A study of notions of participation and discourse in argument structure realisation

Abstract

There is a lack of consensus on the nature of the process of linking; that is, how the conceptual form of thoughts determines the form of the spoken expressions used to describe them. Prominent theories of argument structure realisation include ones based on fixed inventories of participant types (semantic roles, Fillmore, 1968), detailed analyses of the lexical semantics of verbs (predicate decompositions, Jackendoff, 1972; Rappaport and Levin, 1988), characterisations of argument structure positions in terms of prototype semantics (proto-roles, Dowty, 1991), and factors of pragmatics and perspective (Kuno, 1976; Croft, 1991). Given the wide range of views, this study attempts to investigate empirically which semantic and pragmatic factors are the primary drivers of argument structure realisation, without making detailed assumptions about processing models or linguistic representations. The first set of experiments carried out gathered large numbers of acceptability judgements from many non-linguist native-speaker participants over the internet, to quantify the relative acceptability of minimal pairs of diathesis variants. Film script extracts were used to provide authentic materials and contextualisation, and established methodologies from psycholinguistics were employed (Bard et al., 1996; Cowart, 1997). The diatheses examined were the English dative and benefactive constructions; the English and German passive constructions; and the Chinese ba and bei constructions. Effects of aspects of involvement and of discourse structuring were found in all three languages, and crucially the magnitude of these effects was comparable to those elicited by clear syntactic violations. A follow-on corpus study on the British National Corpus gave qualified support to some of these effects. An annotation exercise among linguists found a reasonable degree of consensus on the use of traditional role inventories, but indicated that a coarser grained set of roles might be more appropriate. Overall, the results of these experiments suggested that strict modularity of processing (see e.g. Chomsky, 1986; Bock and Levelt, 1994) may not be a safe assumption; that competition between candidate realisations operates at the clause level; and that the determining factors in argument structure realisation are specific properties of the situation being described in an utterance, rather than default properties of the event or verb in question. On this basis a simple numerical model of argument structure production in English was constructed, integrating elements of Dowty’s proto-roles with the pragmatic notions of information structure and topical saliency. This proof-of-concept model explains much of the linking patterns described in the literature, and provides qualified predictions of judged acceptability. This leads to speculation that such models have the potential to account for argument linking in a range of languages, nominative and otherwise.