[Corpora-List] ProbBank purpose role: PNC or PRP?

Amir Zeldes Amir.Zeldes at georgetown.edu
Wed Mar 2 15:40:16 CET 2022


Hi Andy,

My own impression is that there is no reliable difference between PNC and PRP (at least not in OntoNotes, which is the corpus I know best). The emphasis of “purpose not cause” (PNC) seems to come from disagreements on how to annotate modifiers like “why”, which are often ambiguous between cause and purpose, so the label PNC says “not just purpose but also definitely not cause”. In OntoNotes 3 “PRP” was introduced next to PNC, but looking at examples, I see no real basis for the difference, except maybe as an ambiguity class (if you read “PRP” to mean ‘purpose-maybe-also-cause’). You can also notice that PNC is defined here as follows:

https://catalog.ldc.upenn.edu/docs/LDC2007T21/propbank/english-propbank.pdf

3.8 PNC: “Purpose clauses are used to show the motivation for some action. Clauses beginning with "in order to" are canonical purpose clauses”

But here:

http://clear.colorado.edu/compsem/documents/propbank_guidelines.pdf

1.4.10: PRP “Purpose clauses are used to show the motivation for some action. Clauses beginning with ‘in order to’ and ‘so that’ are canonical purpose clauses.”

So some versions of the guidelines clearly used the exact same definition for the different labels. This has been noticed before, see this post (search for PNC, it’s a bit long):

http://markneumann.xyz/blog/syntax/

The author of this post suggests collapsing PRP an PNC (and also CAU, but that part I think is wrong).

Hope this is helpful!

Amir

------------

Dr. Amir Zeldes

Assoc. Prof. of Computational Linguistics

Department of Linguistics

Georgetown University

1437 37th St. NW

Washington, DC 20057

https://corpling.uis.georgetown.edu/amir

From: corpora-bounces at uib.no <corpora-bounces at uib.no> On Behalf Of Christian Chiarcos Sent: Wednesday, March 2, 2022 5:07 AM To: alueckin <Luecking at em.uni-frankfurt.de> Cc: corpora at uib.no Subject: Re: [Corpora-List] ProbBank purpose role: PNC or PRP?

Am Mi., 2. März 2022 um 11:01 �Uhr schrieb Christian Chiarcos <christian.chiarcos at gmail.com <mailto:christian.chiarcos at gmail.com> >:

Hi Andy,

according to the 2010 guidelines (http://clear.colorado.edu/compsem/documents/propbank_guidelines.pdf), both are two different tags, see Tab. 5.2. And in the current frameset, both continue to be used, e.g., in http://verbs.colorado.edu/propbank/framesets-english-aliases/come.html (PRP) and http://verbs.colorado.edu/propbank/framesets-english-aliases/open.html (PNC).

I guess that should have been https://github.com/propbank/propbank-frames/blob/master/frames/come.xml and https://github.com/propbank/propbank-frames/blob/master/frames/open.xml, but these role identifier haven't changed.

Best,

Christian

In the reality of annotation, the difference can be hard to make at times, so that there will always be an inclination to generalize the one or the other.

Best,

Christian

Am Mi., 2. März 2022 um 10:21 �Uhr schrieb alueckin <luecking at em.uni-frankfurt.de <mailto:luecking at em.uni-frankfurt.de> >:

Dear list,

a clarification question regarding semantic role labeling according to PropBank as used, e.g., in CONLL 2012: there are two labels used for the purpose role, ARGM-PNC and ARGM-PRP. Are they used interchangeably?

Some philology:

The guidelines of CONLL 2014/15 mention only ARGM-PNC. The current PropBank guidelines (http://propbank.github.io) mention only ARGM-PRP. A break occurred in the 2005 CL paper (https://aclanthology.org/J05-1004/) where PNC is used in tab.1, p.77 but PRP is used in tab.13, p.99.

Thanks for clarifying clues!

Andy

_______________________________________________ UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora Corpora mailing list Corpora at uib.no <mailto:Corpora at uib.no> https://mailman.uib.no/listinfo/corpora

-------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: text/html Size: 10468 bytes Desc: not available URL: <https://mailman.uib.no/public/corpora/attachments/20220302/706edc81/attachment.txt>



More information about the Corpora mailing list