Reducing Grounded Learning Tasks To Grammatical Inference

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1416–1425, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Reducing Grounded Learning Tasks to Grammatical Inference Benjamin Börschinger Department of Computing Macquarie University Sydney, Australia benjamin.borschinger@mq.edu.au Bevan K. Jones School of Informatics University of Edinburgh Edinburgh, UK b.k.jones@sms.ed.ac.uk Mark Johnson Department of Computing Macquarie University Sydney, Australia mark.johnson@mq.edu.au Abstract It is often assumed that ‘grounded’ learning tasks are beyond the scope of grammatical in- ference techniques. In this paper, we show that the grounded task of learning a seman- tic parser from ambiguous training data as dis- cussed in Kim and Mooney (2010) can be re- duced to a Probabilistic Context-Free Gram- mar learning task in a way that gives state of the art results. We further show that ad- ditionally letting our model learn the lan- guage’s canonical word order improves its performance and leads to the highest seman- tic parsing f-scores previously reported in the literature.1 1 Introduction One of the most fundamental ideas about language is that we use it to express our thoughts. Learning a natural language, then, amounts to (at least) learning a mapping between the things we utter and the things we think, and can therefore be seen as the task of learning a semantic parser, i.e. something that maps natural language expressions such as sentences into meaning representations such as logical forms. Ob- viously, this learning can neither take place in a fully supervised nor in a fully unsupervised fashion: the learner does not ‘hear’ the meanings of the sentences she observes, but she is also not treating them as merely meaningless strings. Rather, it seems plau- sible to assume that she uses extra-linguistic context 1The source code used for our experiments and the evalua- tion is available as supplementary material for this article. to assign certain meanings to the linguistic input she is confronted with. In this sense, learning a semantic parser seems to go beyond the well-studied task of unsupervised grammar induction. It involves not only learning a grammar for the form-side of language, i.e. lan- guage expressions such as sentences, but also the ‘grounding’ of this structure in meaning represen- tations. It requires going beyond the mere linguistic input to incorporate, for example, perceptual infor- mation that provides a clue to the meaning of the ob- served forms. Essentially, it seems as if ‘grounded’ learning tasks like this require dealing with two different kinds of information, the purely formal (phonemic) and meaningful (semantic) aspects of language. Grammatical inference seems to be lim- ited to dealing with one level of formal information (Chang and Maia, 2001). For this reason, probably, approaches to the task of learning a semantic parser employ a variety of sophisticated and task-specific techniques that go beyond (but often elaborate on) the techniques used for grammatical inference (Lu et al., 2008; Chen and Mooney, 2008; Liang et al., 2009; Kim and Mooney, 2010; Chen et al., 2010). In this paper, we show …

Content truncated to 500 words