What are Little Texts Made Of? A Structural and Presuppositional Account Using Lexicalised TAG

In a series of recent papers (Cristea and Webber, 1997; Webber and Joshi, 1998; Webber et al., 1999a; Webber et al., 1999b), we have argued for starting the analysis of discourse in the same way as one starts the analysis of a clause, looking at how its syntax and semantics project from the lexicon. This is complementary to the issue of discourse pragmatics – how these small syntactic units of discourse are used in achieving communicative intentions – and to other discourse processes that provide additional organisational overlays on these units. Intuitively, these small units of discourse correspond to multi-clause descriptions of entities, including individuals, sets, eventualities, situations, etc. These descriptions can involve both first-order and higher-order predicate-argument relations and modal operators, forming the content of communicative intentions. A key feature of our analysis is that semantic discourse relations are associated with both syntactic structures and anaphoric links, and that the properties of the two are (not surprisingly) different. Together, they allow more complex semantics to be conveyed through simpler structure. For English, we have given linguistic evidence for our analysis in terms of (a) the similar behavior of intra-sentential clausal connectives (i.e., subordinate conjunctions) and inter-sentential connectives (i.e., parallel structures cued by “Not only . . . But also . . . ”, “On the one hand, . . . On the other hand, . . . ”, etc.); (b) the similar behavior of “nominal parataxis” in English (i.e., noun-noun modifiers) and clausal parataxis; and (c) the ability of the analysis to explain the presence of multiple discourse connectives in a clause. In contrast with Rhetorical Structure Theory (Mann and Thompson, 1988), what the current approach offers is a decoupling of discourse semantics from discourse syntax, rather than what is essentially a “semantic grammar” for discourse – i.e., one that merges syntactic-semantic patterns into a single description, with re-write rules such as