Joint Satisfaction of Syntactic and Pragmatic Constraints Improves Incremental Spoken Language Understanding

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 514–523, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Joint Satisfaction of Syntactic and Pragmatic Constraints Improves Incremental Spoken Language Understanding Andreas Peldszus University of Potsdam Department for Linguistics peldszus@uni-potsdam.de Okko Buß University of Potsdam Department for Linguistics okko@ling.uni-potsdam.de Timo Baumann University of Hamburg Department for Informatics baumann@informatik.uni-hamburg.de David Schlangen University of Bielefeld Department for Linguistics david.schlangen@uni-bielefeld.de Abstract We present a model of semantic processing of spoken language that (a) is robust against ill-formed input, such as can be expected from automatic speech recognisers, (b) re- spects both syntactic and pragmatic con- straints in the computation of most likely interpretations, (c) uses a principled, ex- pressive semantic representation formalism (RMRS) with a well-defined model the- ory, and (d) works continuously (produc- ing meaning representations on a word- by-word basis, rather than only for full utterances) and incrementally (computing only the additional contribution by the new word, rather than re-computing for the whole utterance-so-far). We show that the joint satisfaction of syn- tactic and pragmatic constraints improves the performance of the NLU component (around 10 % absolute, over a syntax-only baseline). 1 Introduction Incremental processing for spoken dialogue sys- tems (i. e., the processing of user input even while it still may be extended) has received renewed at- tention recently (Aist et al., 2007; Baumann et al., 2009; Buß and Schlangen, 2010; Skantze and Hjalmarsson, 2010; DeVault et al., 2011; Purver et al., 2011). Most of the practical work, how- ever, has so far focussed on realising the poten- tial for generating more responsive system be- haviour through making available processing re- sults earlier (e. g. (Skantze and Schlangen, 2009)), but has otherwise followed a typical pipeline ar- chitecture where processing results are passed only in one direction towards the next module. In this paper, we investigate whether the other potential advantage of incremental processing— providing “higher-level”-feedback to lower-level modules, in order to improve subsequent process- ing of the lower-level module—can be realised as well. Specifically, we experimented with giving a syntactic parser feedback about whether semantic readings of nominal phrases it is in the process of constructing have a denotation in the given con- text or not. Based on the assumption that speak- ers do plan their referring expressions so that they can successfully refer, we use this information to re-rank derivations; this in turn has an influence on how the derivations are expanded, given con- tinued input. As we show in our experiments, for a corpus of realistic dialogue utterances collected in a Wizard-of-Oz setting, this strategy led to an absolute improvement in computing the intended denotation of around 10 % over a baseline (even more using a more permissive metric), both for manually transcribed test data as well as for the output of automatic speech recognition. The remainder of this paper is structured as fol- lows: We discuss related work in the next section, and then describe in …

Content truncated to 500 words