[Corpora-List] Syntactic parsing performance by humans?

chris brew cbrew at acm.org
Sat May 14 10:20:59 CEST 2016

For the relationship between CFGs and dependency grammars, there is a nice piece of math by Steve Abney http://www.vinartus.net/spa/94g.pdf

The upshot is that dependency grammar is definitely NOT just a notational variant of CFG.

The paper is unpublished, but well worth a read. I particularly commend the following closing section:

In closing, it is perhaps appropriate to discuss the import of the result.
>From a mathematical perspective, equivalences between types of
representation provide useful tools. Some results are easier to discover or prove using one type of representation; some are easier to discover or prove using the other. The equivalence allows one to use either type of

representation interchangeably. Judging from the literature, the linguist’s attitude tends to be rather different. . If one shows that two representations are equivalent in some respect, the immediate response is to seek other arguments for why one or the other representation is right

[Several paragraphs of commentary on possible reasons for this attitude, followed by:]

If the result reported here is of linguistic interest, it is because some constraints on syntactic structures are most easily established using projection-dependency trees, and others are most easily established using characteristic trees. The result does not mean that phrase-structure trees are right and dependency trees are wrong. Indeed, if one is a true

description of a sentence, the other is ipso facto true; though they may be differentially useful in different contexts.

On 14 May 2016 at 09:03, Koos Wilt <kooswilt at gmail.com> wrote:

> Rodolfo raises interesting questions. But to some extent we are all
> prisoners of the prevailing fashion,and the prevailing fashion features
> dependency parsing. It may console him and other to know some claim
> dependency notation is merely a notational variant of context-free grammars
> + tree bank stuff. As for the empty categories. Traditional grammar teaches
> language students about 'infinitival complements'. In *John wants to go
> home* there is no PRO. But there are alternative ways to express *John*
> is in many ways like a subject of the infinitival clause. (The projection
> principle in GB). If you want to do a project in 2016 you are stuck with
> dependency parsers. And yes, things like object or subject are semantic.
> But I don't see a problem with that. A principle that applies to NLP is: be
> pragmatic.
> 2016-05-14 9:58 GMT+02:00 Rodolfo Delmonte <delmont at unive.it>:
>> I don’t find dependency parsing useful at all seen its inherent
>> shortcomings in relation with the problem of using its representation to
>> produce a semantic interpretation that is coherent and complete. This is
>> first of all due to the decision to leave out empty categories, which makes
>> it suitable for languages like English, but a lot less suitable for
>> languages like Chinese or Italian and other Romance languages.
>> However, there’s a number of other shortcomings you are certainly aware
>> of that have to do with treatment of relative and interrogative pronouns.
>> Here below the snapshot of the parse of a sentence containing a pied-piping
>> structure which is heavily wrong in many ways as performed by MATE, the
>> best dependency parser in the survey just posted in this thread.
>> I include also the parser of corresponding non embedded sentence to show
>> where dependencies should be, semantically speaking.
>> Rodolfo
>> --
>> Rodolfo Delmonte Ph.D.
>> Associate Professor of Computational Linguistics
>> Director of Laboratory of Computational Linguistics
>> Department of Linguistic Studies
>> Ca' Bembo - Dorsoduro 1075
>> 30123 VENEZIA (Italy)
>> website: http//project.cgm.unive.it
>> On 14 May 2016, at 04:16, Koos Wilt <kooswilt at gmail.com> wrote:
>> I atempted to post the bibliography twice, but it bounced because of size
>> and is awaiting moderation. Please advise.
>> 2016-05-13 20:13 GMT+02:00 Bob Berwick <berwick at csail.mit.edu>:
>>> would be useful for the list and the community. please do.
>>> On May 13, 2016, at 1:51 PM, Koos Wilt <kooswilt at gmail.com> wrote:
>>> I wrote an overview of the performance of parsers about 4 years ago.
>>> Would sending it somewhere (e.g. to Mr Brew) be helpful to anyone? It's on
>>> my other laptop so I have to dig for it.
>>> Bes
>> _______________________________________________
>> 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 -------------- A non-text attachment was scrubbed... Name: not available Type: text/html Size: 7617 bytes Desc: not available URL: <https://mailman.uib.no/public/corpora/attachments/20160514/701f421d/attachment.txt> -------------- next part -------------- A non-text attachment was scrubbed... Name: pdpng.png Type: image/png Size: 277186 bytes Desc: not available URL: <https://mailman.uib.no/public/corpora/attachments/20160514/701f421d/attachment-0001.png> -------------- next part -------------- A non-text attachment was scrubbed... Name: assert.tiff Type: image/tiff Size: 448390 bytes Desc: not available URL: <https://mailman.uib.no/public/corpora/attachments/20160514/701f421d/attachment-0001.tiff>

More information about the Corpora mailing list