The weight of VOT and f0 in English stop differentiation: 2L1 bilinguals versus monolinguals

2.1 Weight of VOT and f0 in stop differentiation in English

It is generally agreed that VOT is the primary cue in distinguishing voiceless and voiced word-initial stops in English (Abramson and Lisker 1985; Lisker 1978; Raphael 2021), and that f0 is a secondary cue (Abramson and Lisker 1985; Dmitrieva et al. 2015; Francis et al. 2008; Schertz et al. 2020; Whalen et al. 1990, 1993; Yu 2022; but see Francis and Nusbaum 2002, Kwon 2014 arguing that f0 has only a marginal role, and Repp 1982, Shultz et al. 2012, Yu 2022 for the tradeoff relation between VOT and f0).

The voiceless-voiced distinction for English word-initial stop consonants is mainly realized as differences in VOT: the interval between a stop release and the onset of vocal fold vibration (Lisker and Abramson 1964). VOT is considered negative when voicing begins prior to the stop release. Conversely, VOT is considered positive when voicing starts after the stop release. When the voicing onset occurs 25 milliseconds (ms) or more before the release, it is termed as voicing lead (Laver 1994; Lisker and Abramson 1964). Short-lag VOT refers to instances where voicing onset begins approximately 20 ms or less after the release (Laver 1994; Lisker and Abramson 1964). In contrast, long-lag VOT indicates that the voicing onset occurs 25 ms or more after the stop release (Laver 1994; Lisker and Abramson 1964). In English, although both voiceless and voiced stops may have positive VOTs, voiceless stops belong to the long-lag category and voiced stops to the short-lag category (Klatt 1975; Kong et al. 2012; Lisker and Abramson 1964). In addition, voiced stops may also have voicing-lead VOT.

In terms of f0, the general pattern is that voiceless stops are associated with higher f0 and voiced stops with lower f0, irrespective of the specific phonetic realization of the voicing contrast (Dmitrieva et al. 2015; Hanson 2009; House and Fairbanks 1953; Lehiste and Peterson 1961; Ohde 1984; Whalen et al. 1993). This tendency has been found not only in English, but also in Burmese (Shimizu 1996), French (Kirby and Ladd 2016), German (Kohler 1982), Hindi (Shimizu 1996), Italian (Kirby and Ladd 2016), Japanese (Shimizu 1996), Spanish (Dmitrieva et al. 2015), Thai (Gandour 1974, Shimizu 1996), among other languages.

2.2 Weight of VOT and f0 in stop differentiation in Japanese

Japanese has both voiceless (/p/, /t/, /k/) and voiced stops (/b/, /d/, /g/) (Itoh et al. 1980; Shimizu 1989; Vance 1987). These stops are found only in word-initial and word-medial positions (Okada 1999; Vance 1987). Regarding the voiced stops, Itoh et al. (1980) report only negative VOTs for these consonants (see also Kobayashi 1981; Shimizu 1989, 1996, 1999). However, a recent trend in Japanese speakers born after the 1930s involves devoicing word-initial voiced stops. This results in two variants: a voiceless unaspirated variant with a short-lag VOT, and the other variant with voicing-lead VOT. In Tokyo Japanese, fewer than 25 % of voiced stops exhibit voicing-lead VOTs (Kong et al. 2012; Liu and Takeda 2024; Takada 2004, 2011). Similarly to English, VOT is the primary cue in distinguishing the voiceless stops from voiced ones in Japanese (Byun 2019; Gao and Arai 2019; Lisker 1978; Raphael 2021; Shimizu 1996, 1999).

F0 has been identified as a secondary cue for distinguishing between the two sets of stops in Japanese (Homma 1980; Shimizu 1993, 1996, 1999; Takada 2011; Takada et al. 2015). While debates persist regarding the precise f0 contours in various accent contexts, there is a general consensus that Japanese tends to exhibit high f0 at the onset of a vowel after a voiceless stop and low f0 after a voiced stop similarly to English (Gao et al. 2019; Kawasaki 1983; Mizuguchi and Tateishi 2018). The onset f0 in Japanese is shaped by the voicing contrast at the phonological level rather than by VOT, also similarly to English (Gao and Arai 2019; Ishihara 1998; Kawasaki 1983; Shimizu 1993, 1996, 1999; Takada 2011).

4.1 Participants

We recorded six monolingual American English speakers (three female and three male), all born and raised in the United States. Among which, three originated from California and the remaining three from Texas (hereafter the monolingual group). Additionally, six Japanese-American English 2L1 bilinguals (four female and two male; henceforth the JE group) were recorded. Born in the western part of Japan, they moved to America during infancy and spent their formative years there. Two completed their college education in America, while the other four did so in Japan. Despite being residents of Japan during recording, they are fluent in both Japanese and American English. All participants in the JE group acquired both languages naturally before the age of three, and are classified as 2L1 bilinguals (Genesee and Nicoladis 2007; McLaughlin 1978). To minimize the influence of diverse English accents, we exclusively selected American English speakers for the monolingual and JE groups. At the time of recording, all participants were approximately 30–35 years old.

4.2 Acoustic analysis

The first author segmented and annotated the recordings using speech waveforms and spectrograms in Praat, following the guidelines in Lisker and Abramson (1964), Lavoie (2001), and Riney et al. (2007). For voiceless stops, the interval between the stop release and the onset of glottal vibration marked the VOT. Oral closure was identified by the absence of acoustic energy in the formant frequency band, while oral release was indicated by a rapid onset of energy in the formant frequency range. Regarding voiced stops, their VOT values were measured as negative if voicing preceded the release. Stops without audible glottal vibration in word-initial positions during text reading were omitted since it was difficult for us to determine the exact onsets of this kind of stops and thus difficult to measure the exact VOT durations. The stops focused on were word-initial /p/, /t/, /k/, /b/, /d/, and /g/. F0, measured in Hertz, was determined at the onset of the following vowel of a word-initial stop on narrowband spectrograms in Praat with a 0.025 window length. Tokens without a pitch track at the vowel onset were excluded from consideration since it was impossible to measure their onset f0s. In total, 97 data points in the word reading task and 69 data points in the text reading task were excluded from consideration. The final dataset consisted of 1,730 samples, 885 from the monolingual group and 845 from the JE group. The breakdown was as follows: (i) in the word list reading task, 339 tokens from the monolingual group and 332 from the JE group; and (ii) in the text reading task, 546 and 513 from the two groups respectively.

4.3 Prediction of the present study

Both English and Japanese use VOT as the primary cue and f0 as the secondary cue in word-initial stop differentiation to distinguish voiceless stops from voiced ones. According to CLI, the English speech production of the JE group may be influenced by Japanese. However, since the two languages rely on VOT and f0 in stop differentiation similarly, it was predicted that the monolingual and JE groups have no statistically significant differences in terms of the weights of VOT and f0 in stop differentiation in both the word list and text reading tasks.

5.1 Weights of VOT and f0 in word list reading

Table 1 presents the results for the logistic regression model carried on the data of the word list reading task from the monolingual and JE groups.

In Table 1, the Omnibus tests of model coefficients indicate that at least one of the independent variables is statistically significantly correlated with the dependent variable voicedness in the word list reading task (p < 0.001). The variables of group, VOT duration, f0, and VOT_group are statistically significantly correlated with voicedness (p < 0.001, p < 0.001, p < 0.001, p < 0.001). In terms of the variable group, the positive coefficient suggests that the JE group tends to produce stops with characteristics that are more indicative of voicing compared to the monolingual group (B = 6.314). This implies a systematic difference in the speech production patterns between the monolingual and JE groups, particularly regarding the differentiation of stops based on VOT and f0. The negative coefficient value of VOT duration suggests that a longer VOT is associated with a decreased likelihood of the stop being produced as voiced (B = −97.185). This finding aligns with the phonetic characteristics of English, where voiceless stops typically exhibit longer VOTs compared to voiced stops. The negative coefficient value of f0 indicates that stops followed with higher f0s are less likely to be produced as voiced stops, which aligns with the tendency that f0s following voiceless stops are generally higher than those following voiced stops (B = −0.027). The negative coefficient of the VOT_group variable indicates that VOT has a stronger influence on stop classification in the monolingual group compared to the JE group (B = −68.962): VOT has a heavier weight as a cue in differentiating the stops within the monolingual group’s speech production compared to the JE group in the word list reading task. The variable f0_group has not emerged as statistically significant (p = 0.075), which suggests that the two groups do not differ statistically significantly in using f0 as a cue to differentiate English stops in word list reading.

5.2 Weights of VOT and f0 in text reading

Table 2 presents the results for the logistic regression model based on the data of the text reading task from the monolingual and JE groups.

Table 2 shows that the variables VOT duration and f0_group have statistically significant correlations with stop differentiation in the text reading task (p < 0.001, p = 0.048). In terms of VOT duration, the longer the VOT duration of a stop, the more likely it is produced as voiceless, in line with the general tendency of VOT duration in English (B = −155.816). The variable VOT_group has not emerged as statistically significant (p = 0.444). This indicates that the two groups do not have a large difference in the weight of VOT in distinguishing stops in text reading. Plainly, VOT does not significantly differentiate stop production between the monolingual and JE groups in the context of text reading. The variable f0_group has a significant correlation with stop differentiation (p = 0.048), suggesting that f0 has a different weight as a cue to distinguish the stops in the two groups: f0 has a heavier weight in the JE group than in the monolingual group in the text reading task (B = 0.011).

5.3 Comparison of Tables 1 and 2

A comparison of Table 1 with Table 2 gives differences in cue weighting in different tasks in the two groups: (i) the monolingual group utilizes VOT as a cue more than the JE group in the word list reading task, while the JE group utilizes f0 more in text reading; and (ii) the two groups do not have significant differences in the weight of the f0 cue in word list reading and the weight of the VOT cue in text reading. This demonstrates that significant differences as well as some similarities exist between the monolinguals and JE 2L1 bilinguals.

Liu and Takeda (2024) have noted that the monolingual and JE groups do not have significant differences in VOT duration in the word list reading task. The present study reports that they have significant differences in terms of the weight of the VOT cue in word list reading. The difference in the weighting of the VOT cue between the monolingual and JE groups should not be attributed to CLI. Both Japanese and English use VOT as the primary cue and f0 as a secondary cue for distinguishing voiceless and voiced stops. The VOT durations in Japanese are generally shorter than those in English (Chodroff and Wilson 2017; Gao and Arai 2019; Hunnicutt and Morris 2016; Lisker and Abramson 1964; Shimizu 1996). If CLI were the main or sole factor, statistically significant differences in VOT durations should have been witnessed between the two groups, rather than in the weighting of VOT as a cue. Therefore, the difference in VOT weighting is more likely due to bilinguals’ heightened sensitivity to VOT as a cue, rather than CLI from the phonetic system of Japanese. The difference in VOT weighting could indicate that the JE group is more attuned to utilizing VOT as a cue for stop differentiation, even though their VOT durations remain comparable to those of the monolingual group. This could result from the need to maintain a clear contrast between the two languages that both rely on VOT as the primary cue by slightly allocating more cognitive resources or articulatory control toward relying on VOT as the primary cue for English stop differentiation. In short, the strategy chosen by the JE group seems to result from efficiency, as they adapt their use of VOT as a cue to achieve distinctiveness between the two languages.

In text reading, Liu and Takeda (2024) have shown that the VOT duration difference between the monolingual and JE groups is smaller than that in word list reading. This reduction in the VOT duration difference in text reading may make it challenging for the JE group to heavily depend on VOT to differentiate between voiceless and voiced stops. Consequently, they shift their reliance more towards f0 as a cue in text reading.

Liu and Takeda (2024) have reported that the JE group is never completely the same as the monolingual group either in terms of VOT duration, VOT polarity, or f0 value: the JE group has employed an extreme approach to positive VOTs, where they must distinguish among all six stops in Japanese and English, though they adopt an intermediate approach to negative VOTs, where they just need to distinguish the three voiced stops. In the present study, the JE group has different weights for VOT or f0 compared with the monolingual group. However, the JE group has never assigned statistically significant different weights to both VOT and f0 in either word list reading or text reading. It appears that efficiency can explain the choice of strategy by the JE group: emphasizing one cue over the other is enough to distinguish the stops. Therefore, the JE group has only assigned a different weight from the monolingual group to one cue in each task.

In summary, while CLI may play a role in bilingual language processing, it does not sufficiently explain the differences observed in this study. The variations in cue weighting of VOT and f0 appear to stem more from the bilinguals’ adaptive strategies aimed at efficiency in managing phonetic distinctions, rather than from direct CLI from the phonetic system of Japanese. Initially, it was predicted that there were no statistically significant differences in the utilization of VOT and f0 cues for stop differentiation between the monolingual and JE groups. However, our results do not support this prediction. As a result, this suggests that there may be noteworthy distinctions in speech production strategies between monolinguals and 2L1 bilinguals, even in areas where the two languages exhibit similar phonetic patterns. One plausible explanation appears to be that 2L1 bilinguals need to distinguish between their two languages, leading them to make differences in areas where the two languages are similar to each other. In the subsequent sections, we will compare another type of 2L1 bilingual with the monolingual and JE groups to further examine the conclusion drawn in this section.

6.1 Participants and acoustic analysis

Similar to the experiment in Section 4, the same word lists and text from the PAC project were given to participants three weeks before recording for them to familiarize themselves with the material. Six 2L1 bilingual speakers of Mandarin and American English, four female and two male, were recorded (the ME group). They were all born in China and moved to America as infants. They were also residents of California and around 30–35 years old at the time of recording. All of them are college graduates. Procedures of recording, segmentation, and acoustic analysis are the same as those for the first experiment in Section 4.

6.2 Statistical analysis: ME group versus monolingual group

Similarly to Section 5, two logistic regression model tests are conducted on word list reading and text reading data respectively. The dependent variable is still voicedness (voiceless, voiced), with voiceless as the reference level. The independent variables are group (monolingual group, ME group; with the monolingual group as the reference level), VOT duration, f0, VOT_group, and f0_group. The data from the monolingual group are the same as those used in the statistical analysis in Section 5. The variables VOT_group and f0_group were also generated using the Compute Variable function on the Transform menu in SPSS and involved combined information from the columns VOT duration or f0 value and group. Altogether, 826 samples were collected from the ME group, 338 from word list reading and 488 from text reading. In the following, Tables 3 and 4 respectively report results of the logistic regression model for the data of word list reading and of text reading.

The logistic regression model for word list data in Table 3 demonstrates that the VOT_group variable emerges as the sole statistically significant variable (p < 0.001), suggesting that it has a heavier weight in stop differentiation in the monolingual group than in the ME group (B = −27.151), which conforms to our prediction. There are no significant differences in terms of f0_group (p = 0.244). Our provisional explanation is that since VOT has a heavier weight than f0 in Mandarin, the ME group relies on VOT less in English stop differentiation to make the distinguishing of Mandarin and English easier and more efficient. The two groups do not have a significant difference in the weight of f0 as a cue although f0 has a smaller role in stop differentiation in Mandarin. This is not explicable using CLI: if CLI is the main factor here, the ME group is expected to use f0 as a cue less than the monolingual group. However, this result seems to be in line with efficiency: since f0 is not a major cue for stop differentiation in Mandarin, relying on f0 in English stop differentiation as much as the monolingual group will not require unnecessary adjustments. In Table 4, we will take a close look at data from text reading.

In Table 4, the variables group, VOT duration, f0, and VOT_group have given statistically significant differences (p = 0.020, 0.000, 0.045, 0.007). The positive coefficient of the variable group indicates that the ME group tends to produce stops that show more signs of voicing compared to the monolingual group in text reading (B = 0.461). This implies a systematic difference in distinguishing stops based on VOT and f0 between the ME and monolingual groups. The negative coefficient value of VOT implies that a stop with a long VOT is not likely to be produced as a voiced stop, consistent with the observation that voiceless stops in English generally have longer VOTs than the voiced stops (B = −127.615). In a similar vein, the negative coefficient of f0 indicates that stops followed by higher f0s are less likely to be produced as voiced stops, in alignment with the tendency that f0s following voiceless stops are higher than those following voiced stops (B = −0.007). The negative coefficient value for the VOT_group variable indicates that the ME group exhibits a different pattern in the production of stops compared to the monolingual group, with potentially reduced reliance on VOT as a distinguishing cue (B = −11.777). In addition, the variable f0_group has been rejected as a statistically significant variable (p = 1.000). This suggests that the ME and monolingual groups do not have a statistically significant difference in terms of the weight of f0 as a cue in English stop differentiation in text reading.

A comparison of Tables 3 and 4 firstly gives a clear contrast between the results from word list and text reading respectively: while only the VOT_group variable has given a statistically significant result for data from word list reading in Table 3, four variables have emerged as statistically significant for data from text reading in Table 4. This can be explained by the fact that the ME group can meticulously regulate their speech when reading word lists, resembling monolingual speakers more closely. Conversely, because maintaining such precise control during text reading is challenging, greater distinctions between the monolingual and ME groups emerged. Both Tables 3 and 4 suggest that the two groups differ in the weights of VOT as a cue in stop differentiation, in line with our prediction based on the concept of efficiency. The negative coefficients of the VOT_group variable in both Tables 3 and 4 suggest that the monolingual group relies more on VOT than the ME group in both word list reading and text reading tasks (B = −27.151, −11.777). The two groups do not have a significant difference in the weight of f0 in either word list or text reading (p = 0.244, 1.000).

Looking at Tables 3 and 4 together, although VOT is a major cue for stop differentiation in both Mandarin and English, the ME group places less emphasis on VOT than the monolingual group does. Furthermore, while the role of f0 in stop differentiation is notably reduced in Mandarin, the ME group does not show significant differences from the monolingual group in how they use f0 as a cue in either word list or text reading in English. These results from the ME group, concerning both VOT and f0, seem to challenge CLI: if CLI were the primary factor, the ME group would be expected to rely on VOT more than the monolingual group and to use f0 less than the monolingual group. Instead, the concept of efficiency seems to offer a more fitting explanation: like the JE group, the ME group differs from the monolingual group in the use of only one cue in both word list and text reading, suggesting that this distinction is sufficient to differentiate their two languages. More specifically, the ME group relies less on VOT because they can utilize f0 just as much as the monolingual group does, reducing the need to heavily depend on VOT. In this context, efficiency seems to be the key principle guiding their speech production.

Furthermore, similar to the results in Tables 1 and 2, the results in Tables 3 and 4 suggest that there are differences between the monolingual and ME groups, while also revealing similarities between them. Neither the JE nor ME group differs from the monolingual group in both the weights of VOT and f0. While differences may appear in either VOT or f0, they do not occur in both at the same time. This holds true for both word list and text reading tasks.