BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY: Interactive and decomposed approaches for NLP: the case of multi-
 text summarization - Ido Dagan (Bar-Ilan University)
DTSTART:20220429T110000Z
DTEND:20220429T120000Z
UID:TALK173372@talks.cam.ac.uk
CONTACT:Michael Schlichtkrull
DESCRIPTION:Current approaches for NLP tasks often conform to two design p
 rinciples. First\, they address “static” tasks\, where a single input 
 instance is addressed at a time\, independently of other inputs. Second\, 
 outputs are computed via an end-to-end model\, trained directly over input
 -output pairs for the task. In this talk\, I will propose two directions i
 n which NLP research may be systematically extended beyond the static end-
 to-end approach and demonstrate them for the use case of multi-text summar
 ization. In the first part of the talk I suggest that in many realistic us
 e cases multi-text (or long-text) summarization should support an interact
 ive setting\, where users interactively direct summary generation to best 
 fit their information exploration needs.  To promote principled research i
 n this direction\, we propose a systematic evaluation framework for intera
 ctive summarization. This framework  extends summarization evaluation stan
 dards to consider the accumulating information along a user session\, and 
 includes an effective procedure for collecting user sessions. We then pres
 ent a deep reinforcement learning model for interactive summarization\, sh
 owing (using our evaluation framework) that it significantly improves info
 rmation exposure over prior baselines while preserving positive user exper
 ience.\nIn the second part of the talk I suggest that summarization modeli
 ng may be beneficially decomposed to inherent subtasks\, each addressed by
  a targeted model\, rather than employing a single end-to-end model. Such 
 decomposition is enabled through a clever generation of targeted training 
 datasets for specific subtasks\, all derived from the original “end-to-e
 nd” training data. As an additional contribution related to this context
 \, I will describe our Cross-Document Language Model (CDLM)\, which is pre
 -trained specifically to model cross-text relationships\, supporting diver
 se cross-document tasks.\n\nBio:\n\nIdo Dagan is a Professor at the Depart
 ment of Computer Science at Bar-Ilan University\, Israel\, the founder of 
 the Natural Language Processing (NLP) Lab at Bar-Ilan\, the founding Direc
 tor of the nationally funded Bar-Ilan University Data Science Institute\, 
 and a Fellow of the Association for Computational Linguistics (ACL). His i
 nterests are in applied semantic processing\, focusing on textual inferenc
 e\, natural open semantic representations\, consolidation and summarizatio
 n of multi-text information\, and interactive text summarization and explo
 ration. Dagan and colleagues initiated and promoted textual entailment rec
 ognition (RTE\, later aka NLI) as a generic empirical task. He was the Pre
 sident of the ACL in 2010 and served on its Executive Committee during 200
 8-2011. In that capacity\, he led the establishment of the journal Transac
 tions of the Association for Computational Linguistics\, which became one 
 of two premiere journals in NLP. Dagan received his B.A. summa cum laude a
 nd his Ph.D. (1992) in Computer Science from the Technion. He was a resear
 ch fellow at the IBM Haifa Scientific Center (1991) and a Member of Techni
 cal Staff at AT&T Bell Laboratories (1992-1994). During 1998-2003 he was c
 o-founder and CTO of FocusEngine and VP of Technology of LingoMotors\, and
  has been regularly consulting in the industry. His academic research has 
 involved extensive industrial collaboration\, including funds from IBM\, G
 oogle\, Thomson-Reuters\, Bloomberg\, Intel and Facebook\, as well as coll
 aboration with local companies under funded projects of the Israel Innovat
 ion Authority.\n\nTopic: NLIP Seminar\nTime: Apr 29\, 2022 12:00 PM London
 \n\nJoin Zoom Meeting\nhttps://cl-cam-ac-uk.zoom.us/j/96419914999?pwd=RHN4
 TE9KMmdhY3loaE55bHRNTVFodz09\n\nMeeting ID: 964 1991 4999\nPasscode: 48587
 8
LOCATION:Virtual (Zoom)
END:VEVENT
END:VCALENDAR