*MWELex – MWE Lexica of Croatian, Slovene and Serbian Extracted from Parsed Corpora
Abstract
The paper presents *MWELex, a multilingual lexical repository of Croatian, Slovene and Serbian multiword
expressions that were extracted from parsed corpora. The lexica were built with the custom-built
DepMWEx tool which uses dependency syntactic patterns to identify MWE candidates in parse trees. The
extracted MWE candidates are subsequently scored by co-occurrence and organized by headwords producing
a resource of 23 to 48 thousand headwords and 3.2 to 12 million MWE candidates per language. The
evaluation of the lexicon, performed on Croatian and Slovene, shows an overall precision of just over 50%
for Croatian but as high as 85% for Slovene. Similarly, precision over specific syntactic patterns varies
greatly, 0.167-0.859 for Croatian, 0.158-1.00 for Slovene. The possible extension of the tool is demonstrated
on a simplistic distributional-based extraction of non-transparent MWEs and cross-lingual linking
of the extracted lexicons.
expressions that were extracted from parsed corpora. The lexica were built with the custom-built
DepMWEx tool which uses dependency syntactic patterns to identify MWE candidates in parse trees. The
extracted MWE candidates are subsequently scored by co-occurrence and organized by headwords producing
a resource of 23 to 48 thousand headwords and 3.2 to 12 million MWE candidates per language. The
evaluation of the lexicon, performed on Croatian and Slovene, shows an overall precision of just over 50%
for Croatian but as high as 85% for Slovene. Similarly, precision over specific syntactic patterns varies
greatly, 0.167-0.859 for Croatian, 0.158-1.00 for Slovene. The possible extension of the tool is demonstrated
on a simplistic distributional-based extraction of non-transparent MWEs and cross-lingual linking
of the extracted lexicons.
Full Text:
PDF
This work is licensed under a Creative Commons Attribution 3.0 License.