The role of L1 phonology in the perception of L2 semivowels

2.1 The role of L1 on L2 phonology acquisition

When adult learners acquire L2 sounds, their L1 phonological system can exert great influence, contributing to varying levels of ease or difficulty in acquisition. In L2 phonology models such as the PAM (Best 1995; Best and Tyler 2007) and the SLM (Flege 1995, 2005; Flege and Bohn 2021; Flege et al. 2021), the acquisition of given L2 sounds essentially depends on how these sounds are mapped onto the learners’ L1 phonetic categories. In this regard, one straightforward criterion for predicting the relative ease or difficulty of L2 sounds is whether these sounds have corresponding equivalents in the learners’ L1. When learners acquire two sounds in their L2 that constitute a phonemic contrast, but the L1 equivalents of these sounds do not form such a contrast in their L1, they may face challenges in acquiring these L2 sounds. For example, in English, /s/ and /š/ are distinctive phonemes found in minimal pairs such as sip versus ship. However, /s/ and /š/ are allophones of the same phoneme /s/ in Korean, and when L1 Korean speakers learn English, the /s/-/š/ contrast in English can present a challenge (Eckman and Iverson 2013; Eckman et al. 2009; Fox et al. 2009; Idsardi and Son 2004). The empirical evidence from this line of research aligns well with the fundamental tenets of models such as the PAM and the SLM.

In addition to the L1 transfer assumed at the segmental level, some L2 phonology theories have emphasized the role of features. In the framework of Chomsky and Halle (1968), speech sounds can be grouped according to their distinctive features, including cavity, manner of articulation, source, and prosodic features. Feature-based L2 phonology models largely propose that features (associated with L2 sounds) that are also exploited in the learners’ L1 will be acquired more successfully compared to those that are not exploited in the L1 (e.g., Archibald 2005; Brown 1998, 2000). Brown (2000) conducted a series of experiments investigating the perception of English consonant contrasts among L1 Mandarin, Japanese, and Korean speakers of L2 English. The results showed that the L2 learners were more successful in distinguishing the consonant contrasts characterized by features present in their L1 phonological grammar, as opposed to those characterized by features absent in their L1. Research on L2 features has also dealt with suprasegmental features such as quantity. McAllister et al. (2002) examined the acquisition of Swedish vowel quantity by L2 learners of different L1 backgrounds. In Swedish, the same vowel category exhibits a long versus short length contrast, contributing to word meaning differences. Among the L1s spoken by the three groups of L2 learners, Estonian has a similar vowel quantity system while American English and Latin American Spanish do not. However, American English vowels exhibit a tense versus lax distinction (e.g., beet vs. bit), where tense vowels are typically longer in duration. Through production and perception tasks, the study found that the L1 Estonian speakers showed performance similar to the Swedish controls. While both the L1 English and Spanish speakers exhibited differences from the Swedish controls, the English speakers more closely resembled the native controls than did their Spanish counterparts. These results suggest that the presence (or absence) of a quantity feature in the L1 plays a role in the acquisition of L2 quantity. Similar findings have also been reported in studies by Pajak and Levy (2014), Meister et al. (2015), and Lee and Mok (2018), among others.

This study presents a case of L1 Mandarin Chinese and Japanese speakers acquiring L2 English semivowels, which involves both phonetic status and features.

2.2 Semivowels in English, Chinese, and Japanese

Semivowels are common in world languages (Maddieson 1984). In English, there are two semivowels, which are the palatal /j/ (e.g., yes) and labial-velar /w/ (e.g., was). These semivowels are highly resonant, similar to the vowels /i/ (e.g., tea) and /u/ (e.g., too), respectively. However, they distinguish themselves from vowels by involving constriction and requiring less energy in their production (Lander and Carmell 1996). Besides acoustic differences, the functions of semivowels and vowels in English differ in that semivowels are used to release a vowel (or diphthong), and they are always placed next to the nucleus in consonant cluster contexts. Vowels, on the other hand, are only used as nuclei (Borden and Harris 1980). Moreover, in English, /j/ and /w/ can occur immediately before the vowels /i/ and /u/, such as in words year and woozy, and these words are lexically contrastive to words containing only /i/ and /u/, such as in year versus ear, and woozy and oozy.

In Japanese, both /j/ and /w/ exist as phonemes. While Japanese /j/ and English /j/ are highly similar, Japanese /w/ differs from English /w/ in an important articulatory feature – lip rounding. Specifically, Japanese /w/ is unrounded while its English counterpart is rounded (Labrune 2012; Okada 1991; Vance 1987, 2008). The two semivowels in Japanese also have their corresponding vowel counterparts /i/ and /u/. Acoustically, Japanese /i/ and /u/ resemble English /i/ and /u/, except that Japanese /u/ is unrounded whereas English /u/ is rounded (Nishi et al. 2008; Tsujimura 2014). Unlike English, Japanese /j/ and /w/ follow different phonotactic rules, where they cannot precede the vowels /i/ and /u/, respectively. In other words, the syllables */ji/ and */wu/ are not legal according to Japanese phonotactic grammar.

In Mandarin Chinese (hereafter referred to as Chinese), the semivowels [j] and [w] also exist. Phonetically, Chinese [j] is extremely similar to its English counterpart /j/. Chinese [w] is also similar to English /w/, with both being rounded. However, Chinese [j] and [w] are not phonemes but phonetic variants of [i] and [u] (Duanmu 2007; Lin 2001; Lin 2007). For example, the Chinese word yī ‘one’ can be pronounced as either [ji] or [i], and the word wū ‘room’ as [wu] or [u]. Although producing [j] and [w] is not required for words such as yī and wū, it is important to note that The Standard Chinese Proficiency Test Guidelines (Liu 2000), serving as the official guide for standard Chinese assessment, has prescribed that a weak homorganic on-glide should be pronounced before high vowels such as [i] and [u] in standard Chinese. In this regard, a recent study by Chan (2023) showed that native Chinese speakers produced [j] and [w] before the vowels [i] and [u] in their reading of individual words; however, the degree of [j] and [w] articulation varied both across words and speakers.

To summarize, Table 1 presents a crosslinguistic comparison of the two semivowels in the three languages.

2.3 L1 Japanese speakers’ perception of English semivowels

The acquisition of L2 English semivowels by L1 Japanese speakers presents an intricate case for evaluating the influence of the L1 on L2 phonology. Zhou and Nakayama (2023) conducted a study on the perception of English semivowels by L1 Japanese, advanced L2 learners of English. The phonological contexts tested were /j/ and /w/ before high vowels /i/ and /u/, as well as /i/ and /u/ without the preceding semivowels, mirroring English lexical contrasts found in yeast versus east and woof and oof. Through an AX discrimination task and a two-alternative forced choice identification task, the study found that Japanese speakers had much trouble perceiving /j/ and /w/ before /i/ and /u/, respectively. Crosslinguistically, Japanese /j/ is acoustically similar to English /j/, but different from English in its distribution, as Japanese /j/ cannot occur with /i/ due to phonotactic constraints (*/ji/) in contemporary Japanese. In the context of the PAM, it would be assumed that English /j/ is assimilated into the Japanese /j/ category; that is, Japanese speakers would map the phonetics of English /j/ to the Japanese /j/ category. In such a case, Japanese speakers’ perception of English /j/ would be expected to be excellent, irrespective of the phonological context. However, the finding of Japanese speakers’ misperception of /j/ suggested that the L1 phonotactic constraint possibly inhibited their perception of /j/ before /i/. On the other hand, while /w/ is also similar between Japanese and English, Japanese /w/ differs from English /w/ not only in phonotactic rules (Japanese */wu/ vs. English /wu/) but also in the lip-rounding feature. The study found that Japanese speakers’ trouble with /w/ was due to their unfamiliarity with the roundness feature in English /w/, which outweighed the effects of the phonotactic constraint.

The crosslinguistic differences in semivowels in English, Japanese, and Chinese present a fascinating yet complex case for investigating the influence of L1 on the acquisition of L2 phonology. Therefore, the current study extends Zhou and Nakayama’s (2023) study and examines L1 Chinese speakers’ perception of L2 English semivowels, comparing these L2 data with Zhou and Nakayama’s L1 Japanese data. This research provides novel insights into how L1 phonological systems influence the perception of L2 phonemes.

4.1 Participants

Twenty L1 Chinese speakers (6 M and 14 F, average age = 22.05) participated in the study. At the time of the experiment, they were matriculated students in universities in the U.S. and were recruited through advisement. In Zhou and Nakayama (2023), the participants were 22 L1 Japanese speakers (6 M and 16 F, average age = 23.50) who were also degree-seeking students in U.S. universities, and 10 L1 English speakers (4 M, 4 F, and others 2, average age = 20.90). Table 2 shows a language background comparison between the Chinese participants in this study and the Japanese participants in Zhou and Nakayama’s study. It includes measures in length of residence (the amount of time the participants spent living in the U.S., in years), age of arrival (the age at which the participants came to the U.S.), age of English learning onset (the age at which the participants began studying English), and daily English use (the frequency with which participants use English daily in the U.S., as a percentage). The standard deviations of these measures are in parentheses.

Four two-sample t-tests were performed to compare the four metrics. The results revealed a significant difference between the two groups in self-reported daily English use, t(40) = 2.68, p < 0.05, but no difference in length of residence, t(40) = 0.07, p = 0.94, age of arrival, t(40) = 1.61, p = 0.11, and age of English learning onset, t(40) = 1.96, p = 0.06. Given these results, along with the fact that these two groups of participants had similar English language preparation and readiness for studying in the U.S., the two groups of participants were regarded as experienced speakers of English with comparable functional proficiency.

4.2 Stimuli

The stimuli used were identical to those in Zhou and Nakayama (2023), which were 60 disyllabic English-like nonce words (e.g., critical: eesha /ˈiːʃə/, yeesha /ˈjiːʃə/, oosha /ˈuːʃə/, woosha /ˈwuːʃə/; filler: leesha /ˈliːʃə/). A male native speaker of American English produced five tokens of each word in a professional sound studio. Two out of the five tokens were chosen for the study, with their intensity being scaled to 68 dB. The words containing an initial semivowel were, on average, longer than those containing an initial vowel. This duration difference was not normalized, as it would be unnatural for /ji/ and /i/, as well as for /wu/ and /u/, to have the same duration due to the additional segment in /ji/ and /wu/. The average duration of the words containing /i/, /ji/, /u/, and /wu/ in milliseconds are listed in Table 3.

4.3 Procedure

The procedure was also identical to that described in Zhou and Nakayama (2023). The participants completed a consent form, a language background questionnaire, a headphone test, and a word familiarization session. During the familiarization session, the participants viewed the orthographic forms of the words and heard each word pronounced once. Then, they completed two perception tasks: an AX discrimination task (120 trials) and a two-alternative forced choice (2AFC) identification task (120 trials). There was a break between these two tasks. In the AX discrimination task, they were asked to judge whether the two words in each trial were the same word or different words (e.g., “same”: eesha-eesha; “different” eesha-yeesha). The two words within each trial were separated by a 200-millisecond interstimulus interval. After hearing each trial, they were given 3 s to respond. Each “same” trial always comprised two different tokens of the same word. In the 2AFC identification task, they were asked to listen to and identify each heard word from two provided options. After hearing each word, they were again given 3 s to respond. The experimental procedure lasted approximately 45 min, and all the participants received nominal fees for their participation.

5.1 Accuracy in the AX discrimination task

Figure 1 displays the proportions of correct responses for the critical items among different language groups in the discrimination task. The L2 Japanese and L1 English control data is from Zhou and Nakayama (2023). The proportions of correct responses for the filler items can be found in Appendix A.

Figure 1:

Mean accuracies and standard errors by language background, trial type, and word group in the discrimination task.

As expected, the L1 English control group achieved high accuracies across the board, suggesting that English speakers took advantage of their L1 sensitivity to distinguishing these sounds, which form phonological contrasts in their language. On the other hand, both groups of L2 participants achieved lower accuracies across different conditions. Notably, both groups of participants appeared to exhibit a strong bias towards the “same” trial types in the “u/wu” word group, as indicated by the higher accuracy rates for the “same” than the “different” trial types. In other words, the participants were more accurate with /u/-/u/ and /wu/-/wu/ pairs than with /u/-/wu/ and /wu/-/u/ pairs. In the “i/ji” word group, the Chinese speakers again showed some “same” bias, whereas their Japanese counterparts did not exhibit such a tendency.

A mixed-design Analysis of Variance (ANOVA) was conducted for the Chinese and Japanese data, predicting response accuracy from three factors: language background (Chinese and Japanese), trial type (“same” and “different”), and word group (“i/ji” and “u/wu”). The model also incorporated an error term accounting for variability among participants within each trial type and word group condition. The model revealed a significant main effect of trial type, F(1, 40) = 7.20, p < 0.05, and a significant interaction effect between trial type and word group, F(1, 40) = 21.65, p < 0.001. The effect of language background was not significant, F(1, 40) = 0.62, p = 0.43. These results indicate that the effect of word group on response accuracy varied depending on the trial type. Post-hoc pairwise comparisons using Tukey’s Honest Significant Difference (HSD) test revealed that both Chinese and Japanese participants’ accuracy was significantly higher for the “same” trial type than for the “different” trial type in the “u/wu” word group (Chinese: Mean difference = 0.27, 95 % CI [0.05, 0.48], p < 0.01; Japanese: Mean difference = 0.23, 95 % CI [0.03, 0.44], p < 0.05). These results confirmed that both groups of participants exhibited a response bias towards the “same” trial types for the “u/wu” word group but not for the “i/ji” word group.

Based on the statistical analyses, it is evident that a strong response bias is present in the data. To address this and shed more light on the data, d prime, a sensitivity measure that adjusts for response biases, was also computed for additional analysis in the next section.

5.2 Sensitivity (d′) in the AX discrimination task

Rooting from the signal detection theory, d prime (d′) is a sensitivity or discriminability measure assessing one’s ability to distinguish between signal and noise. In the current task, d′s were computed with the following procedure. First, the hit, miss, correct rejection, and false alarm rates of participants’ discrimination responses were calculated. In the context of the current task, “hit” refers to correctly identifying a “different” trial as “different,” “miss” refers to incorrectly identifying a “different” trial as “same,” “correct rejection” refers to correctly identifying a “same” trial as “same,” and “false alarm” refers to incorrectly identifying a “same” trial as “different.” Then, the values for Z(hit) – Z(false alarm) were calculated, where Z is the inverse-normal transformation. To avoid undefined values caused by extreme probabilities from the inverse-normal transformation, a log-linear rule was applied (Hautus 1995). Finally, d’s were identified with the values of Z(hit) – Z(false alarm) according to the Independent Observation Model (Macmillan and Creelman 2005). By this calculation, a larger d’ indicates more successful discrimination. Figure 2 displays the mean d’s and standard errors for the “i/ji” and the “u/wu” word group by the Chinese, Japanese, and English participants.

Figure 2:

Mean d′s and standard errors by language background and word group in the discrimination task.

The d′s by the Chinese and Japanese participants underwent statistical analyses. A two-way, mixed-design ANOVA was conducted to examine the effects of language background (Chinese and Japanese) and word group (“i/ji” and “u/wu”) on d’ while accounting for within-subjects variability among participants. The model found no significant effects of language background, F(1, 40) = 1.25, p = 0.27, word group, F(1, 40) = 0.32, p = 0.57, or their interaction, F(1, 40) = 0.16, p = 0.69. Overall, these results suggest that Chinese and Japanese learners did not differ in their sensitivity to /j/ and /w/.

Based on the accuracy and d’ analyses, it can be concluded that Chinese and Japanese participants did not differ in their discrimination performance. This finding will be discussed along with the results from the identification task in the discussion section.

5.3 Accuracy in the 2AFC identification task

Figure 3 shows the proportions of correct responses for the critical items by the three language groups in the identification task. The L2 Japanese and L1 English control data is from Zhou and Nakayama (2023). The results for the filler items can be found in Appendix B.

Figure 3:

Mean accuracies and standard errors by language background and word group in the identification task.

The response accuracy data was fit with a two-way, mixed-design ANOVA with language background (Chinese and Japanese) and word group (“i”, “ji”, “u”, and “wu”) as factors. The model also included an error term to account for participant variability within each word group. The model showed a significant main effect of language background, F(1, 40) = 8.08, p < 0.01, and a significant main effect of word group, F(3, 120) = 5.31, p < 0.01. However, there was no significant interaction between language background and word group, F(3, 120) = 0.73, p = 0.53. These results suggest that the Chinese and Japanese participants’ identification performance differed and that participants’ identification performance for the four types of words also differed. Post-hoc comparisons using Tukey’s HSD tests confirmed that (a) in terms of language background, Japanese participants’ overall identification performance was significantly better than their Chinese counterparts’ (Mean difference = 0.12, 95 % CI [0.05, 0.19], p < 0.01); and (b) in terms of word type, participants’ identification of /u/ was significantly less accurate than that of /i/ (Mean difference = −0.15, 95 % CI [−0.28, −0.02], p < 0.05), and their identification of /wu/ was significantly better than that of /u/ (Mean difference = 0.16, 95 % CI [0.03, 0.29], p < 0.05).

Overall, the statistical analyses showed that Japanese participants performed reliably better than the Chinese participants at the identification task. Moreover, both groups of participants had varied performance in identifying different word types. Notably, participants’ ability to identify /wu/ was superior to that of /u/. Given the task’s design (a two-alternative forced-choice task), the difference in performance between /wu/ and /u/ suggests a word bias. Specifically, the participants were more inclined to perceive a /w/ before /u/ when hearing /u/, but such an inclination was not observed for /j/ before /i/. To gain deeper insights into participants’ perceptual sensitivity to the two semivowels, d’ was computed for further analysis in the subsequent section.

5.4 Sensitivity (d′) in the 2AFC identification task

Sensitivity d′ was used to measure the participants’ detectability of each semivowel. In the context of this task, “hit” refers to correctly identifying a semivowel before the vowel (detecting the presence of a semivowel), “miss” refers to misidentifying a semivowel as a vowel (failing to detect the presence of a semivowel), “correct rejection” refers to correctly identifying a vowel as a vowel (detecting the absence of a semivowel), and “false alarm” refers to misidentifying a vowel as a semivowel (failing to detect the absence of a semivowel). Figure 4 shows the mean d′s and standard errors by word group and language background in the identification task.

Figure 4:

Mean d′s and standard errors by language background and word group in the identification task.

Statistical analyses were performed on the d’s for the Chinese and Japanese learners. A two-way, mixed-effects ANOVA was conducted to assess the effects of language background (Chinese and Japanese) and word group (“i/ji” and “u/wu”) on d′. The model incorporated an error term for participants within each word group. The model revealed a significant main effect of language background, F(1, 40) = 7.30, p < 0.05, suggesting that the perceptual sensitivity of the two groups of participants differed. Neither the main effect of word group, F(1, 40) = 0.71, p = 0.41, nor the interaction between language background and word group, F(1, 40) = 2.18, p = 0.15, reached statistical significance. Post-hoc pairwise comparisons using Tukey’s HSD tests revealed that the Japanese participants were significantly better than their Chinese peers in perceiving /j/ in the “i/ji” word group (Mean difference = 1.00, 95 % CI [0.15, 1.86], p < 0.05); however, the two groups of participants did not differ in perceiving /w/ before /u/ (Mean difference = 0.58, 95 % CI [-0.28, 1.44], p = 0.29).

In summary, the accuracy and d’ analyses suggest that overall, the Japanese participants were better at perceiving the semivowels than the Chinese participants, and their advantage primarily lies in the semivowel /j/.