[Corpora] [Corpora-List] Phrase similarity

Yannick Versley versley at cl.uni-heidelberg.de
Tue Nov 11 09:18:44 UTC 2014


Alexander,

take_care and give_up should have individual entries in a resource such as
WordNet,
arguably they are more multiword items (single unit of meaning) than
phrases (multiple
meaning-bearing units combined). In that case, neither Mitchell&Lapata nor
ADW will
do you any good.

Handling of multiword items is simple enough to build into the usual
approaches for
distributional similarity by treating them as one item, e.g. word2vec can
use a list
of fixed phrases. (With verb particles or light verbs, it's slightly more
complicated if you want to
turn "He took great care of his aunt" into "He take_care great of his
aunt", also the
modifier 'great' is left dangling when you pretend the light verb
construction is really
one verb).

Best,
Yannick

On Tue, Nov 11, 2014 at 8:54 AM, Alexander Osherenko <osherenko at gmx.de>
wrote:

> Hi all,
>
> there are many approaches to measure words' similarity, for example,
> "donkey-good" using Wordnet. I wonder, is there also an approach to measure
> phrase similarity, for intance, "take care-give up"?
>
> ​Alexander​
>
> --
> Alexander Osherenko, Dr. rer. nat.
> Senior HCI architect
> Homepage at Humboldt-Universität zu Berlin
> <http://www.hu-berlin.de/~osherena/>
> SlimAC <http://www.slimac.de/>
> Humboldt Innovation <http://www.humboldt-innovation.de/>
>
> Founder and R&D
> Homepage at Socioware Development
> <http://www.socioware.de/osherenko_page.html>
>
> Channel: Youtube <https://www.youtube.com/user/MrOsherenko>
> Channel: Twitter <https://twitter.com/mrosherenko>
>
> Standort Adlershof:
> Spin-Off-ZONE Adlershof
> Wegedornstr. 32
> 12524 Berlin
>
> Standort Mitte:
> Ziegelstr. 30
> 10117 Berlin
>
> _______________________________________________
> UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
> Corpora mailing list
> Corpora at uib.no
> http://mailman.uib.no/listinfo/corpora
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20141111/ea41d52f/attachment.htm>
-------------- next part --------------
_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora


More information about the Corpora mailing list