AT deleted A_natural_question_i.tex  almost 9 years ago

Commit id: 7aec0c2f2f1a120ee94ff345bb8cef1b86ed1f54

deletions | additions      

         

A natural question is whether we could simply apply our YodaQA system  on this task. However, while there are some common analysis elements,  generating factoid answers is programatically quite a different task  from deciding whether a sentence is correct or incorrect,%  \footnote{A baseline support for answering yes / no questions is currently  under development within YodaQA as a student internship project, but it is  not finished yet and not meant to be highly reliable at this point.  The initial proposed method is at  \url{https://github.com/brmson/yodaqa/blob/f/textent/doc/ENTAILMENT.md}  but the difference from our scenario is that we are answering questions  on top of a fixed knowledge base, mainly structured one, rather than an  evolving news stream. The baseline techniques are also so simple that  reusing implementation wouldn't help that much --- but reusing experience  definitely will.}  as we explain in more detail below.  We believe the most efficient approach to take for a prototype is thus  to build a system from scratch, reusing mainly our expertise and possibly  some isolated external components (e.g.\ database interfaces or NLP annotators).         

.tex  The_task_of_the.tex  Our_team_at_the.tex  A_natural_question_i.tex  The_rest_of_the.tex  sectionQA_System_in_.tex  When_building_an_NLP.tex