• Seminars & Events
  • Latest
  • Archive
  • RSS Feed


2 June 2011

Extending the logical coverage of Referring Expressions Generation

Location: Meeting Room 10, 2nd Floor, JLB
Time: 12:30pm - 13:45pm
Speaker(s): Dr Kees van Deemter -University of Aberdeen, Computing Science Department

Referring Expressions Generation (REG) is a lively area of Computational Linguistics, devoted to the computer-based generation of English descriptions (like ``the brown dog’’, ``the large green sofa’’) which serve to identify a target referent. Consistent with current trends in Computational Linguistics, existing work on REG is extremely cautious in its use of logic. For example, REG programs start from a Knowledge Base consisting of (almost) purely atomic information, and the descriptions generated by these programs are usually limited to conjunctions of atomic statements without constants. The present talk will explore what happens if the full power of Knowledge Representation is unleashed on REG, by using fragments of Description Logic. REG extensions addressed will include generic rules (e.g., ``every person has 1 father’’), quantified descriptions (e.g., ``the woman with 12 children’’), and proper names (``the author of Waverley’’). We shall also discuss the general lack of interaction between Computational Linguists and researchers in Knowledge Representation, exploring the reasons behind this separation and the prospects for closer interaction. Parts of this talk describes joint work with Yuan Ren and Jeff Pan. Bio: I am an academic working in Computational Linguistics, the area of Artificial Intelligence where Computer Science meets Linguistics. My main areas of expertise are Natural Language Generation and (to a smaller extent) Computational Semantics. I take a lively interest in logical and philosophical issues arising in these areas. More recently I've started to work closely with psycholinguists interested in algorithmic models of language production. My work centers around the question how information can be expressed in a way that is most suitable for human readers/hearers. Examples include the use of computers for generating text from the numbers, facts or formulas in a knowledge base, and the design of Embodied Conversational Agents. Some of this work is supported by the EPSRC Platform Grant “Affecting People with Natural Language” which supports Aberdeen’s Natural Language Generation group. One of my main research interests is the generation of  referring expressions, as when we program a computer to refer to 'your last email to University Registry', or to 'the icon at the top-left of your screen'. Other interests include ambiguity and vagueness: central phenomena in ordinary language whose usefulness -- and whose dangers -- are still incompletely understood. Ambiguity was the topic of the collection "Semantic Ambiguity and Underspecification" (CSLI Lecture Notes 1996). Vagueness is the focus of my recent book "Not Exactly: in Praise of Vagueness", Oxford University Press 2010. For reviews and background, see http://www.csd.abdn.ac.uk/~kvdeemte/NotExactly

Save to your Calendar

Contact: mailto:r.power@open.ac.uk