tag:blogger.com,1999:blog-19803222.post5980174981959828134..comments2024-03-18T01:45:45.724-06:00Comments on natural language processing blog: NAACL 2012 Retrospectivehalhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-19803222.post-40508654283671373202012-06-25T22:24:52.919-06:002012-06-25T22:24:52.919-06:00The Wilensky paper seems to be OCRed at http://ijc...The Wilensky paper seems to be OCRed at http://ijcai.org/Past%20Proceedings/IJCAI-77-VOL1/PDF/003A.pdfJoel Nothmanhttps://www.blogger.com/profile/13572851908191383321noreply@blogger.comtag:blogger.com,1999:blog-19803222.post-27722912406864256942012-06-14T21:41:15.282-06:002012-06-14T21:41:15.282-06:00Glad you liked the talk, Hal! Sorry we didn't ...Glad you liked the talk, Hal! Sorry we didn't get to catch up in Montreal.Nitinnoreply@blogger.comtag:blogger.com,1999:blog-19803222.post-81376580038138808272012-06-14T19:32:05.299-06:002012-06-14T19:32:05.299-06:00Oh absolutely only (language-independent) dependen...Oh absolutely only (language-independent) dependency parsing. It would be interesting to see what happens with French past participles (and if we get more errors in those cases than others).Emily M. Benderhttps://www.blogger.com/profile/12220924508264195256noreply@blogger.comtag:blogger.com,1999:blog-19803222.post-66142029781380820892012-06-14T17:33:12.740-06:002012-06-14T17:33:12.740-06:00Hi,
@"No one has done this before"
i'...Hi,<br />@"No one has done this before"<br />i'm not sure I understand the context (is it restricted to modern dependency parsing?) because it has been done in almost all wide coveraged deep syntax grammars I'm aware of (Pargram's LFG grammar, FRMG (French Tag metagrammar), XTAG project, the one for Dutch and so on..) and in some extent, it also has been used in some "hybrid" frameworks (i'm thinking of the Tsarfaty's Relational realization model, cf. her spmrl 2010 paper). One could also argue that using morphological features (included in the POS labels for example) in a Collins model 2 based parsing model is a way to include some agreements between head and their dependands as well (see the work on spanish by (Cowan and Collins, 2005)) so I'm wondering if the Pos-Pos Label positive impact you've noted couldn't be compared to the one brought by the coarser level of the Collins model's back-off structure.<br /> <br />What I find really nice in your approach is that you're language agnostic but I wonder how this would work with a language with insane agreement rules such as French (eg. when the auxiliary is "avoir" (have) the past participle agrees with the object only if it occures before ), or on user generated content where agreement is apparently not even an option for most speakers.<br /><br />Nice work :)<br /><br />DjaméAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-19803222.post-70705562248730442572012-06-14T10:25:27.297-06:002012-06-14T10:25:27.297-06:00That was more or less our reaction, too ("no ...That was more or less our reaction, too ("no one's done this before? really??") and in fact part of my motivation for suggesting the tutorial. We did try it with auto-generated morphological information for Hebrew (see the paper for details) and it seems to be fairly robust.Emily M. Benderhttps://www.blogger.com/profile/12220924508264195256noreply@blogger.com