06.08.2015 Views

A Wordnet from the Ground Up

A Wordnet from the Ground Up - School of Information Technology ...

A Wordnet from the Ground Up - School of Information Technology ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

70 Chapter 3. Discovering Semantic Relatednessrecognise properly. We have, however, significantly limited <strong>the</strong> range of constructionswhere this constraint is met, so <strong>the</strong> achieved accuracy is relatively good – see Table 3.6.Moreover, in testing <strong>the</strong> presence of adjectival words between <strong>the</strong> target LU and <strong>the</strong>lexical element, we refer to agreement for more accurate recognition. Modificationby a nominal lemma in genitive clearly refers to <strong>the</strong> lexical meaning of <strong>the</strong> modifiednominal. NmgC had a positive influence on <strong>the</strong> accuracy in <strong>the</strong> experiments – see(Piasecki et al., 2007b, Piasecki and Radziszewski, 2009).Verbal LUs are described in plWordNet not in terms of subcategorisation frames,but by <strong>the</strong> semantic and lexical relations. So instead of recognising syntactic frames 14 ,we applied morphosyntactic constraints in a way similar to <strong>the</strong> description of nominalLUs. The description of occurrences of verbal LUs comprises four templates ofmorphosyntactic constraints (<strong>the</strong> lexical elements have been italicised):NSb – a particular noun as a potential subject of <strong>the</strong> given verb,NArg – a noun in a particular case as a potential verb argument,VPart – a present or past participle of <strong>the</strong> given verb as a modifier of some nominalLU 15 ,VAdv – an adverb in close proximity to <strong>the</strong> given verb.NSb is a symmetrical to <strong>the</strong> VsbC constraint applied to nominal LUs. Now nominalLUs are <strong>the</strong> lexical elements searched for. The NArg template is parametrised bytwo values: a case value (<strong>the</strong> nominative value is excluded as covered by NSb) anda nominal lexical element. Because <strong>the</strong>re is no agreement between a verb and itsargument and we had no description of verb subcategorisation frames for Polish, <strong>the</strong>NArg implementation is very straightforward. Having <strong>the</strong> verb in <strong>the</strong> centre (position0) we are looking for <strong>the</strong> first occurrence of <strong>the</strong> given lexical element in <strong>the</strong> given caseunless it is separated by an occurrence of ano<strong>the</strong>r verb (when we cannot disambiguate<strong>the</strong> attachment). VPart explores <strong>the</strong> common use of present and past participles asadjectival modifiers of nominal LUs. Verbs are described via <strong>the</strong>ir occurrences asparticiples and lexical elements are <strong>the</strong> modified nominal LUs. The constraint is verysimilar to <strong>the</strong> AdjC constraint for nominal LUs. For <strong>the</strong> VAdv constraint we test <strong>the</strong>presence of an lexical elements – an adverb – at <strong>the</strong> two closest positions to <strong>the</strong> leftor right. Adverbs have no grammatical categories except degree, so only distance canbe considered.MSRs for adjectives were constructed as a by-product of larger projects in (Hatzivassiloglouand McKeown, 1993, Freitag et al., 2005). Extraction of distributional14 This might be very difficult due to <strong>the</strong> lack of a shallow parser.15 A subtle agreement test and additional structural conditions distinguish such pairs <strong>from</strong> verbcomplementpairs.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!