Hypertagging: Supertagging for Surface Realization with CCG

Proceedings of ACL-08: HLT, pages 183–191, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics Hypertagging: Supertagging for Surface Realization with CCG Dominic Espinosa and Michael White and Dennis Mehay Department of Linguistics The Ohio State University Columbus, OH, USA {espinosa,mwhite,mehay}@ling.osu.edu Abstract In lexicalized grammatical formalisms, it is possible to separate lexical category assign- ment from the combinatory processes that make use of such categories, such as pars- ing and realization. We adapt techniques from supertagging — a relatively recent tech- nique that performs complex lexical tagging before full parsing (Bangalore and Joshi, 1999; Clark, 2002) — for chart realization in OpenCCG, an open-source NLP toolkit for CCG. We call this approach hypertagging, as it operates at a level “above” the syntax, tag- ging semantic representations with syntactic lexical categories. Our results demonstrate that a hypertagger-informed chart realizer can achieve substantial improvements in realiza- tion speed (being approximately twice as fast) with superior realization quality. 1 Introduction In lexicalized grammatical formalisms such as Lex- icalized Tree Adjoining Grammar (Schabes et al., 1988, LTAG), Combinatory Categorial Grammar (Steedman, 2000, CCG) and Head-Driven Phrase- Structure Grammar (Pollard and Sag, 1994, HPSG), it is possible to separate lexical category assign- ment — the assignment of informative syntactic cat- egories to linguistic objects such as words or lex- ical predicates — from the combinatory processes that make use of such categories — such as pars- ing and surface realization. One way of performing lexical assignment is simply to hypothesize all pos- sible lexical categories and then search for the best combination thereof, as in the CCG parser in (Hock- enmaier, 2003) or the chart realizer in (Carroll and Oepen, 2005). A relatively recent technique for lex- ical category assignment is supertagging (Bangalore and Joshi, 1999), a preprocessing step to parsing that assigns likely categories based on word and part-of- speech (POS) contextual information. Supertagging was dubbed “almost parsing” by these authors, be- cause an oracle supertagger left relatively little work for their parser, while speeding up parse times con- siderably. Supertagging has been more recently ex- tended to a multitagging paradigm in CCG (Clark, 2002; Curran et al., 2006), leading to extremely ef- ficient parsing with state-of-the-art dependency re- covery (Clark and Curran, 2007). We have adapted this multitagging approach to lexical category assignment for realization using the CCG-based natural language toolkit OpenCCG.1 In- stead of basing category assignment on linear word and POS context, however, we predict lexical cat- egories based on contexts within a directed graph structure representing the logical form (LF) of a proposition to be realized. Assigned categories are instantiated in OpenCCG’s chart realizer where, to- gether with a treebank-derived syntactic grammar (Hockenmaier and Steedman, 2007) and a factored language model (Bilmes and Kirchhoff, 2003), they constrain the English word-strings that are chosen to express the LF. We have dubbed this approach hy- pertagging, as it operates at a level “above” the syn- tax, moving from semantic representations to syn- tactic categories. We evaluate this hypertagger in two ways: …

Content truncated to 500 words