Inference and Computational Semantics Patrick Blackburn (patrick@aplog.org), Johan Bos (bos@coli.uni-sb.de), Michael Kohlhase (kohlhase@cs.uni-sb.de) and Hans de Nivelle (nivelle@mpi-sb.mpg.de) Universit¨at des Saarlandes, Saarbr¨ ucken, Germany Abstract. This paper discusses inference in computational semantics. We argue that state-of-the-art methods in first-order theorem proving and model generation are of direct relevance to inference for natural language processing. We support our claim by discussing our implementation of van der Sandt’s presupposition projec- tion algorithm in Discourse Representation Theory, an approach which demands sustained use of powerful inference mechanisms. Keywords: Computational Semantics, Theorem Proving, Model Generation, Pre- supposition Projection, Discourse Representation Theory 1. Introduction In this paper we discuss inference in computational semantics. In par- ticular, we argue that state-of-the-art methods in first-order theorem proving and model generation are of direct relevance to inference for natural language processing. This claim is based on our experience of implementing van der Sandt’s approach to presupposition, and much of the paper discusses this application. Incidentally, the reader can experiment with this implementation over the internet: most of what is discussed below is available as part of Johan Bos’s DORIS system (Discourse Oriented Representation and Inference System 1 ). This work has its roots in a textbook entitled Representation and Inference in Natural Language: A First Course in Computational Se- mantics (see Blackburn and Bos (2000a) for the latest draft). The goal of this book is straightforward: to present formal semantics from a computational perspective, and equip students with the basic tools required to perform semantic construction computationally. Modular- ity, reusability, and the use of standard tools is emphasized. Now, as far as representation is concerned, it is more or less clear what an introduction to computational semantics should offer: it is obviously sensible to introduce standard semantic representation formalisms such as Discourse Representation Theory (DRT Kamp and Reyle (1993)), to discuss well-known techniques for handling scope ambiguities, and so on. But inference is far harder to pin down. What exactly is inference in computational semantics? c  2005 Kluwer Academic Publishers. Printed in the Netherlands. tilburg.tex; 9/04/2005; 20:42; p.1