BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Geographically Grounded Language Models - Speaker to be confirmed
DTSTART:20240201T110000Z
DTEND:20240201T120000Z
UID:TALK211666@talks.cam.ac.uk
CONTACT:Panagiotis Fytas
DESCRIPTION:Textual data exhibit pronounced variation along geographical d
 imensions (e.g.\, due to dialect differences). Common practices of trainin
 g and deploying language models do not take this inherent dynamicity into 
 account\, leading to detrimental effects for their robustness and performa
 nce on downstream tasks. In this talk\, I will explore how some shortcomin
 gs of text-only NLP pipelines can be alleviated by grounding language mode
 ls in geography. I will first give a brief overview of prior research on g
 eographical variation in NLP. I will then present geoadaptation\, a method
  for geographically grounding language models that combines language model
 ing with geolocation prediction in a multi-task learning setup. Geoadaptat
 ion leads to consistent performance improvements across a range of tasks a
 nd language areas\, especially in zero-shot settings. Finally\, I will sho
 w that the effectiveness of geoadaptation stems from its ability to geogra
 phically retrofit the representation space of language models.
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR