2.1Approaches to the study of valency classes
All verbs in a language have different meanings, but with respect to their syntactic patterning, they show striking similarities and fall into a relatively circumscribed number of valency classes whose members behave alike. By âsyntactic patterningâ, we refer not just to the coding patterns (i.e. the ways in which the verbal arguments are flagged and indexed), but also to the behavior of verbs with respect to alternations such as causative, passive, applicative and other constructions that are not fully productive for all verbs (e.g., secondary predicates, certain word order phenomena). It has often been noted that these syntactic verb classes typically correlate with semantic classes. The syntactic properties of verbs can be studied separately from the semantic classes, so the strength of the correlation is an empirical question.
The literature on syntactic verb classes is vast, yet with few exceptions it is limited to relatively few languages, most of them European. A classic study in this area is Levinâs (1993) book on English verb classes, where she argues that verb classes are semantically based and can be identified in terms of their argument structures, possible argument structure alternations, as well as further syntactic diagnostics (such as middle alternation, unspecified object deletion, there-insertion, etc.). Unlike earlier studies, which divided the verbal lexicon into a few highly general classes (e.g., stative vs. active verbs, intransitive vs. transitive vs. ditransitive, or, for intransitives, unaccusatives and unergatives), Levin attempts a much more fine-grained classification, which is constructed in a bottom-up fashion, and where verb classes are defined in terms of their overall syntactic distribution. Levinâs study (as well as subsequent work with M. Rappaport Hovav; e.g., Levin & Rappaport Hovav 2005) has been highly influential not only in the theoretical work on lexical semantics, but also in computational linguistics, and underlies verb ontologies in WordNet (Fellbaum 1998) and extensions thereof such as VerbNet (see Kipper-Schuler 2005 for references) and FrameNet (see, e.g., Fillmore et al. 2003; see also Schulte im Walde 2003 for discussion of verb classes in German).
But neither Levinâs study nor the pioneering study Experimental Investigation of the Russian verb by Apresjan (1969), where verb classes were also established on syntactic grounds, have been extended cross-linguistically. Jones (1994) is a small collection of working papers dealing with verb classes in English, German, Korean and Bangla, stemming from an MIT-based project explicitly aiming to extend the Levin-style classification to other languages. There are also occasional studies dealing with other languages (cf. Fukui et al. 1985 on Japanese; Vogel 2003 on Jarawara). Even contrastive studies devoted to a single verb type, such as the study of interaction verbs in English, German, Hungarian and Maori in Blume (1998), are rare.
This lack of an extension to more languages does not seem to be accidental. In spite of its merits, Levinâs approach faces a number of problems which become evident once one attempts to extend it beyond English. First, since Levinâs classes are constructed on syntactic criteria, they are not always semantically coherent. This is a serious drawback for typological studies, where the phenomena to be investigated have to be defined in semantic terms to make a comparison possible. The main question is: Which aspects of this classification are universal and which are language-particular? Clearly, the details cannot be universal as the study refers to specific language forms (e.g., encoding of arguments through specific case forms, prepositions, etc.). Similarly, the encoding of alternative constructions, as well as syntactic diagnostics like the English Middle alternation are clearly not universal. Yet, it is expected that universal cross-linguistic patterns do exist, insofar as both cross-linguistically recurrent coding properties as well as the availability of certain alternations have a semantic motivation. This has already been anticipated in a work on transitivity alternation by Pinker (1989), whose approach is close to Levinâs, but additionally tries to provide explanations for encoding options in terms of semantic properties of verb classes on the one hand and the semantics of the alternation on the other hand (thus, for example, the middle alternation targets EFFECT verbs like break, not CONTACT verbs like hit).
A different tradition in the research on verb classes (or, valency classes) takes its origin in the work on Case Grammar, different versions of which were developed in the 1970s by Fillmore, Gruber, Cook, J. Anderson, Jackendoff, and Chafe, among others. In this tradition, verb classes are identified in terms of the semantic roles of the verbal arguments. A related approach has been developed (particularly in France and Germany) in the work by Tesnière, Gross, Helbig, and others (see Ăgel 2006 for a comprehensive bibliography of valency research and Herbst & GĂśtz-Votteler 2007 for a representative sample of contemporary approaches to valency research). In this approach, however, verbal valency types are defined more in terms of formal than semantic criteria (see, e.g., Somers 1987 for a comparative treatment). In the subsequent literature, the argument structure of verb classes has played an important role in linguistic theories of different persuasions (see, for instance, Levin & Rappaport Hovav 2005 and Butt 2006 for overviews and discussion), yet this research has rarely been carried out systematically. In the generative literature the issue of subcategorization frames of different verb classes has been present for a long time, but it has not been addressed systematically. Thus, while there is a large literature on individual verb classes found to be of particular theoretical interest (cf. the work by Grimshaw 1990; Pesetsky 1995 and others on emotion verbs, which present challenges for argument linking), comprehensive studies of verb classes in other languages have not been attempted.
While the empirical basis of mainstream generative grammar has not been very broad until recently, other theories like Role and Reference Grammar (Van Valin & Lapolla 1997; Van Valin 2001) and Lexical Decomposition Grammar (Wunderlich 2006; Stiebels 2000) have developed a strong typological orientation. These studies pioneered systematic research into semantic argument types of languages of different alignment (in particular, the work by Van Valin has contributed to the study of argument alternations cross-linguistically). Yet, these studies operate in terms of broad valency-based classes rooted in aspectual properties and lexical decomposition, and never reach the level of granularity of Levinâs classification (see, for instance, the study of verb c...