Does Turkish have a two-way voicing contrast?Date Released: Mon, 4 April 2011 10:01 +0200
Dr Zhaleh Feizollahi presented some research from her PhD thesis on the phonetics and phonology of Turkish consonants at a Linguistics Departmental Research Seminar on Tuesday 29 March. This is the abstract for her presentation:
The phonetic implementation of a two-way voicing contrast across languages has been subject to much debate and is cited in arguing for the phonetic transparency of laryngeal features and in the modularity of language-specific phonetic grammar (Chomsky & Halle 1968, Halle & Stevens 1971, Keating 1984, Petrova et al. 2006). Turkish has been cited in this debate as one of the few languages (including Dutch and Swedish) which implements a two-way voicing contrast in stops as prevoiced [b] vs. voiceless aspirated [ph] (Keating et al. 1983, Petrova et al. 2006). In this paper, I present the results of an acoustic study of four native Turkish speakers and review literature (Wilson 2003, Kallestinova 2004), which demonstrate that this classification may be incorrect. The present study finds that orthographically voiced stops lack prevoicing in word-initial position following voiceless consonants (Vt#dV). Prior studies also demonstrate a lack of prevoicing in word-final position (Vd#V, Vd#gV, Vd#]), and variably in phrase-initial position ([#dV). Turkish stops then pattern more like an aspiration language such as German (Jessen & Ringen 2002), since prevoicing only reliably occurs in a voiced context (V#dV, Vm#dV, Vz#dV).
Two perception studies were also conducted in order further investigate voice onset time in Turkish. The results of the first study demonstrate that listeners are able to correctly identify excised VC#CV clips 80% of the time, thus I argue that the lack of prevoicing in word-initial orthographically voiced stops constitutes an incomplete neutralization. The second study digitally manipulated the voice onset time in word-initial stops and asked listeners to categorize the word-initial stop as either voiced or voiceless. Results demonstrate that the crossover points did not necessarily correspond to VOTs produced by speakers. Thus, I argue that voice onset time is only one of several acoustic cues listeners use to distinguish voiced and voiceless stops in Turkish. I demonstrate that phonetically transparent laryngeal features cannot account for the complete and incomplete voicing neutralizations found in Turkish. I, therefore, follow Keating (1984) who proposes that the cross-linguistic differences in a two-way stop-voicing contrast are derived via a separate language-specific phonetic implementation stage.