Cross-language perception of the Japanese singleton/geminate contrasts: comparison of Vietnamese speakers with and without Japanese language experience

1.1 Background

Japanese is a quantity language that uses durational variation contrastively for both vowels and consonants (e.g., Fujisaki et al. 1973; Vance 2008). For example, koko ‘individual, separate’ contrasts with kooko ‘finance corporation’ on the one hand and with kokko ‘national treasury’ on the other hand. While this is essential for efficient communication, it is difficult to learn to perceive and produce the durational contrast especially for non-native speakers from diverse linguistic backgrounds including Vietnamese, which is the target group of this study (e.g., Đỗ 2012, 2015; Sugimoto 2003, 2005, 2007 for perception; Kanamura 1999; Matsuda et al. 2018; Yamakawa et al. 2022 for production). Many studies on durational contrasts in second language (L2) Japanese have been conducted (e.g., Han 1992; Hirata 2004, 2015; Hirata et al. 2007; Tsukada et al. 2018; Tsukada and Hajek 2023) confirming clear and consistent differences between native and non-native speakers. However, valuable findings from some of the previous studies focusing on Vietnamese learners of Japanese are available mostly in Japanese and have not yet reached a wider audience in the field.

Recently, the number of Vietnamese learners of Japanese has been rapidly increasing (e.g., Hashimoto 2022). As of 2021, Vietnam is ranked within the top 6 countries/regions (169,582 or 4.5 % of the entire 3,794,714 learners) of the world in terms of the number of learners of Japanese (The Japan Foundation 2021). Within Japan, Vietnam (31,643 or 14.4 % of the entire 219,808 learners) is the second largest country of origin of non-native learners of Japanese after China (67,027 or 30.5 %) as of 2022 (Agency for Cultural Affairs, Government of Japan 2022). As such, improving our current understanding of how to facilitate communication and pronunciation pedagogy for Vietnamese learners of Japanese is crucial. Given that limited studies have focused on this emergent group of Vietnamese speakers, our aims are (1) to add to the existing literature investigating the role of prior linguistic experience by comparing three groups of native Vietnamese speakers differing in Japanese experience and a control group of native Japanese speakers, and (2) to determine how they may differ in the perception of Japanese consonant length contrasts (i.e., short/singleton vs long/geminate) on account of their experience with Japanese. Below we first provide a brief description of phonetic and phonological characteristics of consonant length contrasts in Japanese. We also briefly describe characteristics of the Vietnamese sound system. Then we review the relevant literature on L2 processing of consonant length.

1.2 Length contrasts in Japanese

As mentioned above, Japanese uses durational variation contrastively for both vowels and consonants. The difference in length (e.g., kite ‘wearing’ vs k ii te ‘listening’ vs ki tt e ‘stamp’) affects the lexical meaning as does the difference in the vowel type (e.g., k i te ‘wearing’ vs k a te ‘nourishment’) or consonant type (e.g., ki t e ‘wearing’ vs ki r e ‘cloth’). While segmental duration is consistently reported to be the primary (though not the only) acoustic cue to differentiate the short and long members of the length contrast in Japanese (e.g., Fujisaki et al. 1973; Kawahara 2015), non-durational cues including fundamental frequency (F0), intensity, voice quality among others have also been examined in various studies (e.g., Idemaru and Guion 2008; Kubozono et al. 2011). Due to the frequent occurrence (e.g., Tamaoka and Makioka 2004) and critical role in determining the meaning, from the very beginning, learners of Japanese are typically exposed to length contrasts both auditorily via instructors’ pronunciation and visually via the explicitly marked orthography (i.e., kana syllabaries). While the difference in vowel/consonant categories is primarily realized at the segmental level via the spectral cues, the difference in length categories is primarily realized at the suprasegmental level via the temporal cues.

Of relevance to the present study, numerous researchers (e.g., Kawahara 2015; Vance 2008) examined phonetic and phonological characteristics of Japanese singletons and geminates. In a mora-timed language such as Japanese, word duration covaries with the number of moras (e.g., Homma 1981). While a geminate sound counts as one mora, it does not form a syllable. It has been suggested that length contrasts may differ phonetically in mora-timed (e.g., Japanese) and syllable-timed (e.g., Italian) languages (Ham 2001; Idemaru and Guion 2008). Specifically, while vowels preceding geminates are known to be shorter (by up to 37 %) than vowels preceding singletons in quantity languages such as Italian (e.g., Esposito and Di Benedetto 1999; Hajek et al. 2007) among others, the exact opposite has been reported for Japanese. In other words, in Japanese, vowels tend to be phonetically longer before geminates than before singletons (e.g., Han 1992; Hussain and Shinohara 2019; Idemaru and Guion 2008; Kawahara 2015).

1.3 Characteristics of the Vietnamese sound system

The participants’ first language (L1), Vietnamese, is a tonal language with five (for the Southern dialect) or six (for the Northern dialect) lexical tones (e.g., Brunelle 2009; Đỗ 2015; Hwa-Froelich et al. 2002; Nguyễn 2009; Phạm and McLeod 2016). The robustness of voice quality in signaling lexical tones has been reported for some dialect speakers in previous research (e.g., Brunelle 2009; Brunelle and Kirby 2016; Hwa-Froelich et al. 2002; Michaud 2004; Phạm and McLeod 2019). Vietnamese is generally classified as a monosyllabic (e.g., Do-Hurinville and Dao 2015; Hwa-Froelich et al. 2002; Thompson 1988; Verdenschot et al. 2022), syllable-timed language (e.g., Kanamura 1999; Nguyễn et al. 2008; Phạm and McLeod 2019; Sugimoto 2003, 2005, 2007; Yamakawa et al. 2022; Yin and Wang 2021) and differs from Japanese, which is classified as a mora-timed language with lexical pitch accents (e.g., Goss and Tamaoka 2019; Vance 2008).

Vowel length contrasts are reported to be limited (e.g., Kirby 2011; Pajak and Levy 2014) or non-existent (Đỗ 2012, 2015) in Vietnamese. Vietnamese learners of Japanese have been shown to have difficulty perceiving and/or producing Japanese vowel length in various studies (e.g., Đỗ 2012, 2015; Sugimoto 2005; Yin and Yasuhara 2020). A recent study (Le et al. in press) focusing on the perception of Japanese vowel length reported that only advanced (but not lower proficiency) Vietnamese learners modestly resembled native Japanese speakers in their use of duration and F0 cues. Crucially, however, unlike Japanese, consonant length is not contrastive in Vietnamese either at the morpheme level or the word level (e.g., Đỗ 2015; Do-Hurinville and Dao 2015; Kanamura 1999; Pajak and Levy 2014; Phạm and McLeod 2016). Not surprisingly, perceiving and producing singleton/geminate contrasts in the context of voiceless consonants has been reported to be problematic for Vietnamese learners of Japanese (Sugimoto 2007).

We present data on the discrimination of Japanese consonant length (i.e., short/singleton vs long/geminate) by three groups of native Vietnamese speakers differing in Japanese experience as mentioned above. While previous research involving Vietnamese learners of Japanese in Hanoi (e.g., Sugimoto 2003, 2005, 2007) tends to be subjectively observational by the authors, the present study provides empirical data via speech perception experiments.

1.4 Review of literature

1.5 Acquisition of durational variations within the theoretical framework of L2 phonetics and phonology

Prominent models of L2 speech learning such as an L2 version of the Perceptual Assimilation Model (PAM-L2) (Best and Tyler 2007), the Speech Learning Model (SLM) (Flege 1995) and its successor, the Revised Speech Learning Model (SLM-r) (Flege and Bohn 2021) all focus on the learning of L2 segments (i.e., vowels and consonants). Unlike segmental contrasts signaled by vocalic or consonantal variations which are found in all languages, length contrasts signaled by durational variations are found only in some, but not all, languages. For native Japanese speakers, durational variations are contrastive (phonological) and function at both segmental and suprasegmental levels. For non-native learners without phonological representations for length in their L1, one or both members of the length contrasts may be classified as “new” sounds (Flege 1995; Flege and Bohn 2021). Their perception of durational variations is likely to be unstable, i.e., initially continuous/phonetic and gradually developing to become discrete/phonological as they become experienced in Japanese. Extending theories of L2 speech learning such as the SLM-r (Flege and Bohn 2021) from segmental to durational variations, Vietnamese speakers, for whom geminates are non-existent, may be expected to have no preexisting L1 phonetic categories which interfere with the formation of new phonetic categories for L2 geminates.

However, as pointed out by Best (2019) for the perception of pitch variations by speakers whose L1s do not employ lexical tones, it is important to recognize that non-tonal language speakers “are not lacking entirely in experience with phonological information being conveyed by pitch variations”. The same logic may apply to “false” (aka fake, post-lexical, concatenated) geminates (e.g., cat tail, Lee et al. 2023) which occur in many languages even though they may not always transfer positively to L2.

Further, according to the “full access hypothesis” adopted by the Revised Speech Learning Model (SLM-r) (Flege and Bohn 2021), L2 learners can gain access to the L2 sounds which are defined, at least in part, by features not used in the learner’s L1. Due perhaps to the fact that (1) all languages make use of durational variations for linguistic or non-linguistic purposes and that (2) length categories are usually fewer in number compared to segmental categories, durational cues may be accessible and learnable to speakers of any language. Thus, given that singleton versus geminate contrasts are primarily based on durational variations which all languages make use of at least to some extent, with extensive experience, learners may ultimately be able to process them successfully. The present study aims to contribute to the current L2 speech learning literature by further broadening target sounds and learner populations.

1.6 Present study

The main objective of the present study was to evaluate the effects of Japanese language experience on native Vietnamese speakers’ perception of Japanese short/singleton versus long/geminate contrasts. As pointed out by some researchers (e.g., Altmann et al. 2012; Hallé et al. 2016), previous L2 research on consonant length contrasts mostly focused on non-native learners of Japanese from English (e.g., Hardison and Motohashi Saigo 2010; Hayes-Harb 2005; Tsukada et al. 2018), Korean (e.g., Sonu et al. 2013; Tsukada et al. 2018) or, more recently, Cantonese (e.g., Lee and Mok 2018; Lee et al. 2023) backgrounds. We aim to increase our current understanding on contrastive length perception by focusing on the learner population from a different L1 background that has been substantially growing. Furthermore, due to recruitment difficulties, research involving highly advanced L2 learners in languages other than English is still limited. We hope to fill this gap by examining if and how advanced Vietnamese learners perceive Japanese singleton/geminate contrasts.

The questions addressed in this study are as follows. Firstly, is there a difference across four (three Vietnamese and one Japanese) groups of participants in their perception of Japanese consonant length? Specifically, do the two groups of learners, one in Vietnam (less experienced/proficient) and the other in Japan (more experienced/proficient), differ from each other on the one hand and from the native Japanese speakers on the other hand? We predict that the Vietnamese groups are less accurate than the native Japanese group, because Vietnamese does not use consonant length contrastively and Vietnamese speakers have less experience with consonant length contrast than native Japanese speakers. In addition, we predict that the advanced learners in Japan are more accurate than the intermediate learners in Vietnam, because the former has much higher proficiency of Japanese and much longer LOR in Japan than the latter. Our second research question was to examine if the place of articulation effect which we observed with speakers from diverse quantity and non-quantity L1 backgrounds (e.g., Hamzah et al. 2023; Tsukada and Hajek 2022; Tsukada and Yurong 2023) also applies to the Vietnamese speakers. Specifically, in many cases, length discrimination was more accurate for the alveolars than for the velars. If this finding is replicated with our Vietnamese participants, it may provide further evidence that this place effect is a language-general phenomenon.

2.1 Speakers and speech materials

The experimental stimuli and procedures were identical to those used in our previous research (e.g., Hamzah et al. 2023; Tsukada and Hajek 2022; Tsukada and Yurong 2023). Six (3 males, 3 females) native Japanese speakers participated in the recording sessions, which lasted between 45 and 60 min. The speakers’ age ranged from late twenties to early forties. According to self-report, which was confirmed by the first author who is a native Japanese speaker originally from Tokyo, all speakers spoke standard Japanese, having been born or having spent most of their life in the Kanto region surrounding the Greater Tokyo Area. The speakers were recorded in the recording studio at the National Institute of Japanese Language and Linguistics, Tokyo.

Table 1 shows 12 Japanese real word pairs used in this study. Each word pair had the same pitch accent patterns. However, some word pairs had a different accent pattern from the other pairs (i.e., High-Low (HL) for ka t o/ka tt o, ma t e/ma tt e, sa t e/sa tt e, i k a/i kk a, and ka k o/ka kk o and LH for the rest of the word pairs). Only words with intervocalic stops were considered in this study. As voiced geminates are disfavoured and their occurrence is limited mostly to loanwords in Japanese (e.g., Hussain and Shinohara 2019; Kawahara 2015; Sano 2019), only voiceless stops (/t, k/) were used (see also Lee and Mok 2018). Of all the Japanese consonants, these two stops occur most frequently both as singletons and geminates (Tamaoka and Makioka 2004). Further, voiceless bilabials are limited to loanwords and onomatopoeic words in Japanese (e.g., Tsujimura 2013).

On average, the closure durations were 96 ms and 262 ms for singletons and geminates, respectively. The geminate-to-singleton ratios were 2.7 for alveolars (/t/-/tː/) and 2.8 for velars (/k/-/kː/), respectively. These durational values are generally in good agreement with what has been reported in previous research (e.g., Han 1992; Hayes-Harb 2005) (see, however, Sano 2019 for alveolars).

2.2 Participants

Three groups of native Vietnamese speakers participated in this study. The first group of 12 (6 males, 6 females, mean age = 21.0 years, sd = 2.3) consisted of university students in Ho Chi Minh City and had no Japanese experience (VTN). Seven of them were born in Ho Chi Minh City and the rest in other locations in the South region of the country (e.g., Ben Tre, Dong Nai, Kien Giang, Tay Ninh, Tien Giang). Although this diversity is undesirable, it was not practical to tightly control the participants’ dialectal variations in this study. In each of the dialects, the durational contrast for consonants is not contrastive. The next two groups, one living in Ho Chi Minh City, Vietnam (7 males, 10 females, mean age = 21.5 years, sd = 4.4) (VTI) and the other in Tokyo, Japan (5 males, 8 females, mean age = 28.6 years, sd = 5.1) (VTA) were learners of Japanese. In addition to the countries of residence, the two learner groups substantially differed in their Japanese proficiency based on the Japanese-Language Proficiency Test (JLPT), according to which the easiest level is N5 and the most difficult level is N1 (Nishizawa et al. 2022).

The 13 VTA learners had all passed the JLPT at N1 (roughly equivalent to CEFR (Common European Framework of Reference for Languages) C1) (The Japan Foundation/Japan Educational Exchanges and Services 2025) and were considered highly advanced learners of Japanese. They participated in the study at a university in Tokyo. Broadly speaking, their places of origin in Vietnam were 7 from North, 1 from Central and 5 from South. Their mean LOR in Japan was 8.4 years (sd = 3.4) at the time of participation. As for their self-reported use of Japanese in Japan, six of them reported more frequent use of Japanese, four reported more frequent use of Vietnamese and the rest reported using both languages with comparable frequency. Three learners reported using Japanese always. They started learning Japanese at the mean age of 17.4 years (sd = 4.1). The learners’ Japanese language learning backgrounds varied somewhat. However, most of them reported learning Japanese intensively at a language school in Japan. All reported having native Japanese-speaking teachers. As mentioned above, whether pronunciation instruction is emphasized or not is typically left up to each instructor’s discretion and, therefore, difficult to quantify accurately. Nevertheless, taken together, compared to the nominally advanced learners included in previous research (e.g., Hardison and Motohashi Saigo 2010; Lee and Mok 2018), it is expected that the VTA learners in the present study were more experienced as well as immersed in the Japanese-speaking environment.

The 17 VTI learners were students at a university in or around Ho Chi Minh City and were at the JLPT N3 level (roughly equivalent to CEFR B1) (The Japan Foundation/Japan Educational Exchanges and Services 2025), which requires “the ability to understand Japanese used in everyday situations to a certain degree”. Except for two participants, all were studying Japanese in the same program at the same university and were considered (pre-)intermediate learners. Nine of them were born in Ho Chi Minh City and the rest in the South/Central regions of the country (e.g., Ben Tre, Dong Nai, Gia Lai, Khanh Hoa, Kien Giang, Nha Trang, Quang Nam). None of them had lived overseas for an extended period of time (i.e., more than 6 months).

Unfortunately, it is difficult to objectively measure and/or control learners’ levels of proficiency in L2 speech research due to large individual variations and different curricula at different educational institutions in different countries. While we used the JLPT as an index of the Vietnamese participants’ proficiency in Japanese, this is not intended to imply that it is the optimal measure of non-native learners’ Japanese experience or capability.

The Vietnamese participants’ results were compared to that of a control group of 10 (2 males, 8 females) native Japanese (NJ) speakers (mean age = 21.0 years, sd = 0.8) who were students at a university in the US where the first author was based at that time. All NJ speakers were from geographical areas where no absence or neutralization of consonant length contrast has been reported. Their mean LOR in the US was 0.4 years (sd = 0.22) at the time of participation. None of the NJ speakers participated in the recording sessions. According to self-report, all four groups of participants had normal hearing.

All participants were tested individually in a session lasting approximately 30–40 min in a sound-attenuated laboratory or a quiet room at their own university. The experimental session was self-paced. The participants heard the stimuli at a self-selected, comfortable amplitude level over the high-quality headphones on a computer.

2.3 Procedure

The 12 pairs of words in Table 1 containing singleton (S) or geminate (G) consonant intervocalically (underlined and bolded) were arranged in triads in four patterns (S-S-G, S-G-G, G-S-S, and G-G-S, 48 trials each). Each pair was presented 16 times (4 times using different combinations of tokens for each of the four patterns), resulting in a total of 192 trials. This was preceded by eight practice trials, which were not analyzed.

The participants completed a two-alternative forced-choice AXB discrimination task, in which they were asked to listen to trials arranged in a triad (A-X-B). The presentation of the stimuli and the collection of perception data were controlled by the PRAAT program (Boersma and Weenink 2016). In the AXB task, the first (A) and third (B) tokens always came from different length categories, and the participants had to decide whether the second token (X) (i.e., “word” in a language which may be unfamiliar to them) belonged to the same category as A (e.g., kato ₂  – kato ₁  – katto ₃ ) or B (e.g., kako ₁  – kakko ₃  – kakko ₂ ; where the subscripts indicate different speakers).

The participants listened to a total of 200 trials. The first eight trials were for practice and were not analyzed. The three tokens in all trials were spoken by three different speakers, who appeared equiprobably in each within-AXB position. Thus, X was never physically identical to either A or B. This was to ensure that the participants focused on relevant phonetic characteristics that group two tokens as members of the same length category without being distracted by audible but phonemically (or phonologically) irrelevant within-category variation (e.g., in voice quality). This was considered a reasonable measure of participants’ perceptual capabilities in real world situations (Strange and Shafer 2008).

The participants were given two (‘A’, ‘B’) response choices on the computer screen. They were asked to select the option ‘A’ if they thought that the first two tokens in the AXB sequence were the same and to select the option ‘B’ if they thought that the last two tokens were the same. No feedback was provided during the experimental sessions. The participants could take a break after every 50 trials if they wished. The participants were required to respond to each trial, and they were told to guess if uncertain. A trial could be replayed as many times as the participants wished to reduce their anxiety, but responses could not be changed once given. The interstimulus interval in all trials was 0.5 s.

3.1 Overall

The overall mean d′ scores were 1.0, 1.9, 3.1 and 4.5 for the VTN group, two groups of learners (VTI, VTA) and the control NJ group, respectively. Figure 1 shows overall d′ scores for the four groups of participants. The NJ group was at ceiling with little individual variation. The two groups of learners clearly outperformed the non-learners in perceiving Japanese consonant length contrasts. Regardless of Japanese experience, all Vietnamese groups were apparently much more variable than the NJ group. It is noteworthy that two non-learners achieved the d′ scores of 2.1 and 2.6, respectively. These scores exceed the mean score of the VTI group and are well within the range (1.52–4.65) set by the VTA group. While none of the 12 non-learners reached the range of d′ scores (3.77–4.65) set by the NJ group, 5 Vietnamese learners (4 in VTA, 1 in VTI) reached the NJ range.

Figure 1:

Overall discrimination index (d′) by four groups of participants. The horizontal line and the red circle in each box indicate the median and mean, respectively. The bottom and top of the box indicate the first and third quartiles. The horizontal position of each empty dot represents randomly added variation to reduce overplotting.

Levene’s test for homogeneity of variance was significant [F(3, 48) = 4.8, p < 0.05] when all four groups were included, but non-significant [F(2, 39) = 0.86, p = 0.43] when the NJ group was excluded, suggesting that homogeneity of variance can only be assumed for the three Vietnamese groups. This qualitative as well as quantitative difference is clearly seen in Figure 1. Based on the results of Levene’s test, between-group comparisons were conducted non-parametrically including the NJ group and parametrically via repeated-measures analysis of variance (ANOVA) excluding the NJ group, respectively.

First, a non-parametric Kruskal-Wallis test with Group (VTN, VTI, VTA, NJ) reached significance [χ²(3) = 34.4, p < 0.001, η² = 0.65]. Dunn’s test for multiple comparisons (Bonferroni adjusted) showed that the NJ group outperformed the VTN and VTI groups [p < 0.001], but did not significantly differ from the VTA group.

Next, we ran a one-way repeated-measures ANOVA excluding the NJ group to determine if the three Vietnamese groups differed in their Japanese consonant length discrimination. The results reached statistical significance [F(2, 39) = 15.6, p < 0.001, η G 2  = 0.44]. According to the post-hoc t-tests, all between-group differences were statistically significant [t(22.3–26.6) = −5.8 – −2.8, p < 0.05].

3.2 Comparison of length contrast discrimination at alveolar (/t/-/tː/) and velar (/k/-/kː/) places of articulation

Next, we examined if the overall pattern of results remained the same for different places of articulation. Table 2 and Figure 2 show d′ scores for the four groups of participants when the place of articulation (Alveolar, Velar) of the target token (i.e., X in the AXB sequence) was taken into consideration. On average, all four groups performed better when the alveolar rather than velar stops occurred in the target position (VTN: 1.0 vs 0.9, VTI: 2.1 vs 1.8, VTA: 3.5 vs 2.9, NJ: 4.65 vs 4.3 for Alveolar and Velar, respectively). The NJ group never misperceived the length categories when the target stop was alveolar. Thus, this group was excluded in the following analysis.

Figure 2:

Discrimination index (d′) by four groups of participants for trials differing in the place of articulation (Velar, Alveolar) of the target token. The horizontal line and the red circle in each box indicate the median and mean, respectively. The bottom and top of the box indicate the first and third quartiles. The light lines connect individual participants’ scores.

Two-way repeated-measures ANOVA with Group (VTN, VTI, VTA) and Place (Alveolar, Velar) reached significance for the main effects of Group [F(2, 39) = 15.7, p < 0.001, η G 2  = 0.42] and Place [F(1, 39) = 10.2, p < 0.01, η G 2  = 0.03] only. The overall pattern of between-group differences seen in Figure 1 was retained in Figure 2. According to the post-hoc t-tests, all between-group differences were significant except for the comparison between the VTN and VTI groups for the target velar. Although all groups performed slightly better for the alveolar than for the velar contrasts on average, the difference in the place of articulation did not reach statistical significance perhaps due to large variability especially for the VTA and VTN groups.

4.1 Between-group difference: advanced learners’ performance

Whilst not completely target-like, the VTA learners’ performance was closer to that of the NJ speakers than to that of the VTI or VTN speakers. Undoubtedly, the VTA learners in this study benefitted from L2 Japanese learning unlike the learners in some previous research (e.g., Đỗ 2012, 2015; Lee and Mok 2018).

We speculate that the VTA learners’ superior performance may be due to various reasons. One possibility is that not only the cross-language speech perception skills, but high frequencies of occurrence of the target sounds (i.e., /k/ and /t/) in the Japanese lexicon (Tamaoka and Makioka 2004) and/or orthography (i.e., kana syllabaries) that explicitly marks length contrasts, i.e., auditory and/or visual reinforcement in classrooms, may have facilitated the VTA learners to a greater extent than the VTI or VTN speakers (see Alarifi and Tucker 2024 for the role of orthography in non-native vowel length learning). This possibility needs to be verified by examining how participants from quantity L1 backgrounds who are familiar with consonant length in their L1 (e.g., Arabic, Finnish, Italian), but unfamiliar with Japanese might perform in the same task.

The VTA learners’ success may also be related to how Japanese geminates are related to the Vietnamese sound system. Presumably, geminate sounds are non-existent and “new” to native Vietnamese speakers. If so, they should be free from “equivalence classification” (e.g., Flege 1987). Thanks, perhaps, to the absence of preexisting L1 phonetic categories which might interfere with the formation of new phonetic categories for L2 geminates as proposed in the SLM-r (Flege and Bohn 2021), the VTA learners with sufficient exposure to Japanese may have been able to learn to perceive Japanese consonant length successfully.^[3]

Alternatively, the VTA learners’ advantage may be more to do with a combination of their extensive experience in Japanese and an extended stay in Japan, i.e., quality and/or quantity of L2 Japanese input. Indeed, the VTA learners in this study had a much longer LOR of 8.4 years in Japan compared to the advanced learners (LOR of 1 year) in Lee and Mok (2018). However, factors such as L2 experience and LOR may be ambiguous and unreliable as a predictor of L2 pronunciation. Thus, it would be necessary to search for measures other than LOR that can accurately characterize learners’ capacity. For example, in the SLM-r, Flege and Bohn (2021) used full-time equivalent (FTE) years of L2 input by multiplying years of residence in the target country by self-reported proportion of target language use as possibly “a somewhat better estimate of quantity of L2 input than LOR alone (p. 7, also p. 32)”. However, as they point out themselves, “it says nothing regarding the quality of L2 input (p. 7)”. Given that “the quality of L2 input may matter more than the age of first exposure to an L2 (Flege and Bohn 2021, p. 8)”, securing the quality as well as quantity of L2 input may remain a pedagogical challenge especially for those learning Japanese outside Japan.

4.2 Influence of the place of articulation on consonant length discrimination

Regardless of Japanese experience, all groups discriminated consonant length slightly better for the alveolar than for the velar contrasts on average. We frequently observed the same pattern of results with listeners from diverse quantity and non-quantity L1 backgrounds in our previous studies (Hamzah et al. 2023; Tsukada and Hajek 2022; Tsukada and Yurong 2023). However, in the present study, the difference in the place of articulation did not reach statistical significance perhaps due to large within-group variability especially for the VTA and VTI groups (Figure 2). This differs from previous research (Sugimoto 2007), in which Vietnamese learners in Hanoi at the beginner/intermediate level were reported to have misperceived the alveolar geminates more frequently than the velar geminates. The results obtained may be related to the Vietnamese phonology. Velars in coda position are reported to be phonetically and orthographically more variable than alveolars in L1 Vietnamese (Phạm and McLeod 2016). If so, it may negatively affect how velars are cognitively represented in L2 Japanese, especially for the VTA (and VTI) learners who are expected to have actively competing L1 versus L2 systems.

Following one of the reviewers’ suggestions, we revisited the data to see if some word pairs were discriminated more or less accurately. Although we did not find a consistent pattern according to the word pair, for the VTA and VTI groups, the most poorly discriminated pair was kako ‘past’ versus kakko ‘parenthesis’ with kakko in the target (X) position. It is possible that the learners have been exposed to and/or familiar with these words. However, as we did not collect information about their lexical knowledge or mastery of writing systems, it is unclear if and how these words are internalized by the learners. Given that learners’ L2 vocabulary size positively affects their L2 speech processing (Bundgaard-Nielsen et al. 2011), gaining information regarding learners’ L2 Japanese vocabulary size as well as a tighter control over lexical frequency would be useful in future work.

Based on the finding that consonant length was generally more discriminable (albeit non-significantly) when the target token was the alveolar rather than the velar, potential pedagogical interventions may be to conduct targeted listening training with alveolar singletons/geminates first and delay the introduction of velar singletons/geminates until learners gain sufficient experience with consonant length. As alveolar contrasts such as -te versus -tte or -ta versus -tta occur very frequently in verb conjugations, in particular, gaining a solid grasp of this is useful in grammar as well as pronunciation learning. However, as mentioned above, we need to bear in mind that the opposite pattern (i.e., more misperception for the alveolar than for the velar geminates) has been reported previously (Sugimoto 2007) and that more data are needed to determine generalizability of our results.

4.3 Individual differences

Regarding individual differences, we observed that two of the VTN speakers achieved the d′ scores which were above the mean score of the VTI group and were well within the range (1.52–4.65) set by the VTA group. In addition, the majority (9 out of 12 [75 %]) of the VTN speakers performed just as well as the VTI participants who had some experience learning Japanese. Based on the background information provided, one of the two high-performing VTN speakers was majoring in cultural studies and the other in German linguistics and reported having studied music in junior high school. The former started learning English at the age of 12 and the latter at 8. Neither reported extended stays in foreign countries. While it is possible that these individuals had a keen interest in language and/or culture studies and possessed an exceptional aptitude for processing unfamiliar speech sounds or some other endogenous factors (e.g., superior auditory acuity, early-stage (precategorical) auditory processing, and working auditory memory) proposed in the SLM-r (Flege and Bohn 2021, p. 66), unfortunately, we do not have further details that account for their excellent performance.

Individual learners’ profiles characterized by factors such as prior language learning experience or cognitive abilities including their “L1 category precision” (Flege and Bohn 2021, p. 36, also on p. 65) or “L1 category compactness” (Kogan and Mora 2022) may play an important role in information processing in general and L2 speech processing in particular. For example, regardless of L2 experience or places of residence, the Vietnamese speakers with relatively precise L1 categories may have discriminated Japanese consonant length contrasts more accurately than those with relatively imprecise L1 categories. It is also possible that, due to the partial use (i.e., informativeness) of L1 vowel length as briefly presented in Section 1.3, the Vietnamese speakers with a good control of L1 category precision may be capable of transferring or repurposing their existing (albeit limited) L1 experience with vowel length in the context of processing consonant length in unfamiliar languages (e.g., Pajak and Levy 2014; Tsukada et al. 2021; Tsukada et al. 2024; Tsukada and Yurong 2022).

Finally, globally and especially in many Asian countries, Japanese popular culture such as manga and anime is hugely popular among young people. Anecdotally, some Japanese instructors noted a non-negligible influence of the popular culture on individuals’ L2 Japanese speech processing skills. Strong desire to impersonate characters in manga/anime may drive learners to seek massive auditory as well as visual input and to practice repetitively, which is likely to improve listening/speaking skills. Consumption of popular culture is another area where one individual is likely to vary widely from another.