• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ³í¹®Áö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ³í¹®Áö > Çѱ¹Á¤º¸°úÇÐȸ ³í¹®Áö > Á¤º¸°úÇÐȸ ³í¹®Áö B : ¼ÒÇÁÆ®¿þ¾î ¹× ÀÀ¿ë

Á¤º¸°úÇÐȸ ³í¹®Áö B : ¼ÒÇÁÆ®¿þ¾î ¹× ÀÀ¿ë

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) CRFs¸¦ ÀÌ¿ëÇÑ ÀÇÁ¸±¸Á¶ ºÐ¼® ¹× ÀÇÁ¸ °ü°è¸í ºÎÂø
¿µ¹®Á¦¸ñ(English Title) Dependency Structure Analysis and Dependency Label Annotation Using CRFs
ÀúÀÚ(Author) ÃָͽĠ  Á¤¼®¿ø   ±èÇмö   Maengsik Choi   Seokwon Jeong   Harksoo Kim  
¿ø¹®¼ö·Ïó(Citation) VOL 41 NO. 04 PP. 0302 ~ 0308 (2014. 04)
Çѱ۳»¿ë
(Korean Abstract)
Çѱ¹¾î ¹®ÀåÀÇ ±¸Á¶¸¦ ºÐ¼®Çϱâ À§ÇØ ÀÇÁ¸±¸Á¶ ºÐ¼®À» ¸¹ÀÌ »ç¿ëÇÑ´Ù. ´ëºÎºÐÀÇ ÀÇÁ¸±¸Á¶ ºÐ¼® ¹æ¹ýÀº ¾îÀý »çÀÌÀÇ ÀÇÁ¸ °ü°è À¯¹«¸¸À» °á°ú·Î Á¦½ÃÇϸç ÁÖ¾î, ¸ñÀû¾î ±×¸®°í ¼ö½Ä¾î µîÀÇ Á¤º¸¸¦ Á¦°øÇÏÁö ¾Ê´Â´Ù. º» ³í¹®¿¡¼­´Â ÀÇÁ¸±¸Á¶ ºÐ¼®°ú ÀÇÁ¸ °ü°è¸í ºÎÂøÀ» µ¿½Ã¿¡ ¼öÇàÇÏ´Â ¸ðµ¨À» Á¦¾ÈÇÑ´Ù. Á¦¾È ¹æ¹ýÀº CRFs(Condition Random Fields)¸¦ ÀÌ¿ëÇÑ ´Ù´Ü°è ±¸ ´ÜÀ§È­(cascade chunking) ¹æ¹ýÀ» ÅëÇØ ÀÇÁ¸±¸Á¶¿Í ÀÇÁ¸ °ü°è¸íÀ» °áÇÕÇÑ Å±׸¦ ¹®Àå °¢°¢ÀÇ ¾îÀý¿¡ ºÎÂøÇÑ´Ù. ¼¼Á¾ ±¸¹® ºÐ¼® ¸»¹¶Ä¡¸¦ ÀÌ¿ëÇÏ¿© 10¹è ±³Â÷ °ËÁõ ½ÇÇèÀ» ÅëÇØ ÅëÇÕµÈ ¸ðµ¨ÀÇ ¼º´É(Á¤¹Ðµµ 81.11%)ÀÌ ÀÇÁ¸±¸Á¶ ºÐ¼®°ú ÀÇÁ¸ °ü°è¸í ºÎÂøÀÇ 2´Ü°è ¸ðµ¨º¸´Ù ³ôÀº ¼º´ÉÀ» º¸¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
In Korean, dependency parsing is frequently used to analyze syntactic structures of sentences. Most of the previous dependency parsing methods return only whether dependency relations between eojeols (spacing unit of Korean) exist or not. They do not return the names of dependency relations such as subject, object, modifier, and so on. In this paper, we propose an integrated dependency parsing model that finds dependency relations and annotates with dependency labels at the same time. The proposed model annotates each eojeol in a sentence with various tags, which combine dependency relations and dependency labels, by using a cascade chunking method based on conditional random fields (CRFs). In the 10-fold cross validation experiments with Sejong syntactic parsing corpus, the integrated model showed the better performance (the accuracy of 81.11%) than the previous two-step model that annotates with dependency labels after finding dependency relations.
Å°¿öµå(Keyword) ÀÇÁ¸±¸Á¶ ºÐ¼®   ÀÇÁ¸ °ü°è¸í ºÎÂø   ´Ù´Ü°è ±¸ ´ÜÀ§È­   CRFs   dependency parsing   dependency label   cascade chunking   CRFs  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå