METHODOLOGICAL ASPECTS OF SEGMENT-RELATED AND SPEAKER-RELATED VARIABILITY - A STUDY OF SEGMENTAL DURATIONS IN DUTCH

This paper addresses a number of issues concerning the effects of spea ker idiosyncrasies on phonetic regularities (in this case of segmental durations). First, a comparison is made between the amount of speaker -related variability and the amount of segment-related variability. Ne xt, we show a ranking of segments according to their speaker idiosyncr asy. And, finally, we point out a way in which the observed speaker id iosyncrasies may be incorporated in phonetic rule models. We start out with three durational analyses on 24 /CVC(sic)/ quasi-Dutch nonsense words collected from five male and five female speakers who read the w ords ten times in isolation. The vowels used were /a, i, u/; the conso nants, which appeared once in C-1-position and once in C-2-position, w ere /p, t, k, d, s, m, n, r/. It is demonstrated that a segment-relate d effect may be confounded by speaker idiosyncrasies. We discuss how s uch speaker-related effects can be evaluated with respect to a phoneti c rule system in a methodologically proper manner. For this purpose we introduce ''obligatory'' (speaker-independent) and ''optional'' (spea ker-dependent) rules. However, this can only successfully be carried o ut if some a priori criteria have been defined. These criteria concern the percentage criteron needed to claim that a rule is active and the statistical power that is needed to render reliable results.