• 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 ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)

Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) º¹»ç ¸ÅÄ¿´ÏÁòÀ» ÀÌ¿ëÇÑ Çѱ¹¾î-Á¦ÁÖ¾î ±â°è¹ø¿ª
¿µ¹®Á¦¸ñ(English Title) Machine Translation of the Korean-Jeju Language using the Copy Mechanism
ÀúÀÚ(Author) ¹Ú´Ù¼Ö   Â÷Á¤¿ø   Da-Sol Park   Jeong-Won Cha  
¿ø¹®¼ö·Ïó(Citation) VOL 28 NO. 03 PP. 0190 ~ 0195 (2022. 03)
Çѱ۳»¿ë
(Korean Abstract)
º» ³í¹®¿¡¼­ Çѱ۷Πǥ±âÇϴ Ư¼º»ó Á¦ÁÖ¾î¿Í Çѱ¹¾î¿¡´Â °ãÄ¡´Â ¾îÈÖ°¡ »ó´çÈ÷ ¸¹´Ù´Â Á¡¿¡ ÁýÁßÇÏ¿´°í, Çѱ¹¾î-Á¦ÁÖ¾îÀÇ ´Ü¾î ġȯÀ̶ó´Â µ¥ÀÌÅÍ Æ¯¼ºÀ» °í·ÁÇÏ¿© º¹»ç ¸ÅÄ¿´ÏÁòÀ» ÀÌ¿ëÇÑ´Ù. º¹»ç ¸ÅÄ¿´ÏÁòÀº Æ÷ÀÎÅÍ »ý¼º ³×Æ®¿öÅ©¸¦ »ç¿ëÇßÀ¸¸ç, ÀÌ´Â ÀÔ·Â ¹®Àå ¶Ç´Â »çÀü ³» ´Ü¾î¸¦ ¼±ÅÃÇÏ¿© »ý¼ºÇÒÁö¸¦ °áÁ¤ÇÑ´Ù. ¶Ç Ÿ°Ù ¾ð¾îÀÇ ÅäÅ«À» ¿ì¼±¼øÀ§·Î ¼±ÅÃÇÏ°Ô °áÁ¤Çϸç À̸¦ ÀÌ¿ëÇÑ º¸»ó(reward)À» Àû¿ëÇÏ¿´´Ù. µ¥ÀÌÅÍ´Â Ä«Ä«¿Àºê·¹Àο¡¼­ ±¸ÃàÇÑ JIT(Jejueo Interview Transcripts)¸¦ ÀÌ¿ëÇÏ¿´°í, BLEU¸¦ ÅëÇØ ¾îÀý ´ÜÀ§ ¼º´ÉÀ» ÃøÁ¤ÇÏ¿´´Ù. Çѱ¹¾î¡æÁ¦ÁÖ¾î ½ÇÇèÀº °ËÁõ µ¥ÀÌÅÍ´Â 47.85, Æò°¡ µ¥ÀÌÅÍ´Â 47.12¸¦ º¸¿´°í, Á¦ÁÖ¾î¡æÇѱ¹¾î ½ÇÇèÀº °ËÁõ µ¥ÀÌÅÍ´Â 69.01, Æò°¡ µ¥ÀÌÅÍ´Â 68.54¸¦ º¸¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
In this paper, we focused on the fact that there isare a lot of overlapping vocabulary in the Jeju language and Korean language due to the characteristics of writing in the Korean language, and we used a copying mechanism in consideration of the data characteristic of word substitution in the Korean-Jeju language. The copying mechanism used a pointer generation network, which determines whether to generate by selecting an input sentence or word in the dictionary. In addition, it was decided to select the token of the target language as a priority and applied a reward was applied using this it [¿ÀÀü1] . Fordata, JIT (Jejueo Interview Transcripts) built by Kakao Brain was used, andperformance per eojeol was measured through BLEU. In the Korean-Jeju language experiment, the validation data was 47.85, ¡æ and the test data was 47.12, and in the Jeju language Korean experiment, the validation data was ¡æ 69.01, and the test data was 68.54.
Å°¿öµå(Keyword) ±â°è¹ø¿ª   ±â°èÇнÀ   ÀÚ¿¬¾î 󸮠  º¹»ç ¸ÅÄ¿´ÏÁò   Çѱ¹¾î-Á¦Á־ machine translation   machine learning   natural language processing   copy mechanism   korean-jeju language  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå