BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Improving Speech Translation with Linguistically-Informed Represen
 tations - Elizabeth Salesky (Johns Hopkins University)
DTSTART:20201127T120000Z
DTEND:20201127T130000Z
UID:TALK152386@talks.cam.ac.uk
CONTACT:Guy Aglionby
DESCRIPTION:Join Zoom Meeting\nhttps://cl-cam-ac-uk.zoom.us/j/99290674837?
 pwd=cEYvd0pSSXgvN2VERUpmblZ3QzJiZz09\n\nMeeting ID: 992 9067 4837\nPasscod
 e: 999939\n\nEnd-to-end models for speech translation (ST) more tightly co
 uple speech recognition (ASR) and machine translation (MT) than a traditio
 nal cascade of separate ASR and MT models\, with simpler model architectur
 es and the potential for reduced error propagation. However\, several chal
 lenges still remain to make end-to-end models perform as well as cascaded 
 models\, particularly in low-resource scenarios.  Further\, in the move to
 wards more task-agnostic neural architectures\, inductive biases for each 
 task have largely been removed. In this talk\, I will discuss some importa
 nt considerations for building speech translation models (and why we shoul
 d still draw inspiration from cascades)\, as well as three methods to re-i
 ntroduce model biases through phonologically-informed representations and 
 the situations where they are most beneficial.\n\n
LOCATION:Zoom
END:VEVENT
END:VCALENDAR