Automatically Learning to Use Discourse Information For Disambiguation

this paper we discuss how we apply predictions from our plan-based discourse processor discussed in (Rose et al., 1995) to the problem of disambiguation. The work we report here has been done in the context of the Enthusiast Spanish to English translation system. The Enthusiast system is part of the JANUS Speech-to-Speech translation system (Woszcyna et al., 1993; Woszcyna et al., 1994; Suhm et al., 1994; Levin et al., 1995). We discuss and evaluate two different methods for combining context based predictions with non context based predictions, namely a genetic programming approach and a neural network approach. We demonstrate the advantage of incorporating context-based predictions over the purely non context-based approach discussed in (Lavie, 1995). The results presented here show a significant improvement over our previous results reported in (Levin et al., 1995). Introduction In this paper we discuss how we apply predictions from our plan-based discourse processor discussed in (Ros e et al., 1995) to the problem of disambiguation. The work reported here had been carried out in the context of the Enthusiast Spanish to English translation system (Woszcyna et al., 1993; Woszcyna et al., 1994; Suhm et al., 1994; Levin et al., 1995). The Enthusiast System is part of the JANUS Speech-to-Speech translation system Ambiguity is a major problem in a large scale machine translation system such as Enthusiast. This is because the parsing grammar must be large in order to cover the wide range of constructions which speakers use. Additionally, the flexibility of the GLR* skipping parser (Lavie, 1995) we use magnifies the problem. In this paper we Computational Linguistics Program, Philosophy, Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh PA, 15213, cprose@cs.cmu.ed