A Generative Lexicon Account of Bangla Complex Predicates

TThis work analyzes the complex predicates of Bangla, a new Indo-Aryan language, using a lexical knowledge representation framework named Generative Lexicon developed by James Pustejovsky. ‘Complex predicates can be defined as predicates which are composed of more than one grbdjfwhhammatical element (either morphemes or words), each of which contributes a non-trivial part of the information of the complex predicate.’ (Alex Alsina, Joan Bresnan and Peter Sells (ed.), 1996) The first element of a complex predicate (henceforth CP) may be a verb, a noun, an adjective or even an adposition (preposition or postposition Dasgupta M 1990, Dasgupta P 2012). Illustrative examples from Bangla are the followings: Noun-Verb Conjunct verb: rag kOra anger do ‘to become angry’ Adjective-Verb conjunct verb: bhalo laga good feel ‘to feel good’ Adposition-Verb complex (Dasgupta Malashree 1990): mone pORa mind-loc fall ‘to remember’ Verb-verb compoud verb: bhore aSa fill come ‘to fill gradually’ Among these CPs, V-V construction, known as a compound verb, has been studied extensively in the literature. Other types of CP, viz., NV and AV constructions, known as conjunct verbs in the literature are relatively less studied. One of the major research questions in the vast literature of complex predicates is where to account them: in syntax or in the lexicon. All the major studies done so far had provided mostly syntactic and some semantic accounts of the complex predicates, whether in dominant Chomskyan generative framework or in some alternative syntactic model like LFG or HPSG. Although a specialized Verb Knowledge Base (VKB) exists which takes care of hierchical arrangement of verbs for English and Hindi developed and maintained by IIT Bombay (refer to the web address at the end), the attempt to characterize the lexical semantic as well as the syntactic properties (argument structure, types of arguments and case marking on those arguments) of these predicates and accommodating them in any kind of specialized lexicon for verbs is not considerably enough. Compound verbs have also been made part of the Wordnet. In the Indo WordNet the compound verbs have been proposed to be stored with a link between the light verb and the main verb it combines with (Paul 2010). Paul argues for enlisting V2s under the main verb with which it can be combined with as compound verbs are lexical variants of the V1. It takes care well of the situation type of the verbal predicate or their event semantics, however, it still does not capture the syntactic information about the arguments of the verbs. For that we need some framework where it is possible to provide both syntactic and semantic information of a predicate. The present study tries to capture the overwhelming and most interesting syntactic and semantic behaviours of the complex predicates in a lexicon taking Pustejovsky’s (1995) Generative Lexicon approach. In South Asian languages, study of complex predicates has been a major research area for several years. A good number of works have been produced in Hindi-Urdu and some other Indian languages on CPs. Hook (1974, 2001), Kachru and Pandharipande (1980), Butt (1993, 1995), Kachru (1993) are most important among them on Hindi Compound verbs. Among other works, works of Mohanan (1995), Davison (2004) and Das (2009) on Hindi N-V complex predicates and by Bhattacharya, Chakrabarti and Sharma (2007) on Wordnet representation of Hindi complex predicates are significant. Bashir (1993) on kalasa, Pandharipande (1993) on Marathi, Fedson (1993) on Tamil, Hook (1996) on Gujrarti, Kaul on Kashmiri (1985) and Rajesh Kumar and Nilu (2012) on Magahi are some of the works in other Indian languages. There is also an extensive and significant work on Bangla compound verb structure (Soma Paul, 2004). Among other theoretical works, Dasgupta (2012) is important for rephrasing the question of complex predicates. However, there is not any elaborate study on Bangla conjunct verbs except some research papers like one by Dasgupta Malasree (1990) and one by the author (Ghosh and Chakraborty (2006). On exclusively A-V complex predicates, there is a real lack of studies in any languages of South Asia before the present author’s work on Bangla (Ghosh 2008). Recently there is also an event-based analysis of complex predicates of Bangla by Basu and Wilbur (2010). Therefore, for theoretical linguistics, this kind of account and analysis of the conjunct verbs is very significant. In corpus annotation, identification of the N-V complex predicates for tagging faces serious problems. Often it creates a confusion to distinguish between a nominal argument of a verbal predicate and the first nominal element of a conjunct verb. There is also some N-V structure where the noun is incorporated within the verb making the whole scenario even more puzzling. Although in writing, native speakers use two words with a gap for these constructions, they are intuitively cognized as one predicate or event. The work sketches a detailed account of the conjunct verbs of Bangla (both AV and NV conjuncts). This kind of corpus-based theoretically grounded study on the conjunct verbs is not available in any South Asian language as far as my knowledge goes. However, there is one paper found by Das et al (2010) on the automatic extraction of Bangla conjunct and compound verbs from two types of corpus with a reasonably good output. This work will help to take up similar studies in related languages which may lead to a typological characterization of the Conjunct Verbs in South Asian languages. One of the significant outputs of the AV and NV complex predicates is characterization and classification of the adjectives and nouns of these conjuncts. This leads to ontology of nouns and adjectives, an extremely important work for both cognitive and computational perspectives. Other outcome of this research is making frames for conjunct and compound verbs. Verb frames are again very significant resources for understanding the argument structure of the verbs as well as their case marking properties. The classification of nouns in this work is based on the argument structure of the nouns resulted from verb frames of the N-V complex predicates. For Compound verbs, the different uses of the verb when used as a vector in the CV have been classified according to their senses.