Phonological trends in Seoul Korean compound tensification

Abstract

In a Korean compound composed of two nouns, WA and WB, if the initial onset of WB is a lax obstruent, it often undergoes tensification, as in /tol + tam/ [tol.t*am] stone wall. This tensification process does not occur in every compound. In /tol + kye.tan/ [tol.kye.tan] stone steps, the initial onset of WB /k/ remains lax. Treatment of exceptions differs in generative phonology and the recent phonological research. In generative phonology, rules apply categorically. Many of previous studies attempted to define the application condition of the tensification rule, but exceptional cases were found no matter how rules were defined. Therefore, this tensification phenomenon was considered unpredictable under the rule-based framework. However, the recent phonology accepts that the distribution of the exceptions themselves is phonologically patterned (Zuraw 2000). In other words, exceptional cases might not be a mere exception and might have phonological reasons to be a variant. Therefore, not a single factor, but rather the interaction of various factors, both phonological and non-phonological ones, decides the occurrence of tensification as a whole. Among these factors, the present study focuses on phonological ones. Specifically, I will examine what phonological factors and how significantly these factors contribute to the overall applicability of tensification. Zuraw (2011), the first systematic study on the distribution of Seoul Korean compound tensification, argued that both phonological and non-phonological factors attribute to the overall probability of the tensification. As inspired by Itos (2014) survey study on Yanbian Korean, I performed a similar survey on twenty-one Seoul Korean speakers employing 304 compound words. In the survey results, I identified if the trends caused by each factor are significant. The trends found to be significant are shown in (1).

(1) Trends in Seoul Korean compound tensification a. Tensification is more likely with high frequency items. b. Tensification is more likely when WA ends with an obstruent, followed by a nasal, a liquid and a vowel, in descending order. c. Tensification is more likely when WA ends with a liquid and WB also begins with a coronal consonant. d. Tensification is less likely when WB contains a laryngeally marked consonant.

I also attempted to develop a formal analysis of this phenomenon, in a frame of Optimality Theory (OT). Characterizing the variable pattern of the phenomenon, MaxEnt OT (Hayes & Wilson 2008) was employed. Constraints responsible for the occurrence of tensification are REALIZEMORPHEME (Morphemes are phonologically realized) and IDENT(tense) (Correspondent segments in the input and output are identical for the feature [tense]). Various trends found in the existing words were also formalized into the separate OT constraints. For (1a), following Ito (2014), a constraint *TENSE/LOWFREQUENCY (*T/LF, No tensification for low frequency items) was adopted. For (1b), regarding the highest tensification rate with WA obstruent final compounds, it was speculated that post-obstruent tensification, obligatory within a single accentual phrase in Korean (Jun 1993), still plays a significant role at the juncture of a compound. Therefore, a constraint *obs-lax (No lax obstruent after an obstruent in an accentual phrase) was established. Meanwhile, Ito (2014) mentioned that lenition might militate against the application of tensification. And the different tensification rates after a vowel and a sonorant might result from the different preferences for lenition. In this regard, Kirchner (1998) suggested that the impetus for lenition varies depending on the flanking segment: lenition is the most likely when a vowel precedes a target consonant, less likely when a liquid does, and the least likely when a nasal does. This hierarchy accords with the result of my survey where the tensification rate of the WA vowel final compounds was significantly lower than that of the WA liquid final ones, which in turn was lower than that of the WA nasal final ones. Capturing the different impetus for lenition, or blockage of tensification, among the three sonorous WA coda types, a constraint for each context was established: *TENSE/VOWEL_ (After a vowel, no tensification), *TENSE/LIQUID_ (After a liquid, no tensification), *TENSE/NASAL_ (After a nasal, no tensification). In Korean native monomorphemic words, a coronal consonant right after a liquid coda is always tense. Regarding the trend (1c), it was assumed that the coronal consonants greater impetus for tensification after a liquid might mirror this pattern in a monomorpheme. Martin (2011) proved that a morpheme-internal constraint diffuses its weaker effect across morpheme boundaries and termed it leakage in the sense that a phonotactic generalization somewhat leaks from the tautomorphemic domain to the heteromorphemic ones. Capturing this leakage effect at the compound juncture, a constraint *L(+)C (No sequence of a liquid and a lax coronal) was adopted. As pointed out in Ito (2014), the trend in (1d) is attributed to laryngeal co-occurrence restrictions. The presence of a laryngeally marked consonant, tense ([+constricted glottis]) or aspirated ([+spread glottis]), blocks tensification. Unlike in Ito (2014) in which the presence of a laryngeally marked consonant in both WA and WB considerably lowered the tensification rate, in my study, the presence in WB only showed this effect. The laryngeal co-occurrence restriction holds only within a stem, at least in the Seoul Korean compound tensification phenomenon. Thus the constraint OCP(stem) (No co-occurrence of laryngeally marked consonants in a stem) was adopted. Using Maxent Grammar Tool (Hayes 2009a), I took the survey results as the training data and tried to find the specific weights of those constraints, shown in (2).

(2) Constraint weight obtained OCP(stem) (1.676), *obs-lax (1.533), REALIZEMORPHEME (1.227), *TENSE/LIQUID_ (1.102), *TENSE/VOWEL_ (0.983), *L(+)C (0.785), *T/LF (0.429), IDENT(tense) (0.142), *TENSE/NASAL_ (0.0000006)

This weighted set of the constraints reflects the crucial effects observed in the existing lexicon. I also confirmed that the distribution of the data reproduced by this grammar highly accorded with that of the input data (R2=0.88). Therefore, it can be concluded that the proposed analysis can explain the various trends observed in the phenomenon.