Haruo Kubozono - kubozono@lit.kobe-u.ac.jp
Department of Linguistics
Kobe University
Rokkodai, Nada, Kobe, 657 Japan
Summary of Paper 5pSC6
Presented Friday afternoon, December 6 1996
3rd Joint ASA/ASJ Meeting, Honolulu, Hawaii
Embargoed until December 6, 1996
Previous studies on speech disfluencies in Japanese have revealed several interesting parallels between speech errors and stuttering with respect to their linguistic patterning. The most important to the present study concerns the precise location where utterances are disrupted in natural connected speech. This paper attempts to examine (i) the disruption patterns found in stuttering and speech errors in Japanese in comparison with those of English, and (ii) the relationship between these patterns and the linguistic construction assumed in theoretical linguistics.
A statistical analysis of over 700 stuttering samples produced by adult Japanese speakers (Ujihira & Kubozono 1994) reveals, among others, that CV (Consonant- Vowel) forms a basic segmentation unit in which stuttering occurs. In repetition-type stuttering, for example, the pattern in (1b/c) overwhelms other patterns in (1), with this pattern alone accounting for nearly 90% of the samples. The dominance of this pattern shows up most clearly in words beginning with a long syllable: Namely, the pattern in (1c) outnumbers those of (1a) and (1d). In other words, words beginning with a long syllable (CVC or CVV) break typically after the nuclear vowel, resulting in the repetition of the word-initial CV ("." denotes syllable boundaries).
(1) a. s - s - soo `so'
g - g - go.nen `five years'
b. no - no - no.bo.ru `to climb'
c. na - na - nan.de `how come?'
so - so - soo.si.ki `funeral'
d. tai - tai - tai.hen `extremely'
gen - gen - gen.yu `unrefined oil'
e. tabi - tabi - ta.bi.ni (de.ru) `(go on) a trip'
Interestingly, a similar asymmetry is observed in speech errors produced by normal Japanese speakers, with long syllables splitting into CV and C (or V) rather than C and VC (or VV). (2) gives a representative example of each error type (Kubozono 1989, 1995, 1996).
(2) a. Substitution: su.tei.sjon su.ten.sjon `station'
b. Transposition: tek.kin kon.ku.rii.to kok.kin ten.ku.rii.to `
ferro-concrete'
c. Blend: ne.ko `cat' / njan.ko `kitty' nen.ko
The disruption patterns in Japanese show a remarkable contrast with the patterns exhibited by analogous disfluency processes in many other languages. In English, for example, the most common disruption point in stuttering seems to be between the first consonant and the next segment, the latter being either another consonant or the nuclear vowel, as illustrated in (3).
(3) a. n - n - n - n - never
b. d - d - difficult
c. s - s - s - street
A similar difference between Japanese and English can be seen in speech errors. English speech errors generally split syllables between the onset consonant(s) and the nuclear vowel, as illustrated in (4) (Fudge 1987).
(4) a. Substitution: Tom and Jer.ry Jom and Jer.ry
b. Transposition: Ro.man Ja.kob.son Yo.man Ra.kob.son
c. Blend: smart / clev.er smev.er
The data given above suggest that Japanese disfluency processes exploit CV as a basic unit which cannot be split. Generally speaking, CV is a basic unit in Japanese in at least two ways. First, Japanese is a typical open-syllable language where CVs (open syllables) are much more frequent than CVCs (closed syllables) (Kubozono 1995). Moreover, two of the three writing systems in Japanese- -namely, hiragata and katakana--are CV-based systems, where each letter corresponds basically to the sequence of the onset consonant and the nuclear vowel. Given these correspondences, it will not be surprising to find that CV serves as a basic segmentation unit in speech disfluencies and, hence, that disruptions occur without splitting this basic unit into two parts.
Seen from a linguistic point of view, CV plays an indispensable role as a unit of speech rhythm in Japanese, where every CV takes an equal duration of time (in a psychological sense, if not in a strictly physical sense). This timing unit is traditionally called `mora'. In terms of the relationship between the prosodic unit called `syllable' and the mora just defined, long syllables, i.e. CVC and CVV, consist of two moras with a mora boundary lying between CV and C (or V). The disruption patterns illustrated in (1b/c) and (2) can then be generalized as breaking words at a mora boundary.
Assuming that the way we produce speech in our native language is directly related to the prosodic organization of that particular language, it follows that the syllable in Japanese has an internal structure in which the mora forms a constituent of the syllable, thereby predicting the segmentation pattern in (1 b,c) and (2). This syllable structure model marks a departure from the traditional `onset-rhyme model' in which the nuclear vowel and the following consonan t(s) form a constituent called `rhyme'.
References
E. Fudge, `Branching structure within the syllable.' Journal of Linguistics 2 3 (1987).
H. Kubozono, `The mora and syllable structure in Japanese: Evidence from speec h errors.' Language and Speech 32-3 (1989).
H. Kubozono, `Perceptual evidence for the mora in Japanese'. Papers in Laborat ory Phonology IV (Cambridge University Press, Cambridge, 1995).
H. Kubozono, `Speech segmentation and phonological structure' Phonological Str ucture and Language Processing: Cross-Linguistic Studies. (Mouton de Gruyter, Berlin, 1996).
A. Ujihira & H. Kubozono, `A phonetic and phonological analysis of stuttering in Japanese,' Proceedings of 1994 ICSLP, (1994).