A Duration-based Solution to the Problem of Stress Realization in Turkish

Abstract

A duration-based solution to the problem of stress realization in Turkish Anne Pycha University of California, Berkeley Department of Linguistics pycha@berkeley.edu I. Introduction In this paper, I argue that the realization of Turkish stress makes crucial reference not to syllable types (such as heavy versus light), but to syllable durations. In agreement with previous work, I show that fundamental frequency (F0) acts as a primary acoustic correlate of stress in the language. Contrary to what we might expect, however, F0 contours realize themselves fully on syllables closed by a sonorant consonant (abbreviated here as CVR), but are “clipped” on syllables with long vowel nuclei (CVV). This presents a puzzle for any theory of phonological weight, whereby CVV syllables should certainly be heavier, and therefore more capable of hosting prosodic information, than CVR syllables. The puzzle resolves itself when we examine phonetic duration: the F0-carrying portion of CVV syllables is in fact consistently shorter than that of CVR syllables. Following Gordon (1999), these findings suggest that exceptions to typological tendencies in syllable weight can find a principled explanation within phonetic implementation. 2. Background Turkish has at least three ways of assigning stress (Sezer 1981, Inkelas & Orgun 1998, Inkelas 1999, Kabak & Vogel 2001, Inkelas & Orgun 2003, and references cited therein). Final stress is the default. Certain morphemes, however, condition the presence of non-final stress; an example is the negative morpheme -mA, which conditions stress on the syllable preceding itself. Place names undergo a totally different stress placement algorithm, dubbed “Sezer stress” after Sezer (1981): here, stress falls on the antepenultimate syllable if it is heavy and the penultimate syllable is light, otherwise it falls on the penultimate syllable. (1a) git-'ti-m go-P AST -1 SG ‘I went’ (1b) 'git-me-di-m go-N EG -P AST -1 SG ‘I didn’t go’ (1c) 'ankara, is'tanbul ‘Ankara’, ‘Istanbul’ In this paper, I will not consider Sezer stress, but will be concerned only with final and non-final stress. Previous researchers who have investigated the acoustic correlates of Turkish stress have concurred that F0 acts as a primary correlate (Konrot 1981, Konrot 1987, Levi 2005), but have also noted that final and non-final stress seem to get realized very differently. Konrot (1981) identified both F0 and intensity as correlates, but only in words with non-final stress. In words with final stress, he found no robust correlates at all. Levi (2005) identified F0, intensity, and duration as correlates, when averaging over words with final and non-final stress. But she too encountered the non-robustness of correlates for final stress, as revealed by her algorithm for