Á¤º¸°úÇÐȸ ³í¹®Áö B : ¼ÒÇÁÆ®¿þ¾î ¹× ÀÀ¿ë
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
ÀÇÁ¸¹®¹ýÀ» ÈÄÇâ ¾ð¾î¸ðµ¨·Î »ç¿ëÇÏ´Â Çѱ¹¾î ¿¬¼ÓÀ½¼º ÀνĽýºÅÛ |
¿µ¹®Á¦¸ñ(English Title) |
A Korean Continuous Speech Recognition System using the Dependency Grammar as a Backward Language Model |
ÀúÀÚ(Author) |
ÀåµÎ¼º
±¸¸í¿Ï
Duseong Chang
Myoungwan Koo
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 24 NO. 04 PP. 0443 ~ 0449 (1997. 04) |
Çѱ۳»¿ë (Korean Abstract) |
º» ³í¹®¿¡¼´Â Çѱ¹¾î ¿¬¼ÓÀ½¼º ÀνĿ¡ Àû¿ë °¡´ÉÇÑ »õ·Î¿î ¾ð¾î¸ðµ¨ ¹æ¹ýÀ» Á¦½ÃÇÑ´Ù. ÀÌ ¾ð¾î¸ðµ¨Àº ±âÁ¸ÀÇ ¿¬¼ÓÀ½¼º ÀνĽýºÅÛ¿¡¼ »ç¿ëµÇ´ø ±¸¹®·ÐÀû ¾ð¾î¸ðµ¨µéÀÌ ´ëºÎºÐ ÁÖ¾îÁø ÀνĿµ»ó¿¡ ÇÑÁ¤µÇ¾î ±¸¼ºµÈ ±¸±¸Á¶¹®¹ýÀ» »ç¿ëÇÏ´ø °Í°ú ´Þ¸® ¹ÙÀ̱׷¥(bigram)À» ÀüÇâ ¾ð¾î¸ðµ¨·Î, ÀÇÁ¸¹®¹ýÀ» ÈÄÇâ ¾ð¾î¸ðµ¨·Î »ç¿ëÇÑ´Ù. ÀÌ ¾ð¾î¸ðµ¨Àº ´Ü¾î°£ÀÇ Åë°èÀûÀÎ ¿¬¾îÁ¤º¸¿Í ÀÇÁ¸°ü°è¸¦ °°ÀÌ °í·ÁÇϹǷΠ»ý·«°ú µµÄ¡°¡ ¹ø¹øÈ÷ ÀϾ´Â Çѱ¹¾î¿¡¼µµ ÀûÀº °è»ê½Ã°£À¸·Î ´ÙÀ½°ú °°Àº È¿À²¼ºÀ» °¡Áø´Ù. ÀÌ ¾ð¾î¸ðµ¨Àº ÀÎ½Ä µÈ ´Ü¾îÀÇ ¸®½ºÆ®¸¦ ¹®¹Ì¿¡¼ ¹®µÎÂÊÀ¸·Î ºÐ¼®ÇØ ³ª¾Æ°¨À¸·Î¼ ÇǼö½Ä¾î¸¦ ºñ·ÔÇÑ Á߽ɾ°¡ ¼ö½Ä¾îµéÀÇ µÚ¿¡ À§Ä¡ÇÏ´Â Çѱ¹¾î¿¡ ÀÖ¾î¼ À߸øµÈ ÀÎ½Ä °á°ú¸¦ ºÐ¼® µµÁß¿¡µµ Á¦°ÅÇÒ ¼ö ÀÖ´Ù. ¶ÇÇÑ ±âÁ¸ÀÇ ¾ð¾î¸ðµ¨°ú ´Þ¸® ÀνĿµ¿ª¿¡ ±¸¾Ö¹ÞÁö ¾Ê´Â ÀϹÝÀûÀÎ ¹®¹ýÀ» »ç¿ëÇÒ ¼ö ÀÖ´Ù. º» ³í¹®¿¡¼´Â Çѱ¹¾îÀÇ ºñ±³Àû ÀÚÀ¯½º·¯¿î ¹®¸Æ ±¸¼ºÀ» Ãæ½ÇÈ÷ ¹Ý¿µÇÑ ¾ð¾î¸ðµ¨À» »ç¿ëÇÑ Çѱ¹¾î ¿¬¼ÓÀ½¼º ÀνĽýºÅÛ¿¡ ´ëÇØ ¼³¸íÇÏ°í, ±âÁ¸ÀÇ ±¸±¸Á¶¹®¹ý°ú LR Æļ¸¦ »ç¿ëÇÑ ½Ã½ºÅÛ¿¡ ºñÇØ ÀûÀº ¾çÀÇ ¹®¹ý±ÔÄ¢À¸·Îµµ ºü¸¥ ½Ã°£¿¡ Àνİá°ú¸¦ ¼±ÅÃÇÒ ¼ö ÀÖÀ½À» º¸ÀδÙ. ¶ÇÇÑ, ¹ÙÀ̱׷¥¸¸À» »ç¿ëÇÑ ½Ã½ºÅÛ¿¡ ºñÇØ ´Ü¾î ¿ÀÀνķüÀº 10.59%, ¹®Àå ¿ÀÀνķüÀº 6.98% °¨¼Ò½ÃŲ ½ÇÇè °á°ú¸¦ º¸ÀδÙ. ¸¶Áö¸·À¸·Î Á¦¾ÈµÈ ¾ð¾î¸ðµ¨À» »ç¿ëÇÑ Çѱ¹¾î ¿¬¼ÓÀ½¼º ÀνĽýºÅÛÀÌ ±âÁ¸ÀÇ ÀνĽýºÅÛµé°ú´Â ´Þ¸® ±â°è¹ø¿ª ½Ã½ºÅÛ°ú ÀÇÁ¸¹®¹ýÀ» ÀÌ¿ëÇÑ ±¸¹®ºÐ¼®±â¸¦ °øÀ¯ÇÒ ¼ö ÀÖÀ¸¹Ç·Î, À½¼º ¹ø¿ª ½Ã½ºÅÛ(speech translation system)À» °í·ÁÇÒ ¶§ ¸Å¿ì È¿À²ÀûÀÎ Á¡À» º¸ÀδÙ.
|
¿µ¹®³»¿ë (English Abstract) |
In This paper, we propose a language modeling method which can improve recognition rate with a few additional computation. The proposed system uses the bigram as a forward language model and the dependency grammar as a backward language model. This system can exclude ungrammatical sentences earlier than the system which uses LR parser and Phrase Structure Grammar(PSG). The proposed method reduces word recognition error by 10.59% and sentence recognition error by 6.98%. This backward language model can be also used for the syntactic analysis of a machine translation system so that the speech translation system can be efficiently integrated.
|
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|