BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Imitation learning for language generation from unaligned data - G
 erasimos Lampouras\, University of Sheffield
DTSTART:20170210T120000Z
DTEND:20170210T130000Z
UID:TALK69217@talks.cam.ac.uk
CONTACT:Kris Cao
DESCRIPTION:Natural language generation (NLG) is the task of generating na
 tural language from a meaning representation. Rule-based approaches requir
 e domain-specific and manually constructed linguistic resources\, while mo
 st corpus based approaches rely on aligned training data and/or phrase tem
 plates. The latter are needed to restrict the search space for the structu
 red prediction task defined by the unaligned datasets. \n\nIn this talk we
  will discuss the use of imitation learning for structured prediction whic
 h learns an incremental model that handles the large search space while av
 oiding explicitly enumerating it. We will show how we adapted the Locally 
 Optimal Learning to Search (Chang et al.\, 2015) framework which allows us
  to train against non-decomposable loss functions such as the BLEU or ROUG
 E scores while not assuming gold standard alignments. We will show the res
 ults of our evaluation on three datasets using both automatic measures and
  human judgements which achieves results comparable to the state-of-the-ar
 t approaches developed for each of them. Furthermore\, we will present an 
 analysis of the datasets which examines common issues with NLG evaluation.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR