A Hybrid Approach for QA Track Definitional Questions

Abstract

This paper presents an overview of DefScriber, a hybrid goal-driven and data-driven system for definitional questions that was developed at Columbia University. DefScriber combines knowledge-based and statistical methods to answer definitional questions of the form, "What is X?" The authors explain how the system was modified and applied to answer definitional questions in the TREC 2003 QA track. They present DefScriber's results on the definitional questions, which were significantly above median, achieving an average F-score of .338 compared with the median of .192. Finally, they analyze their scores on certain individual questions, discussing areas in which their system performed well and areas in which it could be improved.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 2006
Accession Number
ADA456354

Entities

People

  • Andrew H. Schlaikjer
  • Kathleen R. Mckeown
  • Sasha Blair-goldensohn

Organizations

  • Columbia University

Tags

Communities of Interest

  • Autonomy

DTIC Thesaurus Topics

  • Abstracts
  • Automated Text Summarization
  • Cohesion
  • Computer Science
  • Data Sets
  • Errors
  • Extraction
  • Identification
  • Judgment
  • Language
  • Learning
  • Machine Learning
  • Natural Languages
  • Pattern Recognition
  • Precision
  • Test And Evaluation
  • Universities

Readers

  • Artificial Intelligence
  • Information Retrieval