CROSS-LINGUISTIC INFLUENCE IN L1 PHONETIC CATEGORIES IN KOREAN HERITAGE SPEAKERS AND LONG-TERM IMMIGRANTS

Upon acquiring or learning another language, cross-linguistic influence (CLI) is an inevitable phenomenon with which a bilingual speaker lives. One key aspect of CLI is its bidirectionality, flowing between both the first (L1) and second languages (L2) mutually affecting each other. However, investigations of L1 CLI on L2 have dominated previous bilingual studies, and despite the increasing amount of research on L2 CLI on L1, the phonetic and phonological domains remain relatively underexplored. The primary goal of the present study is to expand our understanding of the underlying mechanisms governing L2 CLI on L1 phonetics and phonology.

The present study investigates L2 CLI on L1 phonetics and phonology by examining both the speech perception and production of L1 sound categories among two different groups of bilinguals, Korean heritage speakers (HSs, n = 30) and long-term immigrants (LTIs, n = 27) group participants in the US, in comparison to L1(Korean)-immersed (L1-i) native speakers residing in South Korea (n = 30). Participants completed a series of three experimental tasks: (1) a three-alternative forced-choice (3AFC) identification task, (2) an AX discrimination task, and (3) a controlled reading paradigm task.

Experiment 1 (3AFC task) was conducted to investigate the extent and direction of L2 CLI in perceptual cue weighting to L1 speech categories. In this task, participants listened to a Korean word in each trial, potentially differing in the word-initial stop, and decided which word they heard from a real-word Korean minimal triplet /pul/ ‘fire,’ /p^hul/ ‘grass,’ and /p^*ul/ ‘horn.’ Specifically, the word-initial stop consisted of an eight-by-eight orthogonal voice onset time (VOT)–onset f0 continuum, created through a speech resynthesis technique. Based on the similarities and differences in the use of the two acoustic parameters between Korean (either onset f0 or VOT is a primary cue) and English stops (VOT is the primary cue), bilingual participants were expected to exhibit different cue-weighting patterns, as compared to L1-i speakers. The results from the mixed-effects logistic regression model analyses indicated that while HSs were less sensitive to the Korean primary cue, onset f0, compared to L1-i speakers—suggesting assimilation to L2 in the perceptual domain—LTIs exhibited greater sensitivity to this cue, indicating dissimilation from L2. It was also found that bilingual participants’ Korean dominance significantly influenced their cue weighting in the perception of Korean stops.

Experiment 2 (AX discrimination task) was administered to assess participants’ perceptual accuracy for L1 stop categories and the potential impact of L1 cue weighting, as estimated in Experiment 1, on their discrimination performance. Notably, the VOT in the stop stimuli used in the AX task were resynthesized to have a consistent VOT of 70 ms across all stimuli. This setup created a condition where participants had no choice but to rely solely on the onset f0 cue—the primary cue to the Korean lenis-aspirated stop contrast, rendering VOT, the primary cue for the voicing contrast in English stops, uninformative. The results from mixed-effects logistic regression models showed that HSs were significantly less accurate in discriminating their L1 stop categories without the VOT cue, while LTIs outperformed the L1-i speakers. That is, the LTI group, the most balanced group in terms of language dominance, had the highest accuracy in discriminating L1 contrasts among the participant groups. Furthermore, individual sensitivity to the onset f0 cue was found to be positively correlated with discrimination performance.

Experiment 3 (Controlled reading paradigm) aimed to examine L2 CLI on the implementation of acoustic parameters for L1 Korean stops, as well as the potential impact of proficiency and dominance on these parameters. Participants read aloud a list of minimal triplet stimuli differing in the word-initial stop within a carrier phrase. A machine-learning-based audio signal detection system was used to analyze the acoustic parameters, and Bayesian mixed-effects linear regression models, along with quadratic polynomial regression models, were implemented for statistical analysis of the processed data. The results of the production task mirrored the perception task (Experiment 1): HSs demonstrated assimilation to L2 via onset f0, while LTIs showed dissimilation, as compared to L1-i speakers. The analysis also revealed that the degree of bilingual balance in dominance and proficiency significantly influenced the implementation of onset f0, with more balanced bilinguals exhibiting greater category contrasts than less balanced bilinguals, regardless of whether they were Korean-dominant or English-dominant.

The findings from these experiments provide concrete evidence of L2 CLI in L1 phonetics and phonology. Importantly, the results demonstrate that not only the timing of L2 acquisition and the quantity and quality of L2 input but also the quality and quantity of L1 acquisition and bilingual balance contribute to the direction and the degree of L2 CLI in L1 speech. These findings align with the predictions of the revised Speech Learning Model (SLM-r, Flege & Bohn, 2021) and expand its scope of application to include both HSs and LTIs. In particular, the evidence of category assimilation and dissimilation lends support to the bidirectional CLI hypothesis proposed by SLM-r. To conclude, the present dissertation expands our understanding of the nature of L2 CLI in L1 phonetics and phonology in bilingual speakers.