Pre-service English teachers’ cue-weighting in L2 stop perception^*

English has the two-way voicing distinction of stops, voiced (/b, d, g/) and voiceless (/p, t, k/). While VOT is the primary acoustic cue to signal the voicing distinction of English stops, the VOT cue correlates with F0 at the onset of the following vowel of the stops (Abramson & Lisker, 1985; Francis et al., 2008; Whalen et al., 1993). English voiceless stops usually have longer VOT, which correlates with higher F0 at the following vowel onset, while voiced stops are preceded by relatively lower F0. English-speaking listeners/speakers have different cue-weightings between the two acoustic dimensions varying by many sources. That is, there are individual differences in the relative importance between VOT and F0.

Researchers have sought to identify the sources of individual variation in cue-weightings, examining both environmental factors and individuals’ internal cognitive traits. The environmental factors affecting listeners’ cue-weighting strategies are primarily related to attentional load during speech processing (e.g., Francis et al., 2000). Consequently, research in this area has investigated speech perception patterns under various distracting conditions, simulating everyday challenges such as noise or multitasking. This line of research typically uses a dual-task paradigm in which listeners respond to auditory stimuli while performing another task. The examples of the additional task include math calculation or letter recall task, which increases cognitive loads during the speech perception task (e.g., Gordon et al., 1993; Kong & Lee, 2018; Mattys & Wiget, 2011).

Gordon et al. (1993) studied how listening conditions influence the use of multiple acoustic cues in speech perception. They focused on the voicing contrast in English stops identified with the primary VOT cue and the secondary F0 cue. By using a dual-task paradigm, Gordon et al. (1993) examined how cognitive distractions manipulated by solving math problems, would affect listeners’ reliance on these cues. In their experiment, participants completed a speech identification task under two attentional conditions: one with no distractions (no-distractor) and the other with a math-solving task embedded (distractor). Results showed that in the distractor condition, listeners’ reliance on the primary VOT cue decreased, while their use of the secondary F0 cue increased. Gordon et al. (1993) concluded that stronger cues like VOT require more attentional resources, while weaker cues like F0 can be more relied upon when attention is limited. They suggested that under such distracting conditions, secondary cues might become more prominent to compensate the reduced role of primary cues.

Many attempts have also been made to explain the sources of the individual variability in cue-weighting by exploring the effect of cognitive control mechanism (executive functions, e.g., Francis & Nusbaum, 2002), listeners’ sensitivity to talkers (e.g., Yu, 2022) or listeners’ personality and autistic traits (e.g., Yu, 2010). For example, Yu (2010) showed that less autistic women adjusted less for phonetic coarticulation compared to men and women with more autistic traits. In his later study, Yu (2022) indicated that variations in how listeners weigh perceptual cues are influenced by listeners’ gender and their personal assessment of a speaker. Hutchinson (2022), for example, indicated that L2 learners who demonstrated higher levels of conscientiousness, including being more careful, attentive, and detail-oriented, tended to show less native-like patterns in their perception.

To summarize, the previous studies showed that the individual difference in cue-weighting tended to be systematic in a way that English listeners varied in the reliance of the secondary F0 cue in the perception of stops. The findings suggest that speech perception is not static but highly adaptable, depending on both environmental and individual factors. The insights from this research emphasize the importance of understanding cognitive mechanisms and personal characteristics in speech perception, which could potentially inform individual approaches in language learning and auditory processing interventions.

A subsequent question arises regarding L2 speech perception which is inherently more cognitively demanding than L1 speech perception due to its unfamiliarity. Extensive L2 research has explored the source of the variability ranging from learner, context and linguistic variables, aiming to identify more efficient learners and the environmental conditions that enhance L2 speech perception (e.g., Munro & Bohn, 2007). A recent study, Mora & Darcy (2023), showed the relationship between cognitive abilities and individual difference in L2 speech perception. They examined how attention control abilities measured by a separate task affected phonological processing of adult L2 learners, focusing on Spanish speakers learning English and English speakers learning Spanish. They showed that learners with more efficient attention-switching skills could discriminate L2 vowels more quickly, and the attention control was linked to better production of L2 vowels.

In addition to the learners’ cognitive functions, L2 learners are also affected by the environmental variables, and they tend to be more native-like perception pattern under less distracting contexts (e.g., Asano 2018; Lee, 2014). For example, Asano (2018) showed that under higher task demands with extended inter-stimuli-intervals (ISIs), the non-native listeners exhibited a decline in the perception, indicating that L2 learners were strongly affected by increased memory load and attentional challenges. Asano (2018) concluded that L2 learners’ ability to use acoustic dimensions diminishes when distracting, suggesting the difficulties that advanced learners often face in real-life listening environments with numerous distractions.

Closely relevant to the current study, using the dual-task paradigm in Gordon et al. (1993), Lee & Kong (2023) examined the attentional modulation on VOT and F0 cues in the perception of English voicing distinction by Korean-speaking learners of English. They examined how the relative importance of VOT and F0 was modulated by two different distracting conditions for the L2 learners. The results showed that while Korean learners’ sensitivity to the primary VOT cue decreased under the distracting condition, the reliance on the secondary F0 cue did not compensate the reduced role of VOT and the sensitivity to F0 was also reduced when distracted. The absence of compensating effect under distracting condition differed from English-speaking listeners’ cue-weighting pattern observed in Gordon et al. (1993). Lee & Kong (2023) suggested two possible explanations for the inconsistency. One is the reflection of L1 Korean’s cue-weighting pattern. Korean has three-way laryngeal contrast of stops which is distinguished both by VOT and F0 (e.g., Lee et al., 2020). The amount of attention to each of the two acoustic cues is almost equal in processing the stop contrast, that is, F0 is not secondary to compensate the reduced VOT. The other explanation is the classroom learning setting where Korean learners of English are normally exposed with English. That is, such non-naturalistic learning context has made Korean learners adverse to process L2 acoustic signals under distracting contexts.

Given that English voicing distinctions typically have a well-established prioritization between the two acoustic dimensions, VOT and F0, this discrepancy prompts further scrutiny of the learner group in Lee & Kong (2023) particularly with respect to potential variables such as varying English proficiency and non-controlled personality traits, which may have influenced the L2 learner group’s reduced sensitivity to the secondary F0 cue. Therefore, the current study examines the effect of attention on the cue-weighting of L2 English voicing contrast (/ta/ vs. /da/) for Korean-speaking prospective English teachers who majored in English Education.

This study specifically tested whether and how prospective English teachers, as a homogeneous learner group, exhibit differences in their reliance on VOT and F0 under varying distracting conditions, compared to L1 English listeners as examined by Gordon et al. (1993). By limiting the observation to future English teachers, we ensure a relatively homogeneous group in terms of educational exposure, which helps control potential factors such as proficiency, psychological motivation and personality traits. This allows us to focus more closely on the target variable—attentional load. The findings offer implications for understanding speech perception differences between classroom-taught learners and native speakers exposed to naturalistic environments.

This is a preliminary study, which is a part of a larger investigation. We will later compare perceptual patterns under attentional conditions between two distinct groups with differing traits in English proficiency, personality, and autistic characteristics, including communicative skills. For instance, we aim to compare the perceptual patterns of English education majors with those of science or engineering majors to observe how these groups respond to attentional load when using multiple acoustic cues. The results of this study could inform more tailored pedagogical approaches that account for the specific cognitive and perceptual challenges faced by different learner groups under varying attentional demands.

The current study specifically questions how Korean pre-service English teachers adjust their reliance on VOT and F0 when perceiving English stop voicing contrasts under distracting conditions. We discuss the results in reference with the results from native English speakers reported in Gordon et al. (1993). We subsequently test how the patterns related to their individual traits measured by Big5.

Twenty-six Korean-speaking university students (four males) participated in the perception experiment for a nominal fee. The participants’ ages ranged from 19 to 26, with a mean age of 21.6 (SD=2.06). To examine the correlation of individual L2 learners’ personality traits and attention with their speech perception patterns, this study limited participants to prospective English teachers at middle and high schools. All of the participants majored in English Education, and online recruitments were made among the English majors at Incheon National University and at Korea National University of Education. While speakers of Seoul/Gyeonggi, Chungcheong, and Jeolla dialects were included in the study, speakers of the Gyeongsang dialect were excluded to reduce the potential effect of tonal dialects. None of the participants reported any language disorders.

We conducted cloze test (Chung & Ahn, 2019) and Big5 to test the homogeneity of the present learner group. cloze test assessed participants’ English proficiency, asking them to fill in the blanks made to assess reading comprehension, grammar, and vocabulary of English. Big5 measures five major personality traits: Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness (e.g., Goldberg, 1993, 2001; John & Srivastava, 1999). In the Big5 survey, respondents rated their agreement with 44 statements on a Likert scale, such as “I am the life of the party.” The scores for each personality trait were calculated by averaging the responses for the statements associated with that specific trait, giving a comprehensive score for each personality dimension.

The auditory stimuli were adapted from Lee & Kong (2023). The stimuli were consonant-vowel (CV) syllables, forming a continuum from /ta/ to /da/, created using words produced by an English-speaking adult male speaker from Wisconsin, USA. A /da/ token was selected, and VOT and F0 were systematically varied. Seven log-scale VOT steps (9, 13, 19, 28, 40, 58, to 100 ms) were created by modifying the burst release/aspiration, while five F0 values (98, 106, 114, 122, to 130 Hz) were applied to the vowel portion. This resulted in 35 stimuli (7 VOT×5 F0 combinations). CV syllables were used to avoid lexical confounds and ensure consistency across tasks.

The speech perception experiment was developed and conducted using an online platform, Gorilla (gorilla.sc, Anwyl-Irvine et al., 2019). Each participant completed a set of tasks, including a language background questionnaire, the Big5 Personality Test, the cloze test, and a 2-alternative forced-choice (2AFC) dual paradigm task (with and without a math calculation prior to category decision), by clicking the experiment URL provided by Gorilla. A progress bar on the computer screen helped participants track their progress throughout the online experiment.

Specifically, the dual paradigm 2AFC task followed the experimental design of Gordon et al. (1993) and Kong & Lee (2018), focusing on attentional manipulation. Participants were asked to decide between the English stop voicing contrasts (/t/ vs. /d/) under two different attentional conditions: with a distractor (Distractor) and without a distractor (No-Distractor). In the No-Distractor condition, participants responded solely to the auditory stimuli. In the Distractor condition, participants performed a simple math calculation (e.g., “Given the three numbers 10, 20, and 30, is 20 minus 10 the same or different from 30 minus 20?”) while also completing a speech perception task (/d/ or /t/?). The auditory stimulus syllable was presented before the math question, and participants categorized the stop sound after completing the calculation. Each participant responded to 280 trials (35 auditory stimuli×2 distractor conditions×4 repetitions) on the Gorilla online platform.

Cloze test: The L2 learners’ responses to 40 questions were manually graded by the authors based on whether the responses were identical to the answer keys (2 points), comparable to them (1 point), or irrelevant (0 points) (Jun, 2018, among others). The graders independently scored the responses and later consulted with each other to resolve any discrepancies (about 10 cases). The L2 learners’ cloze test scores ranged from 21 to 65 out of a possible 80 points (mean=38.54, SD=13.18). Each learner’s score was used to represent their English proficiency in the statistical model.

Big5 survey: Figure 1 shows the partial correlation coefficients and p-values for every pair of Big5 sub-components, with the rest of the components controlled. At the 0.05 significance level, no significant correlations were found between the five sub-components, except for the relationship between Openness and Agreeableness. The prospective English teachers’ Openness and Agreeableness scores were positively correlated, controlling for Conscientiousness, Extraversion, and Neuroticism: r(26)=0.572, p<0.005. Given that the sub-components were relatively independent and represented distinct personal traits, we selected Agreeableness and Extraversion for this analysis to represent the personal traits of prospective English teachers. Participants’ scores for Extraversion ranged from 1.625 to 4.875 (mean=3.375, SD=0.674), and for Agreeableness from 2 to 4.778 (mean=3.765, SD=0.657).

Figure 1. The partial correlation coefficients and p-values of each Big5 sub-component pair with the rest of the components controlled.

Download Original Figure

2AFC with and without a distractor: A mixed-effects logistic regression model was constructed to predict stop category responses (/d/ and /t/) based on acoustic parameters in both distracted and non-distracted listening conditions (using the lme4 and lmerTest packages in R via RStudio (Racine, 2012, Version 2024.04.1+748): Bates et al., 2015; Kuznetsova et al., 2017). Fixed effects included VOT and F0 in normalized units [using scale() in R] and the presence of distracting math tasks (DistCondition: Distractor vs. No-Distractor). Random effects included by-subject intercepts and slopes for VOT and F0 at the speaker level and at the speaker-by-distractor level. The fixed effect coefficients of VOT and F0 in the Distractor and No-Distractor conditions indicate group-averaged weights for VOT and F0 with and without the distractor. The random effect coefficients at the speaker-by-distractor level estimate individual speakers’ coefficients as deviations from the group averages in both conditions.

To investigate the effect of the distractor on individual prospective English teachers’ stop perception, we ran a partial correlation test between VOT and F0 coefficients from the mixed-effects model, controlling for the listeners’ personal traits (Agreeableness and Extraversion) and English proficiency (the cloze test scores). The ppcor package was used for the analysis {Kim, 2015: pcor(data, method=c(“pearson”))}.

Table 1 presents the beta coefficients of the fixed-effect variables estimated from the mixed-effects logistic regression model using a treatment coding scheme. Both the two acoustic variables, VOT and F0, were statistically meaningful factors in L2 learners’ identification of /t/ over /d/ without a distractor (β_VOT=3.79, SE=.41, p<0.0001; β_F0= 1.65, SE=0.13, p<0.0001), and the VOT coefficient was greater than F0 coefficient suggesting that the Korean L2 listeners, as a group, put more perceptual weight on VOT than F0 as English native listeners did (e.g., Abramson & Lisker, 1985). Importantly, both acoustic dimensions were significantly affected by the distractor such that the interaction terms of VOT and F0 with Distractor were significant (β_{VOT×Distractor}=–1.29, SE=0.33, p<0.0005; β_{F0×Distractor}=–0.36, SE=0.11, p<0.005). The coefficients of the interaction terms suggest that not only VOT but F0 were used less when distracted than not. Overall, the L2 learner group weighted VOT over F0 in perceiving the L2 English voiced-voiceless contrast, and the distractor affected them in using both primary and secondary cues.

Figure 2 plots the logistic curves based on the beta coefficients of Table 1. In both Distractor and No-Distractor conditions, the VOT curves (dashed line) are steeper than the F0 curves (solid line), showing that the listeners were more sensitive to VOT than to F0 in identifying /t/ over /d/. In addition, both VOT and F0 curves were less steep in the Distractor condition (black lines) than those in the No-Distractor condition.

Figure 2. Logistic curves of the fixed effect variables (VOT: dashed and F0: solid) based on the mixed-effects regression model coefficients. The curves are color-coded by the listening conditions: Distractor in black and No-Distractor in red.

Download Original Figure

The left panel of Figure 3 plot the individual listeners’ VOT coefficients as a function of their F0 coefficients, illustrating how the individual coefficients changed from No-Distractor condition (circles) to Distractor condition (squares). With or without distractor, most VOT coefficients were greater than F0 coefficients, located under the diagonal line. This suggests that most L2 learners used VOT primarily for the English /d/-/t/ distinction. There were only four datapoints above the line who relied on F0 primarily for the contrast. In terms of the effect of distractor, the arrows connecting each individual learner’s coefficients of No-Distractor condition with those of Distractor condition tend to head toward the lower and left corner of the panel. This indicates that VOT and F0 were used less when distracted than not, in general. There were a handful of listeners whose arrows head for different directions deviant from this general tendency. That is, a single listener used F0 more with its VOT coefficient decreased when distracted than not (a red square, an arrow pointing to an upper left corner of the panel). Another six listeners used VOT more with their F0 coefficients decreased when distracted than not (six blue squares, arrows pointing to a lower right corner of the panel).

Figure 3. Individual L2 learners’ F0 coefficients as a function of their VOT coefficients in the No-Distractor (grey circles) and the Distractor conditions (squares). Arrows connect the individuals’ coefficients of the two listening conditions: from No-Distractor to Distractor.

Download Original Figure

While the L2 learners did not show a reconciliation of a secondary cue under distractor except one participant (red square) because both cues VOT and F0 were reduced, they were not equally affected by the distractor. In Figure 3, the right panel repeats the coefficient distributions of the two distractor conditions, highlighting six datapoints in green whose F0 coefficients were greater than VOT coefficients when distracted, although their F0 coefficients were not greater than VOT coefficients without distractor. In other words, distracting listening condition affected these six listeners using F0 as a primary cue even though they primarily relied on VOT without distractor. They may also support Gordon et al. (1993) in the L2 perception in a way that the role of a secondary cue was boosted in a relative sense under distractor.

Finally, the partial correlation test was conducted to estimate the correlation coefficient between acoustic variables given personal traits and L2 proficiency. The two test variables of the test were individual listeners’ VOT and F0 coefficient differences between Distractor and No-Distractor conditions. Two listener-related variables were set as control variables of the test: (1) Individual learners’ mean scores of Big5 survey, and (2) their mean scores of the cloze test. The test result showed that VOT coefficient differences was significantly correlated with F0 coefficient differences given their personal traits and English proficiency: r(26)=–0.431, p<0.05. The negative correlation coefficients suggest that greater VOT differences were associated with smaller F0 differences (see Figure 4). The correlation test result supports that the L2 learners who lost their attention to VOT tended to lose less attention to F0 when distracted. That is, the role of a secondary cue, F0 for the voicing contrast in English, enhanced or maintained to reconcile the damaged role of a primary cue, VOT.

Figure 4. Individual L2 learners’ F0 coefficient difference (Distractor – No-Distractor) as a function of their VOT coefficient difference.

Download Original Figure