Plosives are, as a group, the most common consonants in the languages of the world, and the greater part of world languages have voiceless plosives (Ladefoged & Maddieson, 1996; Maddieson, 1984). More specifically, Moran and McCloy (2019) found that /p/ had an 86% frequency in the languages they surveyed (2,186 languages); the frequencies for /t/ and /k/ were 68% and 90% respectively. A crucial characteristic that contrasts voiceless plosives among the languages of the world is their voice onset time (VOT) (e.g., Hayward, 2000). This refers to the time that elapses between the release of the plosive and the onset of vocal fold vibration (of the following vowel or approximant). There are three types of VOT: negative VOT, zero VOT, and positive VOT. The latter type can be further subdivided into long and short positive VOT; plosives with a long positive VOT are said to be aspirated (e.g., Wayland, 2019). The release stage of the plosive is evidenced acoustically by a burst or transient on the waveform, and a plosion bar on the spectrogram. As noted by Bóna and Auszmann (2014) and Ladefoged (2003), among others, measuring VOT is often marred with complications. One of them is that voiceless plosives can be released in multiple stages, thus producing multiple bursts (MBs). Figure 1 displays the production of the initial plosive in the word kids by a non-native speaker of English with Hindi as his first language (L1). The velar has two transients (as shown on the waveform) and two corresponding plosion bars (shown on the spectrogram.

A few explanations for the production of MBs have been advanced. According to Plauché (2001), this phenomenon may be due to rapid intraoral pressure during closure, which lowers abruptly upon the release of the plosive, producing multiple transients. As will be illustrated later (see sections 1.1 and 1.2), velar plosives have been shown to present many more MBs than other places of articulation (PoAs). Song, Demuth, and Shattuck-Hufnagel (2012) and Plauché (2001) explain that this phenomenon may result from the large area of the velar obstruction and its accompanying slow release, which gives rise to multiple burst releases. Plauché (2001) also notes that some researchers have posited that the closure stage may be conditioned by the Bernoulli force, whereby the quick airflow initiation creates negative pressure at the obstruction place.

To reiterate, aspiration can only be analyzed by measuring the VOT of the plosive, and, as is in many occasions the case when voiceless plosives are produced with MBs, complications arise. In the presence of MBs, it has to be decided whether VOT measurement will be made from the first burst or the last burst (or some other point, see Gráczi & Kohári, 2014). Prior research has determined that different methods produce statistically significant disparate VOT durations, and that factors such as PoA, vowel height, and speech rate have an effect on MBs to varying degrees (Barrera-Pardo, 2022). In line with these past findings, the aim of the current study is to analyze MBs in non-native English and establish a comparison with native English. First, it explores whether VOT values are affected when measuring VOT from the first or last burst. Second, it investigates the effect of PoA, gender, speech rate, and age on MBs.

MBs have been identified in various languages, although few researchers have quantified them systematically (notable exceptions are Gráczi & Kohári, 2014, for Hungarian; Barrera-Pardo, 2022, and Lavoie, 2001, both for Spanish). The following sections review research on MBs in English and other languages.

The data for this study were collected from native and non-native speakers of English with twelve different L1s. Native speaker data were collected to compare VOT measurement method in the presence of MBs and the effects of PoA, speech rate, age, and gender with non-native data. The selected L1s were listed as having the most native speakers in the world (Ethnologue, 2022); the languages are Mandarin, Spanish, Hindi, Bengali, Portuguese, Russian, Japanese, Vietnamese, Turkish, Korean, Farsi (Persian), and German. Arabic was excluded because this language does not have /p/ in its phonological inventory.

The results presented in the preceding sections will be considered in terms of their effect sizes, rather than referring only to their statistical significance (see Cumming, 2012, especially chapter 2). There is a growing trend in the social sciences to answer research questions not with a dichotomous yes or no, that is, whether a result is statistically significant or not. Rather, the focus of statistical interpretation is shifting to a recognition of the magnitude of the detected effects, instead of solely considering its statistical significance below an arbitrary p value (e.g., p < 0.05; see Cohen, 1990). This statistical reform movement is gathering momentum in areas such as second language acquisition (see, e.g, Gass, Loewen, & Plonsky, 2021), and it firmly encourages answering research questions as how much of an effect was found, not as whether or not there was an effect (Cumming, 2012, p. 34).

One aspect of the statistical analyses that merits attention is the decision to use separate analyses for the native and non-native data. One motivation to do this was that, for the analysis of VOT measurement method in non-native English, the data were not normally distributed, and hence non-parametric tests were opted for. In contrast, the data for native English for the same variable were normally distributed. Thus, the analyses were split by language population, which potentially lessens the interpretation of the results. It may have been more fruitful to make a more general analysis on all the data, for example including nativeness as a predictor (i.e., native English vs. non-native English), together with the other factors and their interactions. This is a limitation of the study and future research may circumvent this methodological problem, perhaps by implementing statistical tests that are robust to violations of statistical assumptions such as normality of the data.

The first research question asked about the effect of measurement method of MBs on VOT, that is, whether the VOT values of the voiceless plosives differed substantially when this acoustic characteristic was measured from the first or the last burst to the onset of voicing. In non-native English, it was found that the measurement method in the presence of MBs had a fair impact on VOT. These two methods differed by about 11 ms, which in terms of VOT can be a substantial difference, for example, when describing a voiceless plosive as aspirated or not (e.g., Docherty, 1992). The effect size for this finding was medium, η ² = 0.09, which means that 9% of the variance in VOT values was associated with the MB measurement method. For native English, it was found that the MB measurement method had as well a medium effect size, η ² = 0.07. The two methods in native English differed by almost 13 ms, which also constitutes a meaningful difference. Both results are to some extent comparable with the findings of Barrera-Pardo (2022), who found that in Western Andalusian Spanish 18% of the variance in VOT values was associated with the measurement method in the presence of MBs. This researcher also reported a difference of 8 ms when the voiceless plosives were measured from the first or last burst. In Hungarian, Gráczi and Kohári (2014) reported on the effects of three VOT measurement methods when MBs were present (the first burst, the most intense burst, and the last burst), finding that these differed statistically significantly and with medium effect sizes, similarly to the results of the current study. It could be argued that, obviously, the two VOT measurements would produce different results; the first measurement will logically yield larger values than the second, in all measurements, and it could also be reasoned that with a large enough sample size a significant difference and an associated substantial effect size will be found. There are, however, two problems with this reasoning. First, it is far from clear which measurement method researchers use in their analyses of VOT, since this is many times not reported. Thus, VOT values described in past research might be affected by this methodological inconsistency. Second, statistically, the effect size is not dependent on sample size; it is derived from two parameters, the mean and the standard deviation (for example, Cohen’s d). Thus, effect sizes are not biased by either large or small sample sizes (Plonsky & Oswald, 2014).

The first research question also inquired about the role that PoA may play in VOT values, in the presence of MBs. The effect of PoA on VOT in non-native English was statistically significant, with a medium effect size (η ² = 0.06). In native English this effect was as well statistically significant and much larger, η ² = 0.18. These findings are in line with the data reported in previous literature, especially in the case of native English. The post-hoc tests revealed that in non-native English, for /p/, when measured from the first or the last burst, there was a mean difference of almost 7 ms, with a small effect size (d = 0.22). In native English, the difference amounted to about 8 ms, in this case with a medium effect size (d = 0.44). The results for /t/ are as follows. In non-native English, the mean difference according to measurement method was around 6 ms, and the associated effect size was small (d = 0.26). In native English, this plosive differed in 11.50 ms as per measurement method, which is a sizable amount; the effect size related to this difference was medium (d = 0.66). Lastly, /k/ yielded the largest mean differences and associated effect sizes in both non-native and native English; for the former, the mean difference was about 14 ms, with a large effect size (d = 0.72), and for the latter, there was the same mean difference of around 14 ms, with an associated large effect size (d = 0.82). The effect sizes reported by Barrera-Pardo (2022) for the Spanish dialect he investigated are to different degrees in line with the findings of the current study. In this research, /p/ presented a mean difference of 6 ms according to measurement method, with a large effect size, d = 0.78; the mean difference for /t/ was 5 ms, with a medium effect size, d = 0.65; and for /k/ the mean difference was 13.51 ms, with a very large effect size, d = 1.75. Thus, the data for this Spanish variety are, to a certain extent, comparable to the findings for non-native English reported in the present study, on account of the mean differences found according to the VOT measurement method in the presence of MBs. These cumulative findings therefore suggest that the measurement method plays a significant role in the VOT values obtained, and future research on VOT in which MBs are found should probably be aware of this important methodological issue, as further discussed in Barrera-Pardo (2022).

The role of PoA, gender, age, and speech rate in MBs was the focus of the second research question. Regarding PoA, in non-native English 41% of the plosives analyzed exhibited MBs, with an associated medium effect size. This percentage is higher than in the Spanish dialect researched by Barrera-Pardo (2022), who found that 25% of the plosives presented MBs. It is also much higher than the percentage reported by Lavoie (2001) for American English, whose analysis found that only 9% of the voiceless plosives had MBs. Further, Lavoie found that 11.5% of the Spanish voiceless plosives she examined had MBs, a figure that is clearly much lower than what was reported in the current study. Lastly, Gráczi and Kohári (2014), who quantified MBs in Hungarian, documented that nearly 48% of the voiceless plosives realizations evidenced MBs; this result is somewhat similar to the data reported above for non-native English. Almost 77% of the plosives produced in native English presented MBs (with a very large effect size) a figure that surpasses all the previous results. The difference between non-native English and native English in MBs yielded a statistically significant result, with a medium effect size (Phi = 0.38).

Proceeding with the number of MBs for the three plosives, it was found that in non-native English slightly over a quarter of the bilabials had MBs, and that in native English the percentage was about 34%. This result is in line with the percentage reported for Hungarian by Gráczi and Kohári (2014), 27.5%, but very different from the data presented for English by Lavoie (2001), 5%, and for Spanish by the same author, 0%. Barrera-Pardo (2022) found that in Spanish only about 10% of the bilabials had MBs. The next plosive, /t/, presented MBs in a quarter of its realizations in non-native English, and in native English about 16% of the productions of /t/ had MBs. Gráczi and Kohári (2014) reported many more MBs for /t/ (45%), and Lavoie (2001) found 0.8% of MBs for this consonant in English, whereas she reported 0% MBs for /t/ in her Spanish data. Also for Spanish, Barrera-Pardo (2022) found that about 15% of /t/ productions presented MBs. Finally, in non-native English in the current study /k/ had MBs in nearly 78% of the realizations, and around 67% of the /k/ productions had MBs in native English. This finding is overall in agreement with the data obtained by Gráczi and Kohári (2014), but very different from the 21% of MBs for /k/ reported by Lavoie (2001) for English. In Spanish, Lavoie found that 32% of the /k/ productions presented MBs, which is coincident with the percentage found by Barrera-Pardo (2022). The present study found that non-native English and native English did not vary in the mean number of bursts, which for both varieties was M = 2.40, with a trivial effect size.

Addressing now the number of bursts by PoA, non-native English /p/ had the majority of bursts in the 2-burst rank, as did this consonant for native English. /t/ in non-native English had almost all of the MBs also in the 2-burst category, whereas native English was more spread between the 2- (72%) and 3-burst (28%) ranks; it is remarkable that this consonant did not show any MBs in the 4- and 5-burst categories for either English variety. /k/ had MBs across the entire series of ranks, and with the same percentages for both non-native and native English. Non-native English and native English differed in the occurrence of MBs by PoA in the bilabials, with a small effect size, a result that also applied to the alveolars; for the velars, however, the difference was negligible, as per the effect size observed. The results reported by Barrera-Pardo (2022) for Spanish and by Gráczi and Kohári (2014) for Hungarian are to a great extent comparable to the data of the current study. Thus, this seems to corroborate a general tendency of the voiceless plosives that exhibit MBs. The PoA of the plosives had a large effect on the number of MBs both in non-native English and native English, explaining 23% and 15% of the variance respectively; this concurs with the data reported by Barrera-Pardo (2022). A related question was whether the L1 had an effect on the number of MBs; it was found that the difference in mean bursts by L1 was not statistically significant and the corresponding effect size was small (only 2% of the variance in MBs could be ascribed to the L1). None of the pairwise comparisons yielded significant differences in the mean number of bursts among the twelve L1s. This result is perhaps unexpected, since the various L1s analyzed present very different VOT patterns in their native speech. For example, Russian voiceless plosives are not aspirated (Yanushevskaya & Bunčić, 2015), but Hindi has a phonological contrast between voiceless unaspirated and voiceless aspirated plosives (Ladefoged & Johnson, 2015).

An additional factor that may play a role in MBs is speaker gender. In line with Plauché (2001) and Barrera-Pardo (2022), the current study did not find an association between gender and MBs. Speech rate, as in previous studies (but partially contrary to Barrera-Pardo, 2022), was not found to have an effect on MBs in either non-native English or native English. Lastly, in non-native English, the factor age had a small effect on the number of bursts (ρ = 0.10,), and it played essentially no role in native English. Parveen and Goberman (2012) also reported a lack of association between age and MBs.

Finally, this investigation of MBs may be relevant in a number of speech research areas. MBs have been repeatedly reported in pathological speech when measuring VOT (e.g., Fischer & Goberman, 2010; Kopkalli-Yavuz, Mavis, & Akyldiz, 2011; Novotný et al., 2015). Lacking full control of their motor capabilities, these speakers tend to evidence a substantially higher number of MBs than speakers without speech disorders, as reported in past studies. The cumulative results of the current study, together with those of previous research (Barrera-Pardo, 2022), suggest that MBs are present in nonpathological populations to a larger extent than what may have been normally assumed. VOT has also been extensively explored in second language (L2) perception (Major, 2001) and production (Zampini, 2008). However, to date, the presence of MBs and the complications that may arise when measuring the VOT of L2 speakers’ productions remains an uncharted area. The current study suggests that in non-native English, as reported for the twelve L1s explored, the presence of MBs needs to be considered when analyzing VOT. Systems of automatic detection of VOT (e.g., Kazemzadeh et al., 2006) should probably take into account the potential occurrence of MBs in the data they probe.

Systematically analyzing and quantifying MBs has been rarely addressed in the literature. The current study aimed at discerning the role of MBs in VOT measurement and the potential role of the factors PoA, gender, speech rate, and age in two varieties, non-native English, with data from twelve L1s, and native English. Expanding on previous studies, it was found that the VOT measurement method in the presence of MBs had a substantial effect in both varieties, and it is therefore concluded that researchers who embark on VOT measurement of voiceless plosives take heed of the findings reported in prior research and in the present study. Since VOT is one of the most studied consonant features in the phonetics literature (Whalen, DiCanio, & Dockum, 2022), this recommendation seems all the more pressing. The results of the present study also supported previous research findings that clearly indicate an effect of PoA on MBs, especially for /k/. In addition, this study quantified MBs of voiceless plosives, something that has rarely been done in the literature: MBs are almost routinely noted, but data of the type reported here add to our knowledge base of this phonetic event. These data also suggest that although there are some sizeable differences between non-native English and native English, overall both varieties exhibited similar tendencies when the factors of PoA, gender, age, and speech rate were subjected to analysis. Together with previous research, these findings seem to unveil aspects of voiceless plosives that are not language-specific. To conclude, a number of limitations of the current study will be noted. Data were obtained from only three words in the elicitation passage, and future research of the type reported here should expand to more tokens, and in different vowel contexts, considering that prior research has suggested a potential effect for vowel height on MBs (Barrera-Pardo, 2022; Yang, 2018). The current study focused on aspirated voiceless plosives; it would be pertinent to address voiceless plosives in unstressed positions, and examine whether they also present MBs. In American English, Lavoie (2001) found no MBs for unstressed /p/ and /t/, and only 10% of unstressed voiceless velars presented MBs. She analyzed voiceless plosives in both initial (as in Tucson) and medial position (as in boutique); the present study obtained data only from three monosyllabic words (peas, toy, kids), and it would be relevant to extend this research to multisyllabic words. All these factors may play a role in the production of MBs. In addition, the analysis of speech rate in the current study was conditioned by the recorded speaking task, the elicitation passage, and future research may avoid this limitation by carrying out a different production task.