Directionality and locality in vowel harmony

This dissertation concentrates on vowel harmony, a well-known process of assimilation where one vowel assumes similarity with regard to a certain feature in a neighbouring vowel. Recent work on vowel harmony claims that directionality is not an independent parameter along which vowel harmony languages are organised. However, primarily focussing on Assamese (based on original research and data), I show that regressive harmony is exclusively unidirectional in Assamese. In Assamese harmony, the high vowels /i/ and /u/ regressively trigger harmony, irrespective of morphological considerations, if the preceding vowels are [-ATR] (ATR-Advanced Tongue Root). This regressive triggering is shown to be the result of a precedence relation, where 'marked' sequences of vowels are prohibited. The requirement of a precedence relation leads to 'contextual neutralisation', where the context of one vowel following another vowel results in a 'marked' sequence of vowels, due to a mismatch in their [ATR] features. Therefore in Assamese, [-ATR][+ATR] sequences are marked, and it leads to regressive harmony resulting in less marked /e…i/ and /o…i/ /u…i/ vowel sequences. Also analysing Pulaar and Karajá I show that the same analysis holds for regressive harmony in these two languages as well. Assuming the constraint based framework of Optimality Theory (Prince and Smolensky 1993/2004), it is shown that regressive [ATR] harmony in Assamese, Pulaar and Karajáan only be analysed with the aid of a higher, though violable constraint *[-ATR][+ATR]. As an extension of the argument that regressive directionality is independent of morphological considerations, I also deal with non-iterative regressive harmony in Bengali and Tripura Bengali. These languages require an analysis similar to that of Assamese, but the restricted domain of harmony is dependent on an additional requirement of [+Hi] apart from [+ATR] and therefore, the relevant constraint is *[-ATR][+HI +ATR]. Further, it is shown that locality is another important factor in the analysis of vowel harmony. Locality in Assamese is apparent from the blocking of harmony by a nasal segment only in the immediate vicinity of the triggering segments /i/ and /u/. Nasal segments anywhere else except in the immediately preceding position of a triggering segment cannot block the spread of harmony. An important aspect of this chapter is that it argues that blocking by consonants in vowel harmony is due to the similarity of the blocking segment. Similarity of the blocking segment can be evaluated in either or both of the following two ways: (a) the blocking segment shares a feature with the trigger (b) the blocker is more sonorous and therefore vocalically compatible. Locality in the vowel harmony process of Assamese is also evident in exceptional cases of vowel harmony. When regressive harmony is triggered by the two morphemes /-iyα/ and /-uwα/, they exceptionally raise the otherwise non-participating vowel /α/ in the stem to either /e/ or /o/ - only if the /α/ is in close proximity to the triggering segment. I analyse this exceptional triggering to be the result of a requirement involving local domains, where the domain includes a part of the exceptional trigger. In a nutshell therefore, this dissertation shows that exclusively leftward vowel harmony in languages is inherently a directional process. Further, there are various locality requirements in a long-distance process like vowel harmony, which has been explored in considerable detail in this dissertation