Orð og tunga - 01.06.2007, Page 33

Anna Björk Nikulásdóttir: Sjálfvirk greining merkingarvensla
23
fsletisk orðabók. 2002 (3. útgáfa). Ritstjóri: Mörður Ámason. Reykjavík: Edda.
Jón Hilmar Jónsson. 2005. Aðgangur og efnisskipan í íslensk-erlendum orðabókum
— vandi og valkostir. Orð og tunga 7: 21-40.
Kristín Bjamadóttir. 1998. Um skýringarorðaforðann. Orð og tunga 4: 33-43.
Mörður Ámason. 1998. Endurútgáfa „íslenskrar orðabókar". Stefna — staða — horf-
ur. Orð og tunga 4:1-8.
Storrer, Angelika. 2001. Digitale Wörterbucher als Hypertexte: Zur Nutzung des
Hypertextkonzepts in der Lexikographie. í: Ingrid Lemberg, Berhard Schröder og
Angelika Storrer (ritstj.). Chancen und Perspektiven computergestiltzter Lexikographie,
bls. 53-69. Tubingen: Max Niemeyer.
Svensén, Bo. 1993. Practical Lexicography. Principles and Methods of Dictionary-Making.
Oxford/New York: Oxford University Press.
Vefbækur Eddu: íslensk orðabók; Ensk orðabók: edda.is/vefbaekur. sótt: 31.08.2006
Þórdís Úlfarsdóttir. 2006. Málfræðileg mörkun orðasambanda. Orð og tunga 8: 117-
144.
Lykilorð
merkingarfræði orða, merkingarvensl, orðabókaskýringar, orðflokkamynstur
Keywords
lexical semantics, semantic relations, dictionary definitions, POS-pattems
Abstract
In the design of electronic dictionaries it is possible to organize the information due
to the meaning of the lexems. In this article a method for automatic extraction of se-
mantic relations from dictionary definitions is demonstrated. Definitions of all noun
lexems of a monolingual Icelandic dictionary, fslensk orðabók, were analyzed. First,
the definitions were tagged with Brants TnT-tagger which had been trained on an
Icelandic corpus. From the tagged data the POS-pattems of the defmitions were ex-
tracted and rules for extracting the semantic relations were developed. The mle al-
gorithm was implemented in Smalltalk, resulting in the tool MerkOr. The results of
the analyzis were promising. The test was made with a random set of lexemes, about
1,34% of the data. For each lexeme the result could be completely right, that is all
semantic relations from the reference data was found in the analyzis of MerkOr, or
it could be partly right, that is MerkOr did not find every relation compared to the
reference data, but did nevertheless not extract any wrong word or relation. The pre-
cision was 82,13% (completely right analyzed) up to 94,77% (completely right plus
partly right).