Global Features for Shallow Discourse Parsing
Sucheta Ghosh, Giuseppe Riccardi, Richard Johansson · 2012
A coherently related group of sentences may be referred to as a discourse. In this paper we address the problem of parsing coherence relations as defined in the Penn Discourse Tree Bank (PDTB). A good model for discourse structure analysis needs to account both for local dependencies at the token-level and for global dependencies and statistics. We present techniques on using inter-sentential or sentence-level (global), data-driven, nongrammatical features in the task of parsing discourse. The parser model follows up previous approach based on using tokenlevel (local) features with conditional random fields for shallow discourse parsing, which is lacking in structural knowledge of discourse. The parser adopts a twostage approach where first the local constraints are applied and then global constraints are used on a reduced weighted search space (n-best). In the latter stage we experiment with different rerankers trained on the first stage n-best parses, which are generated using lexico-syntactic local features. The two-stage parser yields significant improvements over the best performing model of discourse parser on the PDTB corpus. 1