[developers] Ungrammatical Input and the FFTB

Dan Flickinger danf at stanford.edu
Thu Apr 9 21:56:00 CEST 2020


Hi Francis,

I might not quite follow you.  If a sentence doesn't get any parses, then there is nothing to do in treebanking, except move on to the next sentence, since the unparsed one will not offer you any discriminants to choose.  If it does get one or more parses, but is ungrammatical, I usually click "Reject".  But I am now starting in with parsing a set of sentences to train a better robust model for errorful student input, using the grammar with mal-rules, so here I don't reject all ungrammatical sentences, but only those where I still can't find an intended robust parse.

 Dan
________________________________
From: Francis Bond <bond at ieee.org>
Sent: Wednesday, April 8, 2020 9:27 PM
To: Dan Flickinger <danf at stanford.edu>; Woodley Packard <sweaglesw at sweaglesw.org>; developers at delph-in.net <developers at delph-in.net>
Subject: Ungrammatical Input and the FFTB

G'day,

if we are treebanking a profile with ungrammatical sentences (i-wf = 0), what is the best practice?   Currently you cannot annotate them at all.  I don't remember what we did in the fine system.  I feel it might be good to be automatically accept an utterance with i-wf=0 and no parse, and reject it if it has a parse, ....  But I am not really sure.

--
Francis Bond <http://www3.ntu.edu.sg/home/fcbond/>
Division of Linguistics and Multilingual Studies
Nanyang Technological University
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20200409/acf1b0df/attachment.html>


More information about the developers mailing list