven-point scale. The SSBE). The results may not be applicable if other varieties actual scores were converted to a nine-point scale to facilitate the compar- of English and Catalan are considered, something which ison of the two studies. 6 Following the conventions for each language, the phonetic symbol /O/ in would be worth pursuing. Catalan represents the lowest of the two mid back rounded vowels, To conclude, the use of a direct method of assessing whereas the symbol /O+/ in SSBE designates a vowel that is higher than perceptual similarity has been found to be a suitable 7 /`/. approach to measuring crosslinguistic similarity. The inclu- The chance levels of 16.7% (Table IV) and 12.5% (Table V) result from dividing 100% by the number of response options present per PAT (e.g., sion of native-native pairs in the RDT, particularly same- Faris et al., 2016), that is, six for the Catalan listeners and eight for the category native-native pairs, makes it possible to assess the English listeners [see Cebrian (2021) for details]. 2792 J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian https://doi.org/10.1121/10.0014902 Best, C. (1995). “A direct realist view of cross-language speech Empirical Progress, edited by R. Wayland (Cambridge University Press, perception,” in Speech Perception and Linguistic Experience: Theoretical Cambridge, UK), pp. 3–83. and Methodological Issues, edited by W. Strange (York, Timonium, Flege, J. E., Munro, M. J., and Fox, R. A. (1994). “Auditory and categorical MD), pp. 171–204. effects on cross language vowel perception,” J. Acoust. Soc. Am. 95, Best, C. T., and Tyler, M. D. (2007). “Nonnative and second-language 3623–3641. speech perception: Commonalities and complementarities,” in Language Fox, R. A., Flege, J. E., and Munro, M. J. (1995). “The perception of Experience in Second Language Speech Learning: In Honor of James English and Spanish vowels by native English and Spanish listeners: A Flege, edited by M. Munro and O.-S. Bohn (Benjamins, Amsterdam), pp. multidimensional scaling analysis,” J. Acoust. Soc. Am. 97(4), 13–34. 2540–2551. Boersma, P., and Weenink, D. (2018). “Praat: Doing phonetics by computer Holliday, J. J. (2015). “A longitudinal study of the second language acquisi- (version 6.0.43) [computer program],” available at http://www.praat.org/ tion of a three-way stop contrast,” J. Phonetics 50, 1–14. (Last viewed September 8, 2018). IBM Corp. (2021). IBM SPSS Statistics for MacIntosh (IBM Corp., Bohn, O.-S. (2002). “On phonetic similarity,” in An Integrated View of Armonk, NY), version 28. Language Development: Papers in Honor of Henning Wode, edited by P. Kuhl, P. K., Conboy, B. T., Coffey-Corina, S., Padden, D., Rivera-Gaxiola, Burmeister, T. Piske, and A. Rohde (Wissenschaftlicher, Trier, M., and Nelson, T. (2008). “Phonetic learning as a pathway to language: Germany), pp. 191–216. New data and native language magnet theory expanded (NLM-e),” Bohn, O.-S. (2017). “Cross-language and second language speech Philos. Trans. R Soc. B 363, 979–1000. perception,” in The Handbook of Psycholinguistics, edited by E. M. Ladefoged, P., and Ferrari Disner, S. (2012). Vowels and Consonants, 3rd Fernandez and H. Smith Cairns (Wiley-Blackwell, Chichester, UK), pp. ed. (Wiley-Blackwell, Chichester, UK). 213–239. Ladefoged, P., and Johnson, K. (2014). A Course in Phonetics, 6th ed. Bohn, O. S., and Garibaldi, C. L. (2021). “Examining the success of three (Wadsworth, Boston, MA). kinds of cross-language similarity in predicting English listeners’ discrim- Pisoni, D. (1973). “Auditory and phonetic memory codes in the discrimina- Downloaded from http://pubs.aip.org/asa/jasa/article-pdf/152/5/2781/16531940/2781_1_online.pdf ination of Danish vowels,” J. Acoust. Soc. Am. 150(4), A44. tion of consonants and vowels,” Percept. Psychophys. 13, 253–260. Cebrian, J. (2009). “Effects of native language and amount of experience Rallo Fabra, L. (2005). “Predicting ease of acquisition of L2 speech sounds: on crosslinguistic perception,” J. Acoust. Soc. Am. 125(4), 2775. A perceived dissimilarity test,” Vigo Int. J. Appl. Linguist. 2, 75–92, Cebrian, J. (2011). “Cross-linguistic perception of Catalan and English available at https://revistas.uvigo.es/index.php/vial/article/view/10/10. diphthongs,” in Achievements and Perspectives in the Acquisition of Rallo Fabra, L., and Romero, J. (2012). “Native Catalan learners’ percep- Second Language Speech: New Sounds 2010, edited by M. Wrembel, M. tion and production of English vowels,” J. Phonetics 40(3), 491–508. Recasens, D. (1993). Fonètica i Fonologia (Phonetics and Phonology) Kul, and K. Dziubalska-Kołaczyk (Lang, Frankfurt am Main, Germany), (Enciclopèdia Catalana, Barcelona, Spain). Vol. II, pp. 43–51. Schmidt, A. M. (1996). “Cross-language identification of consonants. Part Cebrian, J. (2019). “Perceptual assimilation of British English vowels to 1. Korean perception of English,” J. Acoust. Soc. Am. 99(5), 3201–3211. Spanish monophthongs and diphthongs,” J. Acoust. Soc. Am. 145(1), Schmidt, A. M. (2007). “Cross-language consonant identification,” in EL52–EL58. Language Experience in Second Language Speech Learning: In Honor of Cebrian, J. (2021). “Perception of English and Catalan vowels by English James Flege, edited by M. Munro and O.-S. Bohn (Benjamins, and Catalan listeners: A study of reciprocal cross-linguistic similarity,” Amsterdam, the Netherlands), pp. 185–200. J. Acoust. Soc. Am. 149(4), 2671–2685. Strange, W. (1987). “Information for vowels in formant transitions,” Cebrian, J., Mora, J. C., and Aliaga Garcıa, C. (2011). “Assessing crosslin- J. Mem. Lang. 26, 550–557. guistic similarity by means of rated discrimination and perceptual assimi- Strange, W. (2007). “Cross-language similarity of vowels,” in Language lation tasks,” in Achievements and Perspectives in the Acquisition of Experience in Second Language Speech Learning: In Honor of James Second Language Speech: New Sounds 2010, edited by M. Wrembel, M. Flege, edited by M. Munro and O.-S. Bohn (Benjamins, Amsterdam, the Kul, and K. Dziubalska-Kołaczyk (Lang, Frankfurt am Main, Germany), Netherlands), pp. 35–55. Vol. I, pp. 41–52. Strange, W., Akahane-Yamada, R., Kubo, R., Trent, S. A., and Nishi, K. Chang, C. B. (2019). “The phonetics of second language learning and (2001). “Effects of consonantal context on perceptual assimilation of bilingualism,” in The Routledge Handbook of Phonetics, edited by W. F. American English vowels by Japanese listeners,” J. Acoust. Soc. Am. Katz and P. F. Assmann (Routledge, Abingdon, UK), pp. 427–447. 109, 1691–1704. Cortina, J. M. (1993). “What is coefficient alpha? An examination of theory Strange, W., Akahane-Yamada, R., Kubo, R., Trent, S., Nishi, K., and and applications,” J. Appl. Psychol. 78, 98–104. Jenkins, J. (1998). “Perceptual assimilation of American English vowels Cruttenden, A. (2014). Gimson’s Pronunciation of English, 8th ed. by Japanese listeners,” J. Phonetics 26, 311–344. (Routledge, New York). Strange, W., Bohn, O.-S., Trent, S. A., and Nishi, K. (2004). “Acoustic and Docherty, G. J. (1992). The Timing of Voicing in British English Obstruents perceptual similarity of North German and American English vowels,” (de Gruyter, Berlin, Germany). J. Acoust. Soc. Am. 115, 1791–1807. Elvin, J., and Escudero, P. (2019). “Cross-linguistic influence in second lan- Strange, W., Hisagi, M., Akahane-Yamada, R., and Kubo, R. (2011). guage speech: Implications for learning and teaching,” in Cross- “Cross-language perceptual similarity predicts categorical discrimination Linguistic Influence: From Empirical Evidence to Classroom Practice, of American vowels by na€ıve Japanese listeners,” J. Acoust. Soc. Am. edited by M. J. Gutierrez-Mangado, M. Martinez-Adrian, and F. 130(4), EL226–EL231. Gallardo-del-Puerto (Springer, Cham, Switzerland), pp. 1–20. Thomson, R. I., Neary, M. N., and Derwing, T. M. (2009). “A modified sta- Escudero, P., and Williams, D. (2011). “Perceptual assimilation of Dutch tistical pattern recognition approach to measuring the crosslinguistic simi- vowels by Peruvian Spanish listeners,” J. Acoust. Soc. Am. 129(1), larity of Mandarin and English vowels,” J. Acoust. Soc. Am. 126, EL1–EL7. 1447–1460. Faris, M., Best, C. T., and Tyler, M. D. (2016). “An examination of the dif- Tulving, E. (1981). “Similarity relations in recognition,” J. Verbal Learn. ferent ways that non-native phones may be perceptually assimilated as Verbal Behav. 20, 479–496. uncategorized,” J. Acoust. Soc. Am. 139, EL1–EL5. Tyler, M. D. (2021). “Phonetic and phonological influences on the discrimi- Flege, J. E. (1995). “Second language speech learning: Theory, findings and nation of non-native phones,” in Second Language Speech Learning: problems,” in Speech Perception and Linguistic Experience: Theoretical Theoretical and Empirical Progress, edited by R. Wayland (Cambridge and Methodological Issues, edited by W. Strange (York, Timonium, University Press, Cambridge, UK), pp. 157–174. MD), pp. 233–277. Tyler, M. D., Best, C., Faber, A., and Levitt, A. (2014). “Perceptual assimi- Flege, J. E., and Bohn, O.-S. (2021). “The revised speech learning model lation and discrimination of non-native vowel contrasts,” Phonetica 71, (SLM-r),” in Second Language Speech Learning: Theoretical and 4–21. J. Acoust. Soc. Am. 152 (5), November 2022 Juli Cebrian 2793