Perceptual similarity is not all: online perception of English coda stops by Korean listeners

1 Introduction

When words are incorporated into one language from another, they often go through adaptation to comply with the phonological structure of the borrowing language; yet, some loanword patterns are not easily explained by the grammar of the receptor language. One of those patterns involves unnecessary vowel insertion, where vowel insertion appears unnecessary since the original sound of the source language would be permissible in the target language (Kang 2006; Peperkamp et al. 2008; Shinohara 1997). The present study investigates vowel insertion following word-final stops in English words borrowed into Korean. Vowel insertion in this environment is seemingly unmotivated because Korean native words can end in a stop and thereby an English word ending in a stop would be legitimate in the native language.

This study attempts to account for this vowel insertion by means of the adaptation-in-production approach. This view assumes that loanword adapters store the surface form of the source language and the production grammar performs the adaptation to the native phonology (Fleischhacker 2005; Jacobs and Gussenhoven 2000; Kang 2003; Kang et al. 2008; Kawahara 2006; Kenstowicz 2003, 2007; LaCharité and Paradis 2005; Miao 2006; Paradis and LaCharité 1997, 2008; Paradis and Tremblay 2009; Shinohara 2006; Steriade 2001, 2008; Yip 2002, 2006; among others). That is, the phonetic form of the original structure is faithfully taken as the abstract underlying representation and loanword adaptations are then transformations produced by the phonological process in production.

To explain why even accurately perceived forms are sometimes transformed, some researchers have appealed to perceptual factors and subphonemic details in explaining adaptation patterns using the production grammar (Fleischhacker 2005; Kang 2003; Kang et al. 2008; Kawahara 2006; Kenstowicz 2003; Miao 2006; Shinohara 2006; Steriade 2001; Yip 2002, 2006). On the perceptual similarity approach, originally proposed by Steriade (2001), speakers possess knowledge of perceptual similarity (P-map) between foreign and native sounds, and perceptual factors are incorporated into grammatical constraints that can be ranked with respect to other grammatical constraints. For example, the perceptual similarity approach argues that Korean speakers insert a vowel following an English final released stop to maintain perceptual similarity between the English form and the Korean adaptation since a stop plus a vowel is the perceptually closest Korean structure to the English stop release. That is, this hypothesis assumes that loanword adaptation is conducted by sophisticated adapters who have the ability to accurately perceive foreign sounds and choose the closest native language structure by means of a P-map which exists as a component of their grammar (Steriade 2001).

Different factors have been proposed to increase the possibility of vowel insertion in this context (Boersma and Hamann 2009; de Jong and Cho 2012; Jun 2002; Kang 2003; Kim 2018, 2021; Kwon 2017; Rhee and Choi 2001). The current study examines the effects of six different factors including release, voicing and place of coda stop, tenseness of the vowel preceding coda stop, final stress, and word length. First, the release of the English final stop favors vowel insertion in that stop release in English is acoustically similar to the vowel inserted after an English final stop in Korean (Kang 2003); this is because English word-final stops are variably released while Korean counterparts must be unreleased (Byrd 1992; Chung 1986; Crystal and House 1988; Gimson 1989; Huh 1965; Kim 1971). Second, Kang (2003) argues that stop voicing is relevant since vowel insertion after a voiced final stop maintains perceptual similarity between the English and Korean forms by placing the voiced stop in a position where voicing is legal in Korean (i.e., between sonorants). Third, vowel tenseness can also lead to the tendency to insert a final vowel after an English word with a tense pre-final vowel to create an open syllable before the word-final stop; this is because a Korean vowel tends to be longer in an open syllable than in a closed syllable (Chung and Huckvale 2001; Han 1964; Koo 1998). In addition to the factors that involve acoustic characteristics directly affecting the perception of English final stops by Korean listeners, final stop place and final syllable containing stress have also been found to affect the vowel insertion. Namely, vowel insertion is more likely when the final stop is coronal than when it is labial/dorsal; vowel insertion is also more likely when the final syllable is stressed or when it is unstressed (Jun 2002; Kang 2003). Last, word length can affect the likelihood of vowel insertion by Korean speakers adapting English words; vowel insertion is more likely in monosyllabic than in polysyllabic borrowed words (Kang 2003; Rhee and Choi 2001).

In this study, a similarity judgment task is designed to test the effects of each linguistic factor and to investigate whether an English final stop sounds similar to a stop followed by a vowel to Korean participants when specific phonetic characteristics are present. This task is directly connected to conscious judgments of perceptual similarity between native and foreign forms by allowing participants to compare English versus Korean nonce forms. Table 1 shows the predictions of the perceptual similarity approach that assumes accurate perception. Under this view, even when Korean listeners accurately perceive the English form as consonant-final, they would insert a vowel in their production to maintain perceptual similarity to an English final released or voiced stop. This predicts that listeners will judge an English CVC form ending in a released or voiced stop as more similar to Korean CVCV than Korean CVC. Similarly, it is predicted that an English CVC form preceded by a tense vowel will be judged by Korean listeners as more similar to Korean CVCV than CVC since vowels are longer in open syllables in Korean and inserting a vowel after a form with a tense vowel would maintain the perceptual similarity in vowel length. In addition, the perceptual similarity approach predicts that effects of stop place, final stress and word length will cause Korean listeners to judge an English CVC as more similar to Korean CVCV because these factors are connected in English to the greater likelihood of final stop release, as pointed out by Kang (2003). Concerning stop place effect, for instance, vowel insertion is more likely after a dorsal than a labial final stop since Korean speakers are more likely to hear a released pronunciation of a dorsal final stop than that of a labial final stop (see Table 1).

2 Method

3 Results

The responses of the Korean participants were fitted into a mixed-effects logistic regression model using the glmer function implemented in the lme4 package (Bates et al. 2015) in R (R Core Team 2021). The model was built for the six factors (stop release, stop voicing, vowel tenseness, stop place, final stress, and word size). For the model, the dependent variable was the participants’ answers (whether participants’ response was Korean consonant-final or Korean vowel-final), and it was coded as 0 for responses of ‘English word judged as similar to Korean consonant-final word’ and 1 for responses of ‘English word judged as similar to Korean vowel-final word’.

Fixed effects included six factors, Release (Unreleased or Released), Voicing (Voiceless or Voiced), Tenseness (Lax or Tense), Place (Labial or Coronal or Dorsal), Stress (Unstressed or Stressed), and Length (Monosyllabic or Disyllabic or Trisyllabic). All factors were deviation-coded except Place and Length, which were forward difference coded, i.e., Release: Unreleased = −0.5, Released = 0.5; Voicing: Voiceless = −0.5, Voiced = 0.5; Tenseness: Lax = −0.5, Tense = 0.5; Place 1: Labial = 0.6, Coronal = −0.3, Dorsal = −0.3; Place 2: Labial = 0.3, Coronal = 0.3, Dorsal = −0.6; Stress: Unstressed = −0.5, Stressed = 0.5; Length 1: Monosyllabic = 0.6, Disyllabic = −0.3, Trisyllabic = −0.3; Length 2: Monosyllabic = 0.3, Disyllabic = 0.3, Trisyllabic = −0.6. A model started with all the six predictors and backward elimination was conducted through pairwise model comparisons using ANOVA tests based on p-value. Three predictors (Voicing, Tenseness, and Stress) were removed from the process and two-way interactions were taken into consideration for the remaining predictors (Release, Place, and Length). After the elimination of the non-significant predictors, by-participant random slopes for Release and Place and by-item random intercepts were added to the model as they improved the model. Finally, the model in Table 4 in the Appendix A was selected as the best-fitted model.

The regression model for the similarity judgment task (see Table 4) found a significant main effect of Release (z = 5.276, p < 0.001). This result suggests that Korean participants were more likely to judge an English final stop as similar to a stop followed by a vowel when the final stop was released than when it was unreleased, which is consistent with the prediction of the perceptual similarity approach (see Figure 1). In the summary of fixed effect coefficients in the best-fitted model, the fitted logit value for the effect of Release is 1.852, which corresponds to a fitted proportion of participants’ choosing a vowel-final response of 0.864. That is, this model predicts that 86.4% of participants’ responses are responses of ‘English word judged as similar to Korean vowel-final word’ when Release is Released, Place is Labial, and Length is Monosyllabic (see Table 4). In the regression model, the main effect of Place 1 (Labial vs. Coronal/Dorsal) was also significant (z = −3.391, p < 0.001), indicating that participants were more likely to judge an English final stop as similar to a Korean vowel-final word when the stop was coronal or dorsal than when it was labial; yet, the main effect of Place 2 (Labial/Coronal vs. Dorsal) was not significant (p > 0.1). Thus, although dorsal final stops were more likely to be judged as similar to Korean vowel-final than labial final stops, the difference between coronal versus dorsal final stops was not statistically significant (see Figure 2). This result is consistent with the prediction of the perceptual similarity approach since the approach predicted a significant difference of labial versus dorsal stops in the similarity judgment task. In addition, the best-fitted model found a significant main effect of Length 1 (Monosyllabic vs. Disyllabic/Trisyllabic), indicating that participants were more likely to judge an English final stop as similar to a Korean vowel-final when the English form was monosyllabic than when it was disyllabic or trisyllabic (z = 2.972, p < 0.01). However, the main effect of Length 2 (Monosyllabic/Disyllabic vs. Trisyllabic) was not significant (p > 0.1), which suggests that although monosyllabic items significantly induced more vowel-final responses than disyllabic items, a significant difference was not found between disyllabic versus trisyllabic items (see Figure 3). This result is compatible with the prediction of the perceptual similarity view because the view predicted a significant difference between monosyllabic versus polysyllabic words in the similarity judgment.

Figure 1:

Predicted probability of English word judged as similar to Korean vowel-final word depending on coda stop release; error bars indicate standard error.

Figure 2:

Predicted probability of English word judged as similar to Korean vowel-final word depending on coda stop place; error bars indicate standard error.

Figure 3:

Predicted probability of English word judged as similar to Korean vowel-final word depending on word length; error bars indicate standard error.

4 Discussion

We have looked at similarity judgment choices in terms of each linguistic factor according to the predictions of the perceptual similarity approach. This approach predicted that Korean listeners would choose Korean CVCV as more similar to English CVC than Korean CVC. Although participants accurately perceive English CVC, Korean CVCV would be judged as more similar to English CVC than Korean CVC when the English final stop is released or voiced and when it is preceded by a tense vowel. In addition, factors of stop place, final stress and word length would lead to an increase in perceptual similarity to Korean CVCV since they correlate with the likelihood of release in English. Although this hypothesis predicted significant effects of all the given factors, only three factors including release and place of coda stop as well as word length were found to play a role in the similarity judgment task.

In the current study, we have seen that there was a significant effect of final stop release, indicating that a greater likelihood of vowel-final responses was significantly more likely after final released than unreleased stops. Results of this study showed that items ending in unreleased stops showed relatively lower inaccuracy than the overall inaccuracy in similarity judgments (20 vs. 35%); participants had vowel-final responses in 20% of the forms that ended in an unreleased stop. This result confirms that final stop release is indeed a major factor accounting for Korean speakers’ vowel epenthesis following English final stops. As argued in previous studies, the release factor is important in Korean loan phonology based on different phonetic characteristics of English and Korean (Kang 2003; Kim 2021; Kwon 2017). That is, English coda stops are variably released whereas Korean counterparts are always released. Also, Korean high vowels tend to be devoiced when they follow aspirated stops and thus an English word-final released stop may be acoustically close to a Korean devoiced vowel (Jun and Beckman 1994; Song 2002).

In this task, there was also a significant effect of coda stop place, suggesting that Korean participants were more likely to judge an English word ending in a stop as similar to Korean vowel-final word when the coda stop was dorsal than when it was labial. Although there was no significant difference between coronal versus dorsal final stops, results are consistent with the perceptual similarity hypothesis that predicted a significant difference of labial versus dorsal stops. The dichotomy between labial versus dorsal coda stops shown in participants’ perception nicely matches with the pattern in the average duration of coda stop release (see Table 2). Kang (2003) claims that the more likely a final stop is to be released by English speakers, the more likely it is to undergo vowel insertion by Korean speakers, and that vowel insertion is more likely after a dorsal than a labial final stop because Korean speakers are more likely to hear a released pronunciation of a dorsal final stop than that of a labial final stop. Thus, there is nothing about dorsality itself that can contribute to vowel insertion; the only reason labial versus dorsal stops matters is that it affects the likelihood of release in English pronunciation. Although participants in this task did not hear naturalistic spoken English and they listened to stimuli where stop release was balanced across each category of stop place, coda stop release in the current stimuli was controlled in terms of its presence/absence and it can still vary in its duration. This pattern in release duration can affect listeners’ perception in that a longer release can be perceptually more similar to a stop followed by a vowel.

In addition, this study found a significant effect of word length, indicating that Korean listeners were more likely to judge an English word ending in a stop as similar to Korean vowel-final word when the English word was monosyllabic than when it was disyllabic. Although disyllabic words failed to significantly induce more vowel-final responses than trisyllabic words, the significant difference between monosyllabic versus disyllabic words was predicted by the perceptual similarity hypothesis; it is worth addressing the question of why monosyllabic words had more frequent vowel-final responses. Word length effect has previously been argued to be due to a word size preference in that Korean prefers disyllabic prosodic word structure (Hirano 1994a, 1994b). An effect of Japanese English-to-Korean adaptation was also possibly one reason for Koreans’ tendency toward vowel epenthesis in English monosyllables since many English monosyllabic words were adapted into Japanese as disyllables with open final syllables and Korean borrowed these loanwords from Japanese (Hirano 1994b; Kang et al. 2008). In this respect, the word length effect does not appear to follow from the perceptual similarity approach. However, if listeners are biased toward hearing structures that are legal in their native language, the dispreference for monosyllabic words in Korean could bias listeners toward hearing an extra syllable in monosyllabic forms. For example, Peperkamp and Dupoux (2003) claim that the phonological constraints of a language affect perception; that is, Japanese speakers heard an extra vowel in forms like ebzo because ebzo would not be legitimate in the language. Similarly, it is entirely possible that the Korean participants in the current study heard an extra syllable in monosyllabic forms because monosyllabic words are not preferred in Korean.

The results of this study concerning the main effect of release predictor are compatible with the hypothesis of the perceptual similarity approach; namely, the similarity judgment task provided evidence that is accounted for by the perceptual similarity theory in terms of stop release. Results also showed the significant effects of other factors associated with coda stop release that were predicted to lead to more frequent vowel-final responses by the perceptual similarity view. However, it is surprising that stop voicing and vowel tenseness were not found significant in this study since these factors have been previously claimed to involve acoustic cues that can directly affect second language perception (Kim 2021; Kwon 2017). This result supports the need for further study that can focus more on direct perception rather than judgments of perceptual similarity, asking listeners to hear foreign sounds alone. The findings of this study suggest that the phenomenon of unnecessary repair in Korean loanwords cannot be attributed only to maintaining perceptual similarity in production and that other possible factors can play a role in loanword adaptation such as misperception, orthography, and explicit conventions of the Korean Academy. Unnecessary vowel epenthesis appears to be a complex interplay between production, perception and various other linguistic elements.