[Ltg] SALS-SIG/LTG Seminar [Joerg Tiedemann 07-04-02, E6A 357, 11am]

[Marc Tilbrook]

----
 SALS-SIG/LTG Seminar
  - see: http://www.clt.mq.edu.au/Events/SALS-SIG.html

  Monday, 2nd April, 2007, 11am
  Macquarie University, E6A, Room 357
 ----


Title: A Genetic Algorithm for Optimizing IR for Question Answering
Speaker: Joerg Tiedemann

Information retrieval (IR) is used in open-domain question answering (QA) to
retrieve relevant passages from a large text collection to narrow down the
search for appropriate answers. In this talk we discuss the  possibilities
of integrating linguistic features produced by NLP tools into standard IR in
order to improve the retrieval performance. The basic idea is to enrich IR
indeces with additional features and to optimize queries derived from
natural language questions to obtain the best IR results for the QA system.
For the optimization we apply a genetic algorithm to perform a probabilistic
search for preferable query settings and keyword selections. Experiments
using questions from recent CLEF QA tasks show a significant improvement of
both IR and QA when using linguistic features and optimized queries for the
retrieval of relevant passages.


short CV:

Joerg Tiedemann is a researcher at the University of Groningen currently
working on Dutch question answering using dependency relations. He received
his PhD in computational linguistics at Uppsala University in Sweden in 2003
(thesis: "Recycling Translations - Extraction of Lexical Data from Parallel
Corpora and their Application in Natural Language Processing"). Earlier has
worked on the alignment of multilingual parallel corpora and machine
translation in various projects. In 2001/2002 he was a visiting researcher
at Edinburgh University and in 2004 he joined the Department of Information
Science in Groningen.


---
Please note: This should not be confused with the HAIL/SALS-SIG seminar the
following day by Dr Gosse Bouma. See
http://www.ict.csiro.au/HAIL/Abstracts/2007/GosseBouma.htm for details
---