Dr Steven Kaye

This thesis investigates the origins and behaviour of the non-canonical morphological phenomenon of heteroclisis in the verb paradigms of Latin and the Romance languages. Heteroclisis is the coexistence, within a single paradigm, of forms which pattern according to different inflectional classes existing otherwise in the language: a heteroclite lexeme can thus be seen as 'mixed' or 'undecided' as to its inflectional identity. I begin by examining the development of the theoretical concept of heteroclisis and approaches to the idea of inflectional class in general, before situating heteroclisis in typological space in comparison with better-known instances of non-canonical morphology such as deponency and suppletion; heteroclisis exists at a different level of generalization from these, because its identification presupposes the existence of inflectional classes, themselves generalizations over the behaviour of individual lexemes. I also consider two recent theoretical treatments of the phenomenon and survey recent linguistic studies making use of the notion. I then look at the synchronic and diachronic behaviour of heteroclisis in Latin and Romance verbs: the great time depth of our attestations of these languages gives us the chance to witness the development of successive examples of heteroclisis, and their subsequent treatment within the morphological system, in the history of a single family. Focusing chiefly on data from Latin, Romanian and Romansh, I find that the principal (though not the only) source for new instances of heteroclisis in Latin/Romance lies in regular sound change, and find that speakers can treat these synchronically anomalous patterns as robust models of inflectional behaviour to be extended over the lexicon or brought into line with pre-existing types of paradigm-internal alternation. These findings concur with previous demonstrations that speakers make use of non-canonical phenomena as markers of the internal structure of inflectional paradigms.

In this chapter I look at the diachronic development of the opposition between so-called ‘present’ and ‘past’ stems in the Iranian verb, and at the particular distribution which these stems currently show in the different verb classes of northern Talyshi (north-western Iranian; Azerbaijan). A small group of highly frequent northern Talyshi verbs shows a morphomic pattern of stem alternation over the paradigm, which can be accounted for in historical terms; but all other verbs in the language are said to be losing, or to have already lost, the paradigmatic opposition between the two inherited stems, retaining just one of them as part of a long-term trend towards inflectional simplicity. I suggest that the majority of northern Talyshi verbs show no sign of being reduced to a single stem; and those verbs which genuinely retain only one stem have done so because of the unpredictability of the formal relationship between the two inherited stems, and not because of the morphomic distribution of those stems over the paradigm. I also draw attention to the existence of a group of verbs, as defined by stem behaviour, which has not been identified for northern Talyshi before.

Cyril Graham’s The Avar Language, a treatise consisting of a linguistic description and an extensive English-Avar wordlist, originally appeared in the late nineteenth century in the Journal of the Royal Asiatic Society of Great Britain and Ireland, and hasbeen republished in the early twenty-first century in book form, with Russian translation and commentary by Boris Ataev of the Russian Academy of Sciences, Makhachkala. Welcoming Ataev’s contribution in making it accessible to the modern Russophone audience, I discuss the linguistic qualities and shortcomings of Graham’s article as well as the complex and revealing history of its composition. Engagingly written and in some respects perceptive, while in other respects outmoded even in its own time, it provides an insight into the early development of Caucasian linguistic study in the West.

Andi (Nakh-Daghestanian; Daghestan) presents a striking pattern of agreement in which nominals in the affective case agree with the absolutive argument of their clause. Agreement between arguments is observed with verbs of perception and cognition which require affective case on the experiencer and absolutive case on the stimulus argument. Agreement between the arguments of a predicate is at best unexpected, and accordingly attempts have been made to demonstrate that apparent instances of this behaviour have an alternative explanation. The behavioural properties of agreeing nominals in Andi are explored to elucidate the syntactic status of the item expressing the experiencer. A robust set of evidence is presented (including the results of tests on imperative clauses, control structures and reflexive clauses) to establish that in Andi the experiencer is a genuine argument of the verb and has a range of subject properties.

Andi (Nakh-Daghestanian; Russia) displays a typologically remarkable phenomenon: adverbs of numerous morphological and functional types inflect for agreement with a clause-level controller. To the extent that adverb agreement has been observed elsewhere, it is commonly taken to signal that the items involved are semantically oriented towards the participants they agree with, aligning the phenomenon with secondary predication. This paper demonstrates that Andi works differently: the widespread clausal agreement seen on Andi adverbs is insensitive to participant orientation. While agreement exponence on adverbs is morphologically complex, a simple structural principle (modelled here in Minimalist terms) ensures that clause-level agreement is always with the absolutive-case argument. The Andi facts thus provide evidence for a typological distinction between those languages where clausal agreement on adverbs can serve a semantic function and those where it cannot. A potential challenge is posed by the exceptional “biabsolutive” construction, where both subject and object appear in absolutive case and either may control adverb agreement, suggesting a role for some additional non-structural factor. However, on independent grounds this paper identifies the two arguments as belonging to distinct structural layers; this apparent flexibility in controller choice merely reflects the ability of certain adverbs to modify either layer.