Hybrid Parsing: Using Probabilistic Models as Predictors for a Symbolic Parser

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 321–328, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Hybrid Parsing: Using Probabilistic Models as Predictors for a Symbolic Parser Kilian A. Foth, Wolfgang Menzel Department of Informatics Universit¨at Hamburg, Germany {foth|menzel}@informatik.uni-hamburg.de Abstract In this paper we investigate the benefit of stochastic predictor components for the parsing quality which can be obtained with a rule-based dependency grammar. By in- cluding a chunker, a supertagger, a PP at- tacher, and a fast probabilistic parser we were able to improve upon the baseline by 3.2%, bringing the overall labelled accu- racy to 91.1% on the German NEGRA cor- pus. We attribute the successful integra- tion to the ability of the underlying gram- mar model to combine uncertain evidence in a soft manner, thus avoiding the prob- lem of error propagation. 1 Introduction There seems to be an upper limit for the level of quality that can be achieved by a parser if it is confined to information drawn from a single source. Stochastic parsers for English trained on the Penn Treebank have peaked their performance around 90% (Charniak, 2000). Parsing of German seems to be even harder and parsers trained on the NEGRA corpus or an enriched version of it still perform considerably worse. On the other hand, a great number of shallow components like tag- gers, chunkers, supertaggers, as well as general or specialized attachment predictors have been devel- oped that might provide additional information to further improve the quality of a parser’s output, as long as their contributions are in some sense com- plementory. Despite these prospects, such possi- bilities have rarely been investigated so far. To estimate the degree to which the desired syn- ergy between heterogeneous knowledge sources can be achieved, we have established an exper- imental framework for syntactic analysis which allows us to plug in a wide variety of external predictor components, and to integrate their con- tributions as additional evidence in the general decision-making on the optimal structural inter- pretation. We refer to this approach as hybrid pars- ing because it combines different kinds of linguis- tic models, which have been acquired in totally different ways, ranging from manually compiled rule sets to statistically trained components. In this paper we investigate the benefit of ex- ternal predictor components for the parsing qual- ity which can be obtained with a rule-based gram- mar. For that purpose we trained a range of predic- tor components and integrated their output into the parser by means of soft constraints. Accordingly, the goal of our research was not to extensively op- timize the predictor components themselves, but to quantify their contribution to the overall pars- ing quality. The results of these experiments not only lead to a better understanding of the utility of the different knowledge sources, but also allow us to derive empirically based priorities for fur- ther improving them. We are able to show that the potential of WCDG for information …

Content truncated to 500 words