Language and Reasoning for Question Answering:
State of the Art and Future Directions
Farah Benamara
Institut de Recherche en Informatique de Toulouse
118, route de Narbonne, 31062 Toulouse, France
benamara@irit.fr
Abstract
The research community states that QA
systems should support the integration of
deeper modes of language understand-
ing as well as more elaborated reason-
ing schemas in order to boost the perfor-
mances of current QA systems as well as
the quality and the relevance of the pro-
duced answers.
Depending on the complexity of the ques-
tion and the associated passages, more or
less complex strategies can be used, such
as :
• deep semantic analysis of NL ques-
tions such as anaphora resolutions,
• context and ambiguity detection,
• responses to unanticipated questions
or to resolve situations in which no
answer is found in the data sources,
• models for answer completeness,
• dialogueandinteractiveQAscenario,
• models for answer fusion from differ-
ent sources, etc.
We focus in this talk on the role of rea-
soning in a QA process by answering the
following questions: what kind of reason-
ing capabilities can be used ? on what kind
of resources can they be built on ? at what
extend they can be used when developing
realistic systems ?
Based on the synthesis of the current state
of the art, the second part of the talk de-
scribes a general typology of inference at
different levels: deep semantic analysis of
both NL questions and passages, textual
entailment, pragmatic inference and con-
text detection, etc. The related formalisms
as well as a description of some current
QA systems based on these techniques are
also presented.
1 KRAQ06
