New Directions in Question Answering
AAAI Spring 2003 Symposium
24-26 March 2003
A dream since the first investigations into artificial intelligence has been to converse with a machine in
natural language, in part to get answers to questions. Question answering (QA) promises an important new
way of information access for all, a natural step beyond the keyword query and document retrieval of
today’s web search. Several significant question answering activities currently underway include the
ARDA AQUAINT Program, a TREC QA track, the ARDA NRRC (nrrc.mitre.org) summer 2002
workshops on temporal and multiple perspective question answering, and an LREC 2002 workshop to
develop a question answering roadmap. In spite of many activities, the potential richness of question
answering has still only been partially investigated and includes challenges such as:
-
the heterogeneity of questions and sources (e.g., multilingual, multimedia)
a broad range of (query and document) processing possibilities and answer retrieval
mechanisms
methods for answer extraction, integration, and presentation generation
new application areas (e.g., question answering from manuals and for help desks)
Accordingly, this symposium focuses on new directions in the burgeoning area of question answering.
Question answering is challenging, in part, because it lies at the intersection of several scientific fields
including natural language processing (understanding and generating natural language text), information
retrieval (query formulation, document analysis, relevancy feedback), and human computer interaction
(interface design, user modeling). Figure 1 illustrates the relation of the three areas and their intersection in
systems that support question answering. Several additional scientific disciplines may support question
answering are not shown, such as knowledge representation and reasoning for question and answer
analysis, or recommender technology to find preferred answers, or multimodal information processing to
help extract answers from audio or video sources, or information visualization for results display.
Natural Language
Processing
- question/document analysis
- information extraction
- language generation
- discourse analysis
Question
Answering
Information
Retrieval
- query formulation
- document analysis
- relevancy feedback
Human
Computer
Interaction
- user modeling
- user preferences
- displays
- interaction
Figure 1. Relationship of Question Answering to other Disciplines
Apart from government program reviews and formal government evaluation conferences for question
answering (e.g., TREC), there are no specific forums for addressing advanced research directions in an
open scientific form such as the AAAI Symposium series offers and so we convened a Spring Symposium.
The symposium focuses on bringing together researchers from the range of recent government, industrial,
and academic initiatives in question answering, presenting and demonstrating recent results, and working
together to foster new research directions. The program consists of invited speakers, paper presentations, a
poster session, a panel on web-based question answering, and two roadmapping sessions.
Based on submissions from the US, Europe, Japan, and Egypt, a selected set of paper, posters, and invited
presentations was selected to present new findings in areas including:
-
temporal question answering
multiple perspective question answering
multimedia question answering
multilingual question answering
usability and habitability of question answering systems
re-use in question answering
interactive and/or dialogue based question answering
advances in specific tasks within question answering such as question analysis, information
integration, and answer presentation generation
methods and systems for question answering on the web
evaluation of question answering
The entire group will engage in two sessions to create a roadmap of the future of question answering. The
roadmapping activity will leverage prior work and articulate necessary resources for, impediments to, and
planned or possible future capabilities.
Key Web References
This workshop will draw directly upon these recent programs, activities and resources:
-
ARDA AQUAINT Program - http://www.ic-arda.org/InfoExploit/aquaint/index.html
ARDA Q&A Roadmap – http://www-nlpir.nist.gov/projects/duc/papers/qa.Roadmap-paper_v2.doc
LREC Q&A Roadmap Workshop - http://www.lrec-conf.org/lrec2002/lrec/wksh/QuestionAnswering.html
ARDA NRRC Summer 2002 workshops on reuse in question answering and temporal and multiple perspective
question answering – http://nrrc.mitre.org
TREC QA track - http://trec.nist.gov/presentations/TREC10/qa
Acknowledgements
I would like to thank Paula MacDonald at MITRE for her administrative excellence in supporting the
organization of the workshop and its working notes. I would like to thank the organizing committee for the
energy in evaluating submissions and helping shape the workshop, notably Dr. David Day at MITRE
(day@mitre.org), Dr. John Prange at ARDA (jprange@nsa.gov), Prof. James Pustejovsky at Brandeis
University (jamesp@cs.brandeis.edu) and Prof. Janyce Wiebe at Univ. of Pittsburgh (wiebe@cs.pitt.edu).
I would also like to thank each of the participants, the panelists, and the invited speakers for their excellent
and diverse contributions to the workshop. Thank you for your advancements to question answering.
Mark Maybury
Chelmsford, MA
January 7, 2003