I understand what you said about the problem of circularity, but I think that in
a dialog between persons usually we do not find long distance reference because of
the limit of our short term memory. We can't keep long context in mind. May be
theoretically the problem would exist but practically, I do not think.
Thanks for your input.
Narjčs.
Marco Antonio Esteves da Rocha a écrit :
> Hello, Ruslan, Narjés and all in the list.
>
> I decided I would first post a message to the list as a whole, but in case the
> discussion becomes too specialised so as to be of interest for anaphora
> buffs only, DO complain.
>
> It is a bit risky to think of anaphora resolution as input to topic
> segmentation because the resolution of complex anaphora cases - mostly
> those which cannot be resolved by a plain first-candidate search with
> minor adjustments - often relies on topic tracking in order to select
> preferred antecedents and bypass candidates which would be chosen on a
> strictly recency-and-agreement basis.
>
> I understand Narjés is training an HMM, and thus wants to keep it as
> knowledge poor as possible. But he may find further down on the way
> towards completion of his research that a number of references cannot be
> resolved without the sort of discourse information of which topic tracking
> is a part. The circularity problem (pointed out by Fox (1987)) becomes
> then a nontrivial difficulty. You need anaphora resolution for topic
> segmentation, but you need topic segmentation for anaphora resolution.
>
> I haven't had the chance to test the tools developed by the Wolverhampton
> people and I have no idea whether this circularity problem or any
> consideration of topic tracking matters are part of the processing for
> anaphora resolution using these tools. But I would be careful not to
> assume without further thought that resolving anaphora is an input to
> topic segmentation, even if an HMM is used, and especially in dialogues.
>
> I hope you don't think I am too picky.
>
> Marco Rocha
>
>
> On Fri, 2 Feb 2001, Ruslan Mitkov wrote:
>
> > May I suggest that you visit
> > http://www.wlv.ac.uk/sles/compling/
> > and have a look at our 3 anaphora resolution demos.
> > We could offer you MARS but you will need
> > the FDG shallow parser for it.
> >
> > You will see that members of the Research Group
> > in Computational Linguistics at the University of
> > Wolverhampton have also developed
> > tools for annotating coreference.
> >
> > I hope this helps
> > Ruslan Mitkov
> >
> >
> > -0500, Narjčs Boufaden wrote:
> > >Hello Everyone,
> > >
> > >Can anybody point me to a free anaphora resolution module for English? I
> > >am training an HMM for topic segmentaion for dialogs it is a part of my
> > >PhD thesis. I need this module to resolve references and give them as
> > >input of my system. Any information is appreciate.
> > >
> > >Thanks,
> > >
> > >--
> > >Narjčs Boufaden
> > >Étudiante en doctorat,
> > >RALI, Université de Montréal
> > >
> > >http://www.iro.umontreal.ca/~boufaden
> > >
> > >
> >
> >
> > \\\ Ruslan Mitkov, PhD
> > /// Professor of Computational Linguistics and Language Engineering
> > \\\ School of Humanities, Languages and Social Sciences
> > /// University of Wolverhampton
> > \\\ Stafford St.
> > /// Wolverhampton WV1 1SB
> > \\\ United Kingdom
> > /// Telephone (44-1902) 322471
> > \\\ Fax (44-1902) 322739
> > /// Email R.Mitkov@wlv.ac.uk
> > \\\ Website http://www.wlv.ac.uk/~le1825/
> >
> >
-- Narjčs Boufaden Étudiante en doctorat, RALI, Université de Montréal
This archive was generated by hypermail 2b29 : Fri Feb 02 2001 - 20:15:26 MET