Orð og tunga - 01.06.2012, Blaðsíða 21

Matilieiv Whelpton: From human-oriented dictionaries
11
them but a computer must be provided with the information explic-
itly and systematically.
These changes to the original WordNet relation set are essential-
ly extensions and elaborations. However, DanNet also addresses a
fundamental problem with the original relation which provides the
backbone to Princeton WordNet: hyponymy. To be useful for infer-
encing, the hyponymy relation should: (i) hold of subkinds (a dog is
a kind of animal; a cat is a kind of animal), where (ii) co-hyponyms
are mutually exclusive (if something is a dog then it is not a cat and
vice versa). Pedersen et al. (2009: 277) point out that the traditional
use of hyponymy covers a broader range of relations than this, in a
way which is problematic for nlp applications: so-called, hyponymy
overload. Consider the following examples:
(14) oliemaleri 'oil painting', blomstermaleri 'flower painting',
fidusmaleri 'pseudo-art', akvarel 'water colour', marine-
billede 'seascape', klatmaleri 'daub'
Each of these terms is a hyponym of maleri 'painting'. However, if one
thinks in terms of mutually-exclusive subkinds, two candidate pairs
emergé:
(15) oliemalcri 'oil painting' vs akvarel 'water colour'
• subkinds of painting distinguished by the paint used
(16) blomstermaleri 'flower painting' vs marinebillede 'sea-
scape'
• subkinds of painting distinguished by the subject depicted
It is perfectly possible to have an oil painting which is also a flower
painting - and it is perfectly possible that that item is also a "daub".
Notice that the two mutually exclusive pairs are exclusive along a
particular dimension: paintings classified by the kind of paint used;
paintings classified by the subject depicted. Pedersen et al. therefore
adopt a proposal by Huang et al. (2008) which allows hyponyms de-
fined by a particular dimension of description to be grouped together:
oliemaleri 'oil painting' and akvarcl 'water colour' are said to be para-
nyms, terms associated with the same dimension of description. The
paranym relation allows co-hyponyms to be clustered into mutually-
exclusive subsets.
The terms fidusmaleri 'pseudo-art' and klatmaleri 'daub' remain
problems, however, because any painting can in fact be a daub: this