BEGIN:VCALENDAR
VERSION:2.0
PRODID:ILLC Website
X-WR-TIMEZONE:Europe/Amsterdam
BEGIN:VTIMEZONE
TZID:Europe/Amsterdam
X-LIC-LOCATION:Europe/Amsterdam
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:/NewsandEvents/Archives/2022/newsitem/13813/30
 -August-2022-Computational-Linguistics-Seminar-Alb
 erto-Testoni
DTSTAMP:20220901T143925
SUMMARY:Computational Linguistics Seminar, Alberto
  Testoni
ATTENDEE;ROLE=Speaker:Alberto Testoni (University 
 of Trento)
DTSTART;TZID=Europe/Amsterdam:20220830T160000
LOCATION:Room B0.204, Science Park 904, Amsterdam 
 / Online (Zoom)
DESCRIPTION:Recent years have witnessed an explosi
 on of NLP models for many different tasks, both in
  text-only and multimodal (vision & language) sett
 ings. Impressive results have been obtained on mul
 timodal encoders, whereas decoders have received l
 ess attention. In my work, I focus on the latter a
 iming to study the problem-solving reasoning behin
 d natural language generation. To this end, I take
  referential grounded dialogue games as a testbed.
  I will discuss the main issues affecting generati
 ve systems and explore how the weaknesses of the e
 ncoder affect the choice of the decoder by focusin
 g on the interpretation of negatively answered que
 stions. I will then present a cognitively-inspired
  re-ranking decoding strategy for promoting the ge
 neration of strategic questions. I will compare th
 is strategy to a wide variety of different decodin
 g algorithms proposed in the literature, together 
 with an in-depth analysis of their hyper-parameter
  configurations. Finally, I will briefly mention s
 ome ongoing works on exploring how modeling human 
 uncertainty can lead to better natural language ge
 neration systems and an investigation of pragmatic
  phenomena that allow humans to efficiently solve 
 referential games.
X-ALT-DESC;FMTTYPE=text/html:\n  <p>Recent years h
 ave witnessed an explosion of NLP models for many 
 different tasks, both in text-only and multimodal 
 (vision & language) settings. Impressive results h
 ave been obtained on multimodal encoders, whereas 
 decoders have received less attention. In my work,
  I focus on the latter aiming to study the problem
 -solving reasoning behind natural language generat
 ion. To this end, I take referential grounded dial
 ogue games as a testbed. I will discuss the main i
 ssues affecting generative systems and explore how
  the weaknesses of the encoder affect the choice o
 f the decoder by focusing on the interpretation of
  negatively answered questions. I will then presen
 t a cognitively-inspired re-ranking decoding strat
 egy for promoting the generation of strategic ques
 tions. I will compare this strategy to a wide vari
 ety of different decoding algorithms proposed in t
 he literature, together with an in-depth analysis 
 of their hyper-parameter configurations. Finally, 
 I will briefly mention some ongoing works on explo
 ring how modeling human uncertainty can lead to be
 tter natural language generation systems and an in
 vestigation of pragmatic phenomena that allow huma
 ns to efficiently solve referential games.</p>\n
URL:https://projects.illc.uva.nl/LaCo/CLS/
END:VEVENT
END:VCALENDAR