BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A Graph-Based Framework for Structured Prediction Tasks in Sanskri
 t - Amrith Krishna (University of Cambridge)
DTSTART:20210205T130000Z
DTEND:20210205T140000Z
UID:TALK156112@talks.cam.ac.uk
CONTACT:James Thorne
DESCRIPTION:Join Zoom Meeting\nhttps://cl-cam-ac-uk.zoom.us/j/96198886046?
 pwd=Ui8rRG1UTkZtdVQyZSswcWN6T0hVUT09\n\nMeeting ID: 961 9888 6046\nPasscod
 e: 695236\n\nWe propose a framework using Energy-Based Models for multiple
  structured prediction tasks in Morphologically rich free-word order langu
 ages\, with a focus Sanskrit. Ours is an arc-factored model\, similar to t
 he graph-based parsing approaches\, and we consider the tasks of word-segm
 entation\, morphological parsing\, dependency parsing\, syntactic linearis
 ation and prosodification\, a prosody level task we introduce in this work
 . Ours is a search based structured prediction framework\, which  expects 
 a graph as input\, where relevant linguistic information is encoded in the
  nodes\,  and the edges are then used to indicate the association between 
 these nodes. Typically the state of the art models for morphosyntactic tas
 ks in morphologically rich languages still rely on hand-crafted features f
 or their performance. But here\, we automate the learning of the feature f
 unction. The feature function so learnt along with the search space we con
 struct\, encode relevant linguistic information for the tasks we consider.
  This enables us to substantially reduce the training data requirements to
  as low as 10 \\% as compared to the data requirements for the neural stat
 e of the art models. Our experiments in Czech and Sanskrit show the langua
 ge agnostic nature of the framework\, where we train highly competitive mo
 dels for both the languages. Moreover\, our framework enables to incorpora
 te language specific constraints to prune the search space and to filter t
 he candidates during inference. We obtain significant improvements in morp
 hosyntactic tasks for Sanskrit by incorporating language specific constrai
 nts into the model. In all the tasks we discuss for Sanskrit\, we either a
 chieve state of the art results or ours is the only data driven solution f
 or those tasks.  \n
LOCATION:Virtual (Zoom)
END:VEVENT
END:VCALENDAR