This dissertation deals with some basic theoretical problems concerning the phenomenon of vowel harmony in a framework combining insights from Government Phonology, Optimality Theory and Lexical Phonology. Vowel harmony is a process whereby all vowels in (roughly speaking) a word are required to agree with each other with respect to one of their properties. For example, in a language with palatal harmony, like Hungarian, every vowel in a word is either front or back. On the one hand, this means that all vowels of polysyllabic roots are either front or back. On the other hand, all affixes containing a vowel have two allomorphs, one with a front vowel and one with a back vowel, the choice of which depends on the root to which the affix is attached (e.g. város-ban ‘in a/the city’ vs. tömeg-ben ‘in a/the crowd’). That is, the vowels harmonise with one another. This phenomenon is interesting for several reasons for any phonological theory. One of its most challenging aspects is that it looks like a process which operates non-locally, because it ‘skips’
(or disregards) intervening consonants. In a theory like Government Phonology (cf. Kaye, Lowenstamm & Vergnaud 1985, 1990), however, there is an independently motivated level where vowels are in fact adjacent to each other. Since a vowel forms the head of the syllable that contains it, vowels in their function of syllable heads can be projected to a separate level where they can “see” each other. Consonants, on the other hand, cannot be projected in the same way, and this enables us to explain why consonant harmony of a similar sort does not exist. In a theory like this, the claim can be maintained that all phonological processes operate locally. Such a restriction constrains the number of possible grammars considerably, and this is the main reason why I chose to employ the representational theory of Government Phonology in this dissertation.
Apart from this basic problem, research on vowel harmony involves three main areas. The first concerns the question of what types of vowel harmonies exist in the world’s languages, and which feature theory can account for this typology in the best way. According to modern theories of phonology, sounds can be divided into smaller ingredients, called distinctive features. Since vowel harmony involves the agreement of vowels within a certain domain with respect to a particular property, or feature, a given feature theory predicts that there are as many possible types of harmony as there are vocalic features recognised by the theory. Feature theories thus can be tested on the basis of whether they make correct predictions about the typology of vowel harmony systems. In this dissertation, I argue that the feature theory of Government Phonology can account for the possible types of harmony. The second issue concerns the domain of vowel harmony. This domain is usually defined as the “word”. One question we need to answer here is whether this domain is defined in terms of morphology or phonology (since members of compounds, for example, constitute separate harmonic spans); and if it is defined phonologically, whether it is a prosodic domain or something else.
In this dissertation, I argue for a phonotactic definition of the domain of harmony; more precisely, that it coincides with the analytic domain. Another question concerns the existence of disharmonic roots and disharmonic affixes. The former can be exemplified by the root kosztüm ‘costume’ in Hungarian, and the latter by the suffix -kor (cf. öt-kor ‘at five o‘clock’, not *öt-kör). The
domain of harmony should be defined in such a way that systematic characteristics of disharmonic strings are accounted for
as well. The third research area concerns neutral vowels. These are those vowels in a given system that do not have a harmonic
counterpart. Their neutrality is manifested by the fact that they can co-occur with vowels of both harmonic sets. Affixes containing
neutral vowels have only one allomorph, and they do not alternate depending on what type of root they are attached to. Not all
neutral vowels behave in the same way, however. On the basis of their behaviour, two main types can be distinguished. One type is
called ‘transparent’, because harmony goes through these vowels as if they were not there. That is, if a suffix vowel follows a stem
that ends in a neutral vowel, the suffix vowel will harmonise with the non-neutral vowel to the left of the transparent vowel, so to
speak ignoring what is intervening. The other type is called ‘opaque’, because these neutral vowels stop the harmony. In these cases, the following suffix vowel harmonises with the neutral vowel itself, ignoring what is preceding in the stem. The issue of transparent vowels is connected to the problem of locality mentioned above, because it seems as if harmony had ‘skipped’ the transparent vowels. Van der Hulst & Smith (1986) solve this problem, and they further claim that the two types of behaviour exhibited by neutral vowels can be predicted from the segmental make-up of these vowels themselves. In this dissertation, I test their theory, and show that not all the possibilities predicted by it actually occur in the world’s languages. I propose that it is possible to predict which possibilities do not occur if we take into account certain properties of the vowel systems involved. To be able to offer a solution for these three problems, I had to take a position concerning some more general issues of phonological theory. This forms the first part of the dissertation. In
chapter 2, I argue that considerable advantages can be gained if the framework of Government Phonology is combined with Optimality Theory (cf. Prince & Smolensky 1993). I show that ranking is necessary to be able to account for certain types of phenomena. I use the “principle” of government licensing as an example, which is in conflict with the “principle” of proper government word-internally. Certain languages resolve this conflict in favour of one of the principles, while other languages resolve it in favour of the other. If we accept that ranking of principles (or rather constraints) is necessary, this makes it possible to carry out some further positive changes in the theory. I argue that language variation can now be expressed exclusively by ranking, and consequently the notion of parameters can be abandoned. I illustrate this by replacing the parameter that licenses domain-final empty nuclei by a violable constraint. As a
consequence, consonant final words now end in an onset, instead of ending in an empty nucleus.
The theory developed in chapter 2, however, is still not rich enough to solve certain types of problems. Processes of the lexical phonology can show ‘derived environment effects’. This means that they can be restricted to apply only in a derived environment, but not within monomorphemic forms. That is, we need to retain some version of the Lexical Phonological notion of the Strict Cycle Condition. I formulate a non-derivational version of this condition in the form of a violable constraint and call it DERIVED
ENVIRONMENT CONSTRAINT. This constraint prohibits changes within a single analytic domain. To avoid having to refer to
‘neutralising’ changes, which would complicate the evaluation procedure of this constraint considerably, I argue that in fact all
lexical processes are neutralising. This means that the principle of Structure Preservation is fully adhered to in the lexicon. I show on the basis of some representative examples that processes which are not structure preserving can be argued not to be lexical on
independent grounds. They are either ‘word level’ rules (in the sense of Borowsky 1993), or they are postlexical. After this general introduction, the second part, comprising chapters 4, 5 and 6, deals with the phenomenon of vowel harmony. In chapter 4, I look at the issue of harmonic features. Given a particular theory of (vocalic) features, we expect as many types of vowel harmony as there are features allowed for by the theory. In an element-based feature theory, such as Government Phonology, where there are only three elements, I, A and U, supplemented by the property of headedness, this means that we expect four types of vowel harmony. In the first part of the chapter, I give an example of each type.
In a theory like this, there is no primitive corresponding to the feature [high], thus we do not expect to find cases of raising harmony. In the second part of the chapter, I discuss the case of Pasiego Spanish that has been argued to exhibit exactly this type of harmony. I argue that it is possible to analyse the process in Pasiego without referring to the absence of the element A in a framework that combines the insights of Optimality Theory with those of Government Phonology. I propose a constraint that requires that the combination of elements in a governed position is licensed by a governing A. This constraint is ranked above PARSE (A). Thus the element A will always be deleted from a complex expression, if it is not supported by an A in the governing position without having to state this negative condition as the trigger of the process. Moreover, raising is not regarded as the same type of harmony as spreading harmonies, since only sequences of mid vowels followed by high vowels are ruled out by it, and complete uniformity is not required in sequences of non-low vowels. In chapter 5, I investigate the issue of the harmonic domain, and how disharmonic roots and affixes should be handled in the adopted framework. On the basis of a detailed analysis of Turkish vowel harmony, I propose that harmony applies with reference to the domain introduced in chapter 3, the ‘analytic domain’.
Furthermore, I argue that vowel harmony is no longer active in Turkish roots, and this is why there are so many disharmonic roots in the language. In other words, vowel harmony is one of those lexical processes that can exhibit derived environment effects. This can be expressed by ranking the DERIVED ENVIRONMENT CONSTRAINT, introduced in chapter 3, above the HARMONY constraint responsible for spreading. For disharmonic suffixes I claim that they can be of one of the following two types: (i) they either behave as parts of compounds,in this way falling into the category of ‘compounding analytic’ suffixes; or (ii) they are unproductive derivational suffixes, belonging to the category of ‘synthetic’ suffixes. The first type of suffixes do not undergo harmony, because they form an analytic domain of their own. The second type of suffixes, on the other hand, form one unit phonologically with the stem they attach to, and harmony does not apply to them for the same reason why it does not apply in non-derived roots. This means that harmony only applies if analytic suffixes are added to the root. In chapter 6, I test the predictions which the theory proposed by Van der Hulst & Smith (1986) makes concerning the typology of neutral vowels. According to this theory, neutral vowels are expected to behave in one of two ways on the basis of the segmental make-up of these vowels: (i) they either possess the harmonic feature and they are transparent to harmony (since the harmonic feature is compatible with them and thus can spread through them); or (ii) they lack the harmonic feature and they are opaque (because harmony cannot skip any vowels).
I show that not both of these possibilities occur in all types of harmony systems. I propose to account for the non-existent cases on the basis of particular properties of the vowel systems in question. More precisely, I claim that harmony cannot involve those elements that reside on a fused line. This recaptures the original autosegmental idea that only features specified on separate autosegmental tiers can exhibit long-distance dependencies. As a consequence, I- and U-harmony are only possible in vowel systems that contain front rounded vowels; whereas A- and ATR-harmony can occur in triangular systems, as well. Finally, I discuss the role of an apparently very powerful
constraint that prohibits the combination of the element A with the property of headedness.
In summary, the combination of a theory of representations with a theory of constraint interactions and with a theory of phonological derivations can handle several phenomena in a satisfactory way that cannot be accounted for in either theory when they stand on their own. Since these theories cover different empirical domains, it never has been claimed that they are inherently incompatible. However, it has not been suggested either that they should be combined.