Semantic Role Labeling Improves Incremental Parsing

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 1191–1201, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Semantic Role Labeling Improves Incremental Parsing Ioannis Konstas and Frank Keller Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh {ikonstas,keller}@inf.ed.ac.uk Abstract Incremental parsing is the task of assign- ing a syntactic structure to an input sen- tence as it unfolds word by word. Incre- mental parsing is more difficult than full- sentence parsing, as incomplete input in- creases ambiguity. Intuitively, an incre- mental parser that has access to seman- tic information should be able to reduce ambiguity by ruling out semantically im- plausible analyses, even for incomplete in- put. In this paper, we test this hypothesis by combining an incremental TAG parser with an incremental semantic role labeler in a discriminative framework. We show a substantial improvement in parsing per- formance compared to the baseline parser, both in full-sentence F-score and in incre- mental F-score. 1 Introduction When humans listen to speech, the input becomes available gradually as the speech signal unfolds. Reading happens in a similarly gradual manner when the eyes scan a text. There is good evidence that the human language processor is adapted to this and works incrementally, i.e., computes an in- terpretation for an incoming sentence on a word- by-word basis (Tanenhaus et al., 1995; Altmann and Kamide, 1999). Also language processing systems often deal with speech as it is spoken, or text as it is typed. A dialogue system should start interpreting a sentence while it is spoken, and an information retrieval system should start retrieving results while the user is typing. Incremental processing is therefore essential both for realistic models of human language pro- cessing and for NLP applications that react to user input in real time. In response to this, a number of incremental parsers have been devel- oped, which use context-free grammar (Roark, 2001; Schuler et al., 2010), dependency grammar (Chelba and Jelinek, 2000; Nivre, 2007; Huang and Sagae, 2010), or tree-substitution grammars (Sangati and Keller, 2013). Typical applications of incremental parsers include speech recognition (Chelba and Jelinek, 2000; Roark, 2001; Xu et al., 2002), machine translation (Schwartz et al., 2011; Tan et al., 2011), reading time modeling (Demberg and Keller, 2008), or dialogue systems (Stoness et al., 2004). Incremental parsing, however, is considerably harder than full-sentence parsing: when process- ing the n-th word in a sentence, an, the parser only has access to the left context (words a1 ...an−1); the right context (words an+1 ...aN) is not known yet. This can lead to local ambiguity, i.e., pro- duce additional syntactic analyses that are valid for the sentence prefix, but become invalid as the right context is processed. As an example consider the sentence prefix in (1): (1) a.The athlete realized her goals ... at the competition b. were out of reach The prefix could continue as in (1-a), i.e., as a main clause …

Content truncated to 500 words