My main research interest lies in speech processing and production. More specifically I am interested in prosody processing, L1 & L2 speech production and perception, phonetics and articulation. I am a graduate student in the department of Linguistics at KU and work with Dr. Annie Tremblay in Second Language Processing and Eye-Tracking (L2 PET) lab. My current research focuses on the influence of prosodic information on online spoken word recognition. I speak Seoul Korean with some jargons that I mostly picked up from my mom. In my free time, I love reading books•cartoons, watching clay (or puppet) animations, and making fun of my cat /ᐠ.ꞈ.ᐟ\.




KU-ARTLEX: A single-speaker EMA database for modeling the articulatory structure of the lexicon

Charles Redmon, Seulgi Shin & Panying Rong

Download paper

Articulatory data was recorded on a 26,793-word database of English words - replicating the set used in the Massive Auditory Lexical Decision project (Tucker et al., 2018) - along with two repetitions of 1,200 controlled CVC syllables based on the California Syllable Test (Woods et al., 2010), from a single male native speaker of Midwestern American English. The primary aim of this database is to serve as a window on the articulatory structure of the lexicon. Further, using this open-access database, comparable perception experiments testing the predictability of word recognition patterns from articulatory and acoutic profiles can now be run by multiple research groups, providing greater clarity to broader theoretical debates on the nature of the encoding of the speech signal.

Context-dependent hyperarticulation of the Korean three-way laryngeal stop contrast to clear speech

Seulgi Shin & Allard Jongman

Download poster

We investigate whether and how speakers' adaptation to feedback is reflected in hyperarticulation by comparing the production of Korean aspirated, lenis, and fortis stops in plain versus clear speech. Ten Seoul Korean speakers were asked to read a stop-initial nonword and repeat the nonword after receiving misrecognition feedback that conatined a contrasting stop. f0 and VOT were examined as main cues to the Korean stop distinction. Results showed that VOT was lengthened for aspirated and lenis stop in clear compared to plain speech. This difference in VOT was attributable to speakers' modification in response to different types of feedback. VOT for aspirated stops was lengthened after both fortis and lenis feedback whereas VOT for lenis stops was lengthened only after fortis feedback. f0 did not show difference between plain and clear speech. However, f0 enhancement for lenis and fortis stops depended on the type of feedback. f0 for lenis stops was lowered after aspirated feedback and raised after fortis feedback. Fortis stops' f0 was lowered after aspirated and raised after lenis feedback. Overall, speakers can make local modifications in VOT and f0 to response to specific feedback, suggesting precise control in their effort to distinguish sounds and maintain intelligibility. We plan to collect more data and expand the findings to Kyunsang Korean.

Use of tonal information in Korean lexical access

Annie Tremblay, Seulgi Shin, Taehong Cho, & Sahyang Kim

Download paper

Prominence in Seoul Korean is realized at the level of the Accentual Phrase (AP), with the AP-final high (H) tone signaling word-final boundaries and the AP-initial Low (L) tone signaling word-initial boundaries. Using word-spotting experiments, Kim and Cho (2009) showed that Korean speech segmentation benefits from both the AP-final H and AP-initial L tones, but it is unclear whether (and if so, how) tonal information also contrains lexical access in Korean. The present study investigates this issue using a visual-world eye-tracking experiment. Native Korean listeners hear sentences containing a temporary lexical ambiguity between a disyllabic target word in AP-initial position (e.g., [saesinbu-ga]AP [masul-eul]AP 'the-new-bride-subj magic-obj') and a disyllabic competitor word spanning the AP boundary (e.g., gama 'palanquin'). The auditory stimuli were resynthesized to create four tonal boundary conditions: H#L, H#H, L#L, and L#H, where # represents an AP boundary. Listeners' eye movements to the printed target and competitor words were monitored as they heard the auditory stimuli. The results showed independent effects of the AP-initial and AP-final tones on lexical access, suggesting that the post-lexical intonational system of Korean modulates lexical activation.

Effect of prosodic context on lexical access: An investigation of Korean denasalization

Seulgi Shin & Annie Tremblay

Download paper

The present study investigates whether prosodic context modulates listeners’ interpretation of prosodically driven variations in lexical access by examining the recognition of words whose initial nasal consonant is denasalized in Korean. Nasals in Korean undergo nasal weakening in phrase-initial prosodic positions such as at the beginning of the Accentual phrase (AP). Because of this weakening, denasalized nasals share acoustic characteristics with both nasals and plosives. If Korean listeners exploit prosodic context in lexical access, they should activate nasal-initial words more when hearing denasalized nasals in AP-initial position than when hearing them in AP-medial position, and they should activate plosive-initial words more when hearing denasalized nasals in AP-medial position than when hearing them in AP-initial position. Participants completed two cross-modal priming experiments with lexical decision. The targets were visually presented Korean words beginning with a nasal (e.g., noru ‘roe deer’) (Exp. 1) or plosives (e.g., toru ‘stealing a base’) (Exp. 2). The experimental primes were auditory words that rhymed with the target but began with a denasalized nasal (e.g., /noru/); the control primes were auditory words that were phonologically and semantically unrelated to the target (e.g., /tʃotɛ/ ‘invitation’). The primes were heard in either AP-initial or AP-medial position. In Exp. 1, listeners showed sensitivity to prosodic context such that denasalized primes facilitated lexical decisions in relation to the control condition in AP-initial position but not in AP-medial position. Exp. 2 showed no effect of prosodic context or priming condition, indicating that listeners interpreted denasalized nasals differently from plosives in lexical access regardless of prosodic context (unlike previous offline studies). These results suggest that listeners take prosodic context into account and are sensitive to prosodically driven phonetic details in lexical access.

No adaptation to forign-accented voice-onset-time distribution in lexical processing

Tifani Biro, Seulgi Shin, Yuyu Zeng & Annie Tremblay

Download poster

This study investigates whether foreign-accent adaptation is less limited when the mapping between the accented and underlying words does not require higher-level inferencing. English listeners completed an eye-tracking experiment in which they heard a (female) French talker and a (male) English talker. Target and competitor words began with a voiced or voiceless stop and were otherwise temporarily ambiguous (e.g., timber and dimple). All stops were resynthesized: The French talker's stops were prevoiced (voiced) and short-lag (voiceless) and the English talker's stops were short-lag (voiced) and long-lag (voiceless). Preliminary results suggest an effect of voicing only for the French talker, with voiceless-stop targets generating more competition than voiced-stop targets. Importantly, this effect decreased only slightly from the first to the second half of the experiment, suggesting that there was no foreign-accent adaptation.

Effects of native language on the use of segmental and suprasegmental cues to stress in English word recognition: An eye-tracking study

Katrina Connell, Simone Hüls, Maria Teresa Martínez-García, Zhen Quin, Seulgi Shin, Hanbo Yan & Annie Tremblay

Download paper

This study investigate whether the presence of lexical stress in the native language (L1) determines second-language (L2) learners' ability to use stress in L2 lexical access. It focuses on (standard Mandarin) Chinese and (Seoul) Korean listeners' (and native English listeners') use of segmental and suprasegmental cues to stress in English word recognition. Stress placement in English is signaled by segmental (vowel reduction) and suprasegmental (fundamental frequency, duration, and intensity) cues. Chienese has full-full and full-reduced words that differ in stress placement, with segmental and suprasegmental cues signaling stress. By contrast, Korean does not have lexical stress. Participants completed an eye-tracking experiment. They heard stimuli containing a target word with initial stress (parrot), and saw four orthographic words in the display, including the target and one of two competitors (stress match: parish ; stress mismatch: parade). The first syllable of the target and stress-mismatch competitor differed in both segmental and suprasegmental information (parrot=parade) or only in suprasegmental information (mystic-mistake). Growth-curve analyses on fixations revealed that Chinese listeners used stress to recognize English words. Korean listeners did not use stress in the recognition of English words, but made use of stress in the presence of vowel reduction, confirming L1 effects on the use of stress in L2 lexical access. Furthermore, English speakers used stress regardless of the presence or the absence of vowel reduction, but showed greater use of stress in the presence of vowel reduction.

What is special about prosodic strengthening in Korean: Evidence in lingual movement in V#V and V#CV

Seulgi Shin, Sahyang Kim & Taehong Cho

Download paper

The present study investigates the effects of prosodic boundary and prominence (focus) on the /a/-to-/i/ tongue movement in Korean in two contexts: V#V and V#/m/V. The Electromagnetic Midsagittal Articulography (EMA) experiment was conducted. Results show that the tongue movement at an Intonational Phrase (IP) boundary is larger, longer and faster. The observed boundary-induced strengthening pattern in Korean is clearly different from that in English where the pie-gesture (Byrd, 2000; Byrd & Saltzman, 1998, 2003; Saltzman, 1995) is assumed to influence temporal realization of articulatory gestures at a prosodic juncture, slowing down the gestural movement. Korean, a language without constraints from the lexical stress system, has more freedom to strengthen articulation at prosodic junctures, creating strengthening patterns which are often encountered with prominence marking in English. Prominence effects show a relatively weaker but comparable pattern to the boundary effect, showing a larger, longer, and faster movement. The effect of boundary closely interacted with the effect of prominence in that the prominence effect was only found at a word boundary, suggesting the boundary effect largely took precedence over the prominence effect in Korean. Results also reveal that the presence of a consonant influences transboundary vocalic movement, and that the consonantal influence is further modulated by boundary strength. Those results show that the effects of boundary and prominence in Korean is systematically realized, reflecting the language-specific prosodic system and have further implications for the pi-gesture model.