BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Overview of: Measuring the non-compositionality of multiword expre
 ssions. [best paper award at COLING] - Laura Rimell (University of Cambrid
 ge)
DTSTART:20101018T113000Z
DTEND:20101018T123000Z
UID:TALK27310@talks.cam.ac.uk
CONTACT:Jimme Jardine
DESCRIPTION:Laura\, recently back from COLING\, will give us a brief summa
 ry of the conference\, and present the paper that won the best paper award
  there.\n\nThe paper is:\nFan Bu and Xiaoyan Zhu. Measuring the non-compos
 itionality of multiword expressions. \n\nAlthough she will not cover them 
 in the talk\, two other papers Laura found interesting are listed below.  
 They may just tickle your paper bits too...\n\nAs always\, please voluntee
 r the dates you would be willing to present a paper or two!!\n\nLater\,\nJ
 imme\n\n-----\n\nFan Bu and Xiaoyan Zhu. Measuring the non-compositionalit
 y of multiword expressions.\n\nMultiword Expressions (MWEs) appear frequen
 tly and grammatically in the natural languages. Identifying MWEs in free t
 exts is a very challenging problem. This paper proposes a knowledge-free\,
  training-free\, and language-independent Multiword Expression Distance (M
 ED). The new metric is derived from an accepted physical principle\, measu
 res the distance from an n-gram to its semantics\, and outperforms other s
 tate-of-the-art methods on MWEs in two applications: question answering an
 d named entity extraction.\n\n\nMark Johnson and Katherine Demuth. Unsuper
 vised phonemic Chinese word segmentation using Adaptor Grammars.\n\nAdapto
 r grammars are a framework for expressing and performing inference over a 
 variety of non-parametric linguistic models. These models currently provid
 e state-of-the-art performance on unsuper- vised word segmentation from ph
 onemic representations of child-directed unseg- mented English utterances.
  This paper in- vestigates the applicability of these mod- els to unsuperv
 ised word segmentation of Mandarin. We investigate a wide vari- ety of dif
 ferent segmentation models\, and show that the best segmentation accuracy 
 is obtained from models that capture inter- word "collocational" \ndepende
 ncies. Sur- prisingly\, enhancing the models to exploit syllable structure
  regularities and to cap- ture tone information does improve over- all wor
 d segmentation accuracy\, perhaps because the information these elements c
 onvey is redundant when compared to the inter-word dependencies.\n\n\nShac
 har Mirkin\, Jonathan Berant\, Ido Dagan\, Eyal Shnarch. Recognising Entai
 lment within Discourse. \n\nTexts are commonly interpreted based on the en
 tire discourse in which they are sit- uated. Discourse processing has been
  shown useful for inference-based applica- tion\; yet\, most systems for t
 extual entail- ment - a generic paradigm for applied in- ference - have on
 ly addressed discourse considerations via off-the-shelf corefer- ence reso
 lvers. In this paper we explore various discourse aspects in entailment in
 - ference\, suggest initial solutions for them and investigate their impac
 t on entailment performance. Our experiments suggest that discourse provid
 es useful\ninforma- tion\, which significantly improves entail- ment infer
 ence\, and should be better ad- dressed by future entailment systems.\n
LOCATION:GS15\, Computer Laboratory
END:VEVENT
END:VCALENDAR