Vowel harmony in isiXhosa: an OT and acoustic study of [ATR]

The vowel harmony system in isiXhosa is centred on a process of vowel raising. All mid-vowels preceding a high vowel take on the feature advanced tongue root (ATR) (e.g. thɛnga ‘buy’ → thengisa ‘sell’; bɔna 'see' → bonisa 'cause to see') (Harris 1987). The process of mid-vowel assimilation for the feature [+ATR] is consistent in all instances in which the mid-vowel occurs preceding a high vowel trigger, unless harmony is blocked by the low opaque vowel [a]. This is the analysis presented in Jokweni & Thipa (1996) the only previous literature to address the vowel harmony process of isiXhosa in detail. As an alternative approach to the rule-based phonology applied in the analysis presented by Jokweni & Thipa (1996), I propose the introduction of Optimality theory (OT) (Prince & Smolensky 1993, Bakovic 2000, and Pulleyblank 2002). I will present a map of the harmony system of isiXhosa using OT, while also presenting acoustic data to supplement the selected examples provided in Jokweni & Thipa (1996). This acoustic investigation will determine whether the harmonic feature is ATR, and how this feature patterns among vowels in different phonological contexts. In this paper vowel harmony is achieved through the implication of numerous rules, and with very specific directional and prosodic limitations on the spread of [+ATR]. Using generalisations based on my own collected data as well as those reported in previous literature, I have developed a constraint ranking to account for the harmony process in isiXhosa. By adapting the No-disagreement approach to harmony (Pulleyblank 2002), the final constraint ranking has the capacity to derive the optimal phonetic candidate for every harmony case. A selection of spread constraints is used to account for the raising as well as blocking processes, by driving either regressive or progressive spreading. Within the original No-disagreement approach a spread constraint would recognised only one feature in its prohibition of disagreeing segments. However, in the adapted approach the spread constraint driving [+ATR] assimilation is combined with a feature of correspondence (Krämer 2001) which considers the height as well as the ATR value of the sequential segments. The constraint is therefore adapted to consider more than one feature and is not activated unless the sequential segments agree for this particular feature. The regressive spread constraint is therefore only activated when the consecutive segments have an agreeing height value. The introduction of this adaptation was necessary to provide a more nuanced OT approach with the capacity to effectively characterise the idiosyncrasies observed in this harmony pattern. The harmony constraints are therefore no longer contradict one another by simultaneously driving harmony in opposite directions. Furthermore, the direct acoustic analysis is completed by means of the PRAAT software, to answer the salient question of the definitive harmonic feature. To provide a multiplicity of empirical evidence I have recorded utterances containing a number of vowel combinations. Each combination positions the alternating mid-vowels in a particular phonological context from which instances of ATR alternations have been extracted and phonetically analysed. Using the generalisations reported in Jokweni & Thipa (1996) as a starting point, the acoustic signal of each mid-vowel within a set phonological context is annotated for a predicted ATR value. Hence, if a mid-vowel occurs preceding a high vowel it is annotated as [+ATR] etc. The data sets representing each of the mid-variants found in a specific phonological context are then plotted into vowel charts and compared by means of statistical analysis (Baayen 2008, Bluman 2000). The results are then used to determine whether any significant phonetic alternation is occurring, and what the acoustic distinction between [+ATR] & [-ATR] variants is essentially comprised of. The final acoustic results indicate a significant difference between the mid-vowel ATR variants extracted from specific phonological contexts. Hence, due to co-articulatory effects or some other phonological influence the realisation of [+/-ATR] variants exist along a spectrum, and are therefore not phonetically consistent, but indicate a different acoustic make-up across the various groups