BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Recurrent Continuous Translation Models - Nal Kalchbrenner (Univer
 sity of Oxford)
DTSTART:20131007T110000Z
DTEND:20131007T120000Z
UID:TALK47008@talks.cam.ac.uk
CONTACT:Rogier van Dalen
DESCRIPTION:Deep learning methods are well-suited for constructing distrib
 uted\, continuous representations for linguistic units ranging from charac
 ters to sentences. These learnt representations come with an inherent\, ta
 sk-dependent notion of similarity that allows the models to overcome spars
 ity issues and to generalise well beyond the training domain.\nIn this tal
 k we extend these methods to the problem of machine translation and introd
 uce a class of probabilistic translation models (RCTMs) that rely purely o
 n continuous representations of the source and target sentences. We explor
 e several model architectures and we see that the models obtain translatio
 n perplexities that are significantly lower than those of state-of-the-art
  alignment-based translation models. We also investigate the models' abili
 ty to generate translations directly and solely from the underlying contin
 uous space.\n\n*Bio*\n\nNal is a second-year PhD student in the Computatio
 nal Linguistics and Quantum groups at Oxford. Before joining Oxford\, he s
 tudied CS\, maths and logic at the ILLC and at Stanford. He is a recipient
  of the Clarendon fellowship.
LOCATION:Department of Engineering - LR12
END:VEVENT
END:VCALENDAR