BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Neural text generation from rich semantic representations - Michae
 l Goodman (University of Washington)
DTSTART:20190716T112000Z
DTEND:20190716T113500Z
UID:TALK127492@talks.cam.ac.uk
CONTACT:Guy Edward Toh Emerson
DESCRIPTION:We propose neural models to generate high-quality text from st
 ructured representations based on Minimal Recursion Semantics (MRS). MRS i
 s a rich semantic representation that encodes more precise semantic detail
  than other representations such as Abstract Meaning Representation (AMR).
  We show that a sequence-to-sequence model that maps a linearization of De
 pendency MRS\, a graph-based representation of MRS\, to English text can a
 chieve a BLEU score of 66.11 when trained on gold data. The performance ca
 n be improved further using a high-precision\, broad-coverage grammar-base
 d parser to generate a large silver training corpus\, achieving a final BL
 EU score of 77.17 on the full test set\, and 83.37 on the subset of test d
 ata most closely matching the silver data domain. Our results suggest that
  MRS-based representations are a good choice for applications that need bo
 th structured semantics and the ability to produce natural language text a
 s output.
LOCATION:FW11\, William Gates Building
END:VEVENT
END:VCALENDAR