BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Long Form Question Answering - Angela Fan (Facebook)
DTSTART:20191206T120000Z
DTEND:20191206T130000Z
UID:TALK134101@talks.cam.ac.uk
CONTACT:James Thorne
DESCRIPTION:We will discuss long-form question answering\, a task requirin
 g elaborate and in-depth answers to open-ended questions. The dataset comp
 rises 270K threads from the Reddit forum ``Explain Like I'm Five'' (ELI5) 
 where an online community provides answers to questions which are comprehe
 nsible by five year olds. Compared to existing datasets\, ELI5 comprises d
 iverse questions requiring multi-sentence answers. We provide a large set 
 of web documents to help answer the question. Automatic and human evaluati
 ons show that an abstractive model trained with a multi-task objective out
 performs conventional Seq2Seq\, language modeling\, as well as a strong ex
 tractive baseline. However\, our best model is still far from human perfor
 mance since raters prefer gold responses in over 86% of cases\, leaving am
 ple opportunity for future improvement. In subsequent work\, we propose co
 nstructing a local graph structured knowledge base for each query\, which 
 compresses the web search information and reduces redundancy. We show that
  by linearizing the graph into a structured input sequence\, models can en
 code the graph representations within a standard Sequence-to-Sequence sett
 ing. We apply this approach to long form question answering. By feeding gr
 aph representations as input\, we can achieve better performance than usin
 g retrieved text portions.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR