Intervocalic -v- deletion in Tamil: evidence for aspect as a morphological category.

• Author: Schiffman, Harold F.

1. LEXICAL PHONOLOGY AND TAMIL

   1.1. Introduction

   Tamil possesses a rule of intervocalic -v- deletion that has eluded systematic attempts at formulation because of the diverse morphological and syntactic environments in which it occurs. In a previous analysis (Schiffman 1979: 6) I have stated the rule as occurring after the second syllable, but said little about various morphological exceptions. The problem is complex, since there are underlying -v-s in some lexemes that are deleted, but -v- is also introduced in inflection and in syntax by a sandhi rule that prevents vowels from coalescing. In particular, -v- is inserted after a back vowel and before any other vowel.

   The theory of Lexical Phonology (LP) as it has developed in the last decade offers some insights into a number of phonological processes that have hitherto been thought of as irregular, and its application to Tamil seems to shed some light on -v- deletion, showing it to be quite regular after some other phonological processes have occurred.

   The purpose of this paper is to examine these phenomena in detail. The most important conclusion it makes is that certain supposedly syntactic processes in Tamil must be considered instead to be morphological processes; to be more precise, they must be part of derivational morphology. Otherwise, phonological rules must be sensitive to syntactic rules of the language, which leads to the paradox of phonological rules relying on syntactic information that is no longer available, having already been deleted in earlier derivational processes.

   1.2. Theoretical Background

   Tamil has already been the subject of analysis using the LP model (Christdas 1987, 1988), as has its close sister, Malayalam, in the pioneering work of K. P. Mohanan (1986), Mohanan and Mohanan (1984), and T. Mohanan (1989).(1)

   Since the LP model is relatively recent and has not been applied to many Dravidian languages, or indeed many South Asian languages, it may be useful to summarize the theory briefly. A thumbnail sketch of this theory is given in Christdas 1988:

   The theory of Lexical Phonology (LP) centers around the assumption that phonological rules apply in the lexicon and interact with morphological rules. The lexicon, in this account, is not merely a repository for lexical entries, but is the domain of all the morphology as well as the phonological rules that are sensitive to the morphology (the morphophonemic rules of structuralist accounts).

   In all models of LP, the lexicon is organized into a hierarchy of levels, each constituting a well-defined and independent domain of morphological rules. In some models, each morphological level (or stratum) has one or more phonological rules that apply uniquely at that level.... In other models, the phonology constitutes a separate module within the lexicon, independent of the morphology.... Each phonological rule is specified for its domain of application, which can range over more than one level....

   ... [W]ord formation rules are generally assigned to discrete strata, unlike phonological rules, the domain of which may be a series of continuous strata....

   Most versions of Lexical Phonology assume that brackets delimiting lexical entries are erased at the end of each stratum. Bracket erasure thus is a blocking convention that prevents a later rule from referring to the internal structure of words at an earlier stratum (Christdas 1988: 46-49).

   Christdas goes on to point out that while there are no uniform accounts of LP (i.e., its practitioners disagree with each other in many ways), there is agreement on the need for strata or levels, with different kinds of word formation occurring at different levels because of the different kinds of morphological processes involved. In English, noun-formation suffixes like -ity and -al must occur at a different stratum than suffixes like -ness, -ism, and -dom. The reason is that the first set of suffixes affect stress in English, which in turn has an effect on the vowel quality of stressed vowels, whereas the second set can be added to stems or roots without such effect. But note that when national (derived from the stem nation and having a different vowel quality, i.e., [ae], than the vowel in the latter) is affixed with -ity, stress shifts and so does vowel quality; but when national is affixed with -ism, no stress or vowel changes occur. Phonological theories that do not allow for such levels are hard put to explain these vowel and stress shifts. Note also that there needs to be a notion of cyclicity, such that the output of the rules can either be recycled through the first level (nation [right arrow!! national [right arrow] nationality) or can form the input to the second level (national [right arrow] nationalism).

   All these processes must precede that part of morphology commonly called inflection, since inflection tends to be quite regular and applies to all appropriate categories, i.e., all count nouns can be inflected for plural (e.g., nationalisms, nationalities) regardless of how they have been derived.

   Levels also help to differentiate morphological processes that are in effect dependent on borrowed vocabulary, since in many languages different sources of vocabulary have different word-formation processes and different morpho-phonemic rules. Thus in English, Latinate vocabulary exhibits word-formation processes, affixes, and rules that differ from Germanic vocabulary; Dravidian languages show differences dependent on whether the vocabulary source is Indo-Aryan or native Dravidian.

   Finally it should be noted that there are also rules that apply after the lexicon, i.e., post-lexically. These are the kind of rules that apply everywhere, to all forms irrespective of their grammatical category, and have been referred to by some phonologists as "automatic." These post-lexical rules also tend to be gradient, i.e., there are differing degrees of lenition, flapping, vowel-rounding, or whatever the process is, rather than an either-or situation.

   1.3. Implications of LP Theory for Tamil

   When the foregoing assumptions about LP are sorted out and applied to the issues we are examining in Tamil, what appears to be a rather complex phonological rule or rules governing VDEL is actually a rather regular process once a number of other more problematical issues have been taken care of. The theory of LP allows us to see that some of these rules actually apply more than once, at different levels, and with different surface results. But the application of this theory also appears to reveal some new insights about the morphology and syntax of Tamil that are more difficult to reconcile with standard views of these matters, and require a departure from our previous ways of thinking about how Tamil is structured. In particular, the treatment of Tamil VDEL within the LP model seems to indicate that processes involved in deriving aspectually marked verbs, which have always been thought to be syntactic, now appear to be part of derivational morphology. Since the data adduced to illustrate the operation of VDEL rely on an understanding of not only the verbal aspectual system, but also such diverse phenomena as derived causatives and negative modals, this should not be surprising. Still it remains theoretically challenging to try to deal with these disparate phenomena with the same kinds of rules.

2. THE FACTS OF INTERVOCALIC -V- DELETION

   2.1. Complications Raised by Tamil Diglossia

   Any description of Tamil is complicated by the fact that the language is severely diglossic (Britto 1986); many descriptions of Tamil do not state whether the norm being described is Literary Tamil (LT), some variety of Spoken Tamil (ST), or some kind of mixture. Researchers also disagree as to whether a standard form of ST exists, and puristically inclined native speakers are known to deny even the existence of ST and/or skew their pronunciation in the direction of LT when questioned about Tamil. All these factors complicate any analysis of Tamil that focuses on a feature of ST that is not shared by LT. The rule in question, VDEL, is one of these features. Since it is not found in LT, some find it tempting to describe it as a historical rule that has resulted in the absence of intervocalic -v- in certain places in ST where LT still retains it, and let it go at that. Fortunately or unfortunately, the picture is not that simple. LT gives us evidence of environments where intervocalic -v- once occurred, but its deletion in ST is not the result of a regular across-the-board historical process.

   The rule Of VDEL at first glance appears to delete intervocalic [v]s in most environments with the exception of [v]s that mark the future of certain verbs, [v]s that are the markers of deverbal-nouns (of certain classes of verbs), and [v]s that arise in sandhi processes. There are exceptions, however, and they have not lent themselves to simple analysis in the past.(2) As mentioned, on the surface the rule seems to apply after the second syllable, with various morphological exceptions, but this is oversimplistic.

   The environments in which VDEL occurs are as shown in table 1. It should be noted that when VDEL occurs it almost always takes the subsequent vowel with it; i.e., it is a syllable-deletion rule that operates on certain syllables that begin with -v-, but only intervocalically. Morphemes that begin with -v- occurring in isolation are not subject to the rule unless by some morphological or syntactic process they are attached to other morphemes, such that an intervocalic environment arises.(30 But as we stated, not all surface -v-s are deleted in ST, so our task is to find some constraints on this rule that are not arbitrary and capricious.

   [TABULAR DATA OMITTED]

   We will begin by attempting to establish whether all surface occurrences of -v- have the same underlying source, and then go on to deal with morphologicial problems that must be solved before the phonological issues can be sorted out. We...