The two glides w and j exist in both English and Korean. In English, these two glides form syllable-initial glide-vowel (GV) sequences with any of the following twelve vowels (i, ɪ, eɪ, ɛ, ʌ, ə, œ, u, ʊ, oʊ, ɔ, a). In Korean, assuming seven monophthongs (i, e, ə, ɨ, u, o, a), fourteen GV sequences are logically possible, but only nine occur; the following five GV sequences are absent: *ji, *jɨ, *wu, *wɨ, *wo. Researchers who have proposed phonological explanations for this gap unanimously point to the homorganicity between the two segments in these absent sequences. In English, however, homorganicity seems to be disregarded; five GV sequences--GV[HO] sequences--consist of homorganic segments: wu, ji, wʊ, jɪ, wo. This difference in phonotactics between the two languages constitutes the source of difficulty for Korean ESL learners in mastering the L2 glides and GV[HO] sequences.
In this study, I first provide detailed phonological and phonetic characterizations of glides. I review phonological representations of glides, as well as corresponding high vowels. Then, I perform a series of acoustic analyses of a set of production data collected from Korean and English monolingual speakers. The acoustic parameters under analysis include the first three formants (F1-F3) and the duration of the glide steady state and the glide-to-vowel transition. These analyses reveal that the F2 of English [w] is consistently lower than that of any of the twelve vowels, while the F2 of Korean [w] depends significantly on the quality of the following vowel. Also, English glides exhibit considerably longer steady state durations compared to Korean glides.
Next, I analyze the learners' production data, collected from twenty-two Korean ESL learners. The L2 data reveal that the learners resorted to a few major repair strategies for target GV[HO] sequences, while the vast majority of the non-homorganic GV sequences (GV[N-HO]) are produced target-appropriately. Among these repair strategies, 79% were glide deletion (wound → [und]/[ʔund]), 20% vowel shift (wound → [wənd]), and 1% glide shift (yip → [wɪp]). Interestingly, however, in their L2 glides, many of the learners showed a departure from monolingual Korean glides in the F2 of [w] and the duration of the steady state.
Lastly, an Optimality Theoretic account is proposed for the learners' L2 data. Under the assumption that GV[HO] sequences are marked relative to GV[N-HO] sequences (Kawasaki 1982), I argue that learning English GV[HO] sequences by Korean ESL learners involves constraint reranking, crucially, demotion of a set of markedness constraints below a set of competing faithfulness constraints.