︹ ᠨᠠᠰᠤᠨ᠋ᠤ᠋ᠷᠲᠤ᠂ ᠴᠠᠩ ᠴᠢᠩ᠂ ᠵᠢᠶᠠᠩ ᠸᠧᠨ ᠪᠢᠨ᠂ ᠦ᠋ ᠶᠢᠨ ᠰᠢᠩ᠂ ᠯᠢᠦ ᠴᠶᠦ᠋ᠨ᠂ ᠵᠣᠣ ᠯᠢ ᠯᠢ ︺ - ᠦᠭᠦᠯᠡᠯ - （ᠬᠢᠲᠠᠳ） 蒙古语词法分析的有向图模型

»25 null»5 ù 2011 M9 ÏӐРJOURNAL OF CHINESE INFORMATION PROCESSING Vol. 25, No. 5 Sep., 2011 ÓcI|: 1003-0077 ︵2011︶ 05-0094-07 ÎÔMEs¥µ_m ˜ {Ó\1,ñÁ’1, 2,É ]1,2, * ¨é °m2, •1,u Ø Ý1, 3 ︵1.ÏS SÐý9 Ø/ Œùî î,Ø100190︔ 2. = ÎvÐ ÎÐÐý, = Ϋ„}+010021︔ 3.Š 2 =SvÐ9 ØА/ ŒÐý,Š 2T453007︶ Knull1: 我们为蒙古语词法分析建立了一种生成式的概率统计模型b该模型将蒙古语语句的词法分析结果描述为 有向图结构,图中节点表示分析结果中的词干a词缀及其相应标注,而边则表示节点之间的转移或生成关系b特别 地,在本工作中我们刻画了词干到词干转移概率a词缀到词缀转移概率a词干到词缀生成概率a相应的标注之间的 三种转移或生成概率,以及词干或词缀到相应标注相互生成概率b以内蒙古大学开发的20万词规模的三级标注 人工语料库为训练数据,该模型取得了词级切分正确率95.1%,词级联合切分与标注正确率93%的成绩b 1oM: 蒙古语︔ 词法分析︔ 词语切分︔ 词性标注︔ 词干提取︔ 有向图 Ïms Ë|: TP391null null null nullÓDS M ’: A Directed Graph Model for Mongolian Lexical Analysis JIANG Wenbin1,WU Jinxing1, 2, CHANG Qing1, 2,Nasanurtu2,LIU Qun1,ZHAO Lili1, 3 ︵1. Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China︔ 2. Inner Mongolian University, Huhhot, Inner Mongolia 010021, China︔ 3. Henan Normal University, Xinxiang, Henan 453007, China︶ Abstract: We proposea generative statistical model for Mongolian lexical analysis. This model describes the lexical analysis result as a directed graph, wherethe nodes represent thestems, affixes and their tags, while theedges re- present thetransition or generation relationships between nodes. Especially in this work, we adopt three kinds of transition or generation probabilities: a︶ probabilities of stem-stem transition, affix-affix transition and stem-affix generation︔ b︶ the transition or generation probabilities between the corresponding tags︔ and c︶ thegeneration prob- abilities between stems or affixes and their tags. Using the 3rd-level annotated corpus with about 200 000 words as the training data, this model achieves a word-level segmentation accuracy of 95.1%, and a word-level joint segmen- tation and tagging accuracy of 93%. Key words: Mongolian︔ lexical analysis︔ segmentation︔ POS tagging︔ stemming︔ directed graph là ° ù: 2010-08-29 nullçà ° ù: 2011-02-17 Á[ “:SE1 – SÐÁ'ù[ “ ︵Contract60736014︶︔ 863 × Ä[ “ ︵2006AA010108︶︔ â†aSEÔ¾ öBÔý Ó3︖ SSy︕ #Ä'ù[ “ ︵MZ115-038︶ T€eº:{Ó\ ︵1984null︶ , 3,p V 3,ö1ùîZ_¹MEsa Es„ Jr︔ ñÁ’ ︵1987null︶ , o, « V 3, ö1ùîZ_¹ ÎԐ︶ Ø︔ É ] ︵1985null︶ , o, « V 3,ö1ùîZ_¹ ÎԐ︶ Øb 1 null„ý qÔ„¸ öBÔý Ÿ ª,MEs ^v ”1 –Ôý︶ Ø ©¥$bqÔ¥M™ ¹e †,’ -¥MEsXÜSž L= V¨¥ £ Ü[1-4] , 7¿™ ÿ¯¥ öBÔý  ÎÔ„»ð:Ô, MEs¥ ’ q ¯µ v¥4 6 bW[ 5-11]b ö BŒ @Ð µ†³ p °m ç M¥C¨ , Jr/ Œ¥ × 1T¨︖ jAb öBÔýMEsT¹ Jr¥A︕ -4,³1¤žùî€÷¥1ÿb ÐqÔ¥3˜ ¨Q Ó¤¥MZ TM1, Î 5 ù{Ó\©: ÎÔMEs¥µ_m ˜ Ô„»ð:Ô©™ ÿ~¶¥ÔýM︖ p÷F¯ bƒ ËÔý¥MÔYÈ®MÄ„ ÏÄ © eT¨ ¥MFî Ž²,MEs¥ ©ü ^³ MÔ¥MÄ„Mî,i OSçz ñ Ì¥ ËYS ÿbƒ“B Ÿ,qÔ rT z¥½ Sÿ  ˜[12-14]ƒ ÚM¤‚ þ a¨,7ùª°¤¹ ¨ƒtCî¥LŸ½  ˜,] H| ©Kç¹Y MsSÿ[7-10] ,ƒ P¤“d¥ Ø ‚N´„ L¨Ÿ vu| hb 6BZ ë,.d¥¿︖ 5¥MEs  T³1 Ê¥ÔýÐ ¦Œ,ªª{nv ¥Ú ï Ø kqyª, ’ q„ × çŸi‚Ð ¦ibyN,á ̵A1/÷¹ ’¥d9 ˜,Ð V︖  ’¹ í ™ ÿ~¶Ôý¥M︖ p,V7 y ÎqyÚŸ︖ ¥MEs“db á ̹ ÎÔMEsy ë BÕ 3î T¥À qd9 ˜b¾ ˜| ÎÔÔ ¥MEs²T í ¹µ_m²,mÏ«ÄV Us²TÏ¥M ÄaM# M‹Sÿ,7H5V U«Ä­W¥M  3î1“, ñ Ì Y MÄaM# M‹Sÿ õ ¤îM¥︖ pb 3î TÀ qd9 ˜¹ƒtM 3î1“®[† a¥À q™ T,MEs¥Vñü ^Ùs  îµÀ qðKv¥µ_mb'ýTÏ á Ì Y MÄžMÄMÀ qaMžMM À qaMÄžM 3îÀ qaM‹¥Sÿ­W¥ ØÕ M 3îÀ q[#MÄMžM‹SÿMº 3 îÀ qbƒtM 3îÀ q[v » –9¥Z TVÞ Ô Ïd9¤žbk¿'ýT¥im¿ d9y ,¹ Ï¥ ÄñM »  V︖ ¥MÔ² ©Ê H,á Ìi Àµ æ¨ ¦ýSÿMEsÔ  o ­“¥ © Ôý'÷,9 Àµ︕ 9 Ê¥ME„Ô E© MÉ› · ,7 ^G V ¦ýÔ  oÏ | ¥MÄV„MV,YV¾B Æ ì k  îµ V︖ ¥ MZ Tb á Ì = ÎvÐ 7︖ ¥20£M︖ ¥ Ø︶ Sÿ ¦ýÔ  o ︵= ÎvÐ ˜àÔ ︶ É› L bá Ì Ûsé5%„5%¥ 0sYT¹ 7 ︖ “„© k“, :Å¥90%¥ 0 ††T¹Þ “b© k“ ,¾ ˜ |¤ M︶ Msž ’ q 95.1%,M︶ ó† MsÐSÿž ’ q93%¥zî b 6“,œñ“d¥Þ Vñº³1+ E ð' V ›î,³ ’VñPC 9 Vrž+äM Ä ð¥ Îb7 O,®¿“d+¬ Àµ¹ù © ÔýЩ M,á ÌMº³“ ¥¿îü V[‹¨ž  ð™ ÿ~¶¥Ôý b [/¥c«Ï,á Ì n5º  ÎÔMEs ¥ ©çl, –ª í á Ì¥ 3î TÀ qd9  ˜,Z U¾“d L²TiÉ›M‹¥s ª ü ª,á ÌÐ - ¦ýTÉ›1,Kª ^9²„Z¬b 2 null ÎÔMEs ]  ð™ ÿ~¶¥Ôý Ë », ÎÔ¥M®M Ä„ V︖ ¥MFîb‚]¥ ^, ÎÔMÄÐM ¥F†³1›V+µ¥ÿ : a︶ Mĺ︖ µBñ Oº︖ CK - ë︔ b︶ sMº︖ õ õM­ª︔ c︶ ] ËMÏ‚]M¶[+ç¥ ¨½Cb ÿ a︶ ︖ çBñ ÎÔMº︖ µBñl[Ï ,ƒBÄлð:Ô‚]︔ 7ÿ b︶ „c︶ ︖ ç  ‚]M¥+çC ¨½,ƒBÄÐÔ7Ô»‚ M]b [ = ÎvÐ ˜àÔ Ï¥ ÎÔMHUUR- NILDU/ HU-DU¹ è, +çÔâ/¥BÕM Es²T¹: HUURNI/Ve2+ LDU/Fe3+ HU/ Ft12- DU/Fc21 Ï, null + null|„null - null|sYV Uª 덓¥ ^ õª„sªbóçBñ ÎÔM,á Ì V[¹ùMÄV„MV,[¾B » ¥Z TÜ V︖ ¥ME²  Ÿb¾Ô  oÏ,sª „B†s õªð SMÏXÜ$S M Ÿb î µ¥sªûbM¥Á†, O[null - null|sï, è ÂMÁ¥null - DUnull︔ B†s õMÊ¿sM­ -, O[null /null|Ð - ë†ssï 7 Ÿbƒ†s õM  - ë¥null /null| ^ = ÎvÐ ˜àÔ „Sÿ VñÏ ¦ýFÉ €¥MÄÐM™¸Fîs­W¥s ï˜|,yNá Ì“d© k­ -| ÷“ { Æ ” Ï¥null /null|,[  E‡ LÌâ/¥ ÎÔÔ Q˜M Es“d¥‡ LŸ︖ b ¸ ÎÔMšµ‚¸BÕMEs©Ê² ,] HSMÄSÿ„MSÿ¥ H©,©Ê² ” M¤÷¹ ”vb ÂTô +ç¥ /ÓÌâ¹ ÎÔMÊ4ž ’¥MEs²,︔ ^ l †³ ¥Ù5,9 ^ ÎÔMEs¥ 4Ä îb¿Ô ýÐ︖ 5¥MEs“d︖ ¹ Äñ†ÿ¥MÚr ¹ » Ð V︖ Úe¥©Ês“†, ‚ üÉ¿ ô /ÓÌâ¹ ÄñMÊ4K ’¥©ÊbÔý ¥d9y  z V[Ð︖ 5ZE LCª ]º€bd 9ZE 4[¹ ÄñM ’çBñÚe¥†E©Ês 95 ÏӐÐ2011 M “,  üÉ¿Úr¹¹œñ 0Ê4K V︖ ¥œ8 s²Tb'ýT¥ × Ä' ÎÔMEs¥d 9y  b 3 nullµ_mÀ q ˜ ¿d9¥µSüy 98  Vs¹  Ë: 3 î Td9y „ ‘Y Td9y b Õ ˜8C  © –‚]¥y  ± ^b 3î ˜] H I° { Æ„s ²T,»sÁ 3À qKÚ¥ { ÆÐs²T ¥F†byN, ñµB†sÀ q© M¨¿ í  { Æ Ô ¥ 3î︖ pb ‘Y ˜5 ë@¿ { Æ I°s ²T,  “¥¿sX© { Æ¥ f ƒ/Kª¥s ²T©ÊbÐ 3î ˜¥ Ø QM1, ‘Y ˜÷ ˜† ¦ Ìs³ %Ù5¥Z Tb Y L9£ ü  ‘Y ˜¥ª ]b½ Sÿ¥ÜÅÙ5MŸSÿ , ‘ Y ˜qÔ„‡Ô û1 3î ˜µ üAÉ„b –7, ‘Y ˜¨¿ ÎÔMEsÎi‚ µ€³ %¥1oÙ5bBZ ë,Ð 3î ˜‚] ¥ ^, ‘Y ˜¥Þ 1︕ 9v ¥ ‘Y+˜,iY ȳ1 ¥É¥ }Ý}VñbЙ ÿe†¥qÔ„ ‡ÔM1, ÎÔM¯¥™ ÿ² P¤ ˜¥ Æ ì bW1v¥,V7³1v¥i%]¨„÷ò¥ Þ HW︔ 6BZ ë, ‘Y ˜YÈÇ a¨¿¹ Æ ì bW²%ç¥ ©y , ÂMŸSÿ„Gis, ñ ÌûµBñ%ç‚M¥MÔ½ b7¿™ ÿ~ ¶Ôý¥MEs Ÿ ª5‚ ︕ ’ç¥ Æ ì bW² ,y¹á ̳1 » òÕ V︖ ¥©ÊMÔ²,i Ê4KD©Ê²¥] H ’ç¾² =MÄ„M ¥Sÿb “ -,á Ì¿™ ÿ~¶ÔýMEs¥ ‘Y Ty 9X +*„ V›¥ZÃ,M1¥ùî Xܵ½¥É›b'ÓÏ,á ÌÇÿ¿  XÜ 3î TÀ qd9 ˜b 3. 1 null†B Ms¥ ˜² ]MÔ™ ÿe†¥qÔ€‡ÔM1,MÔ™ ÿ~¶¥ ÎÔ¥MEs÷^ ^Bñ Ž²É ›Ê4i ŽÏ«ÄÉ›Sÿ¥Vñ,7‚ ^Bñ e†¥LŸ½ SÿÙ5bƒ Ú,á Ìi‚ ” º ︖ ] HÉ› Ms„Sÿ¥KÔMEs ˜, 7 ^5V ¹e†¥ © ª ,'†B Ms¥ ˜ yb á ÌÜÔ ÏòM¥s²Tçl¹ ý² , Âm1 î U: m1 nullMÔ =†²V U null nullƒ Ú,S ︵Stem︶ V UMÄ, A ︵Adjoin︶ V U õ M,D ︵Disjoint︶ V UsMbá ̨´L õ¤ ¥ ñA ︵D︶ V U0ñ õM ︵sM ︶ bMÄžM­W[#MžªÃM­W, phV U 3î€M1“b¿œñÔ ,s ²T5 V í ¹ Ž², Âm2 î U: m2nullœ ¥MÔ =†²V U null nullІñM¥s²T²M1,œ s² Ï9F M #M¥MÄ­W¥ 3îM1“,V 7 îµMÄ„M­W™îBñ‰ êµ½¥ Ž² b ŽÏ«Ä'V UMÄ€M,7«Ä­W¥ H5V UMÄžMÄaMÄžM[#MžM ¥ 3îM1“b À7ý­,í ‚¿︖ 5 ˜Î ^d9 ˜,y ¥'Éû ^ YÂMÄaM# M‹Sÿ ︵ÂT ] H9SSÿ¥Å︶ ­W¥ 3îaM©ÿ ︖ pb ÂTá Ì︖ ¹ ŽÏ¥òÕ‚]¥H︕ 9M‹¥ ƒ  × ,ƒt ƒ × ¥ Q˜ «Ä­W 3îM︖ p¥ •.ÏÓMEsÐ Es µ †¦ {ùî[ J] .ÏӐÐ, 2008, 22 ︵2︶ : 10-17. [ 5] null * ¨é °m,Ò,=A ü.C} ÎÔÔ  oFý / Œ¥ÉZnullB} ÎÔMÔ1î MsÐSÿ“ d[C] //» E½ †S  ” öBÔýÓ3︶ ØÐ Œ ù︶ ö,2005. [ 6] null¥¡¼, •, * ¨é °m,©.¿d9Ôý ˜¥ ÎÓM Ms[ J].  T MYÐ ¦ýÆ︖ , 2009, 22: 108-112. [ 7] nullu¿,¥¡¼,V¿,©.¿Hq ÛÆ¥ ÎÔM Msùî[ J] .ÏӐÐ, 2010, 24 ︵5︶ : 31-35. [8] nullW¿.¿ªß ”: S ˜¥ ÎÔM Ms“d ¥ùî[ D] . = ÎvÐ « V8 ‚Ó, 2009. [9] null£,¦ ® ° Î.¿HMM¥ ÎÓ1îMŸS ÿùî[ J]. = Î =SvÐÐ ︵1 – SÐqÓñ︶ , 2010, 39 ︵2︶ , 206-209. [10] null á ˜null­êY X, Ü4null­ƒ ï Ü4.»ð:Ô M MsZE[ J].ÏӐÐ, 2004, 18 ︵6︶ : 61-65. [11] null Lawrence. R. Rabiner. A Tutorial on Hidden Mark- ov Models and Selected Applications in Speech Recog- nition[ C] // Proceedings of IEEE, 1989:257-286. [12] null John Lafferty, AndrewMcCallum, Fernando Pereira. Conditional random fields: Probabilistic models for segmenting and labeling sequence data[ C] // Proceed- ings of the 18th ICML, 2001:282-289. [13] null McCallum, A., Freitag, D., Pereira, F. Maximum entropy Markov models for information extraction and segmentation[ C] // Proc. ICML, 2000: 591-598. [14] null Stolcke, Andreas. Srilmnull an extensible language modeling toolkit[ C]/ /Proceedings of the Internation- al Conferenceon Spoken Language Processing, 2002: 311-318. ︵¤»93:︶ null null null semantic orientation of adjectives[ C]/ / Proceedings of the Eighth Conference on European Chapter of the As- sociation For Computational Linguistics. European Chapter Meeting of the ACL. Association for Compu- tational Linguistics, Morristown, NJ, 1997: 174-181. [ 6] null Yi,J. ,Nasukawa,T. , Bunescu, R. , Niblack, W. Sen- timent analyzer: Extracting sentiments about a given topic using natural language processing techniques [C] // TheThird IEEE International Conference on Da- ta Mining, November 2003,. IEEE Computer Society Press, Los Alamitos,2003:427-434. [ 7] null Bo Pang, Lillian Lee, Shivakumar Vaithyanathan, Thumbs up︖ Sentiment Classification using Machine Learning Techniques [ C]// Conference on Empirical Methods in Natural Language Processing. 2002:79-86. [ 8] null L. Mangasarian, D. R. Musicant. Lagrangian support vector machines[ J]. Journal of Machine Learning Re- search, 2001,1:161-177. [ 9] null · ¡, Ÿf,Ñw.¿Ôl س¥Ó' `_Ÿ MYÅ[ J] .ÏӐÐ, 2007,21 ︵1︶ : 96-100. [ 10] null Aizerman M, Braverman E, Rozonoer L. Theoretical foundations of the potential function method in pat- tern recognition learning[ J] . Automation and Remote Control, 1964,25:821-837. [ 11] null Mercer J. Functions of positive and negative typeand their connection with the theory of integral equations [ J] . Philosophical Transactions of the Royal Society of London,1909, A209:415-446. [12] null Chapelle O, Vapnik V N, Bacsquest O, et al. Choo- sing multiple parameters for support vector machine [ J] .Machine Learning. 2002, 46:131-159. [ 13] null Cucker F, SmoleS. On the mathematical foundations of learning [ J]. Bulletin of the American Mathemat-i cal Society, 2001:1-49. [14] null Smithsgf, Jordaanem. Improved SVM regression u- sing mixtures of kernels [ C] // Proceedings of the 2002 International Joint Conference on Neural Net- works. Washington, DC: IEEE, 2002, 3: 2785- 2790. [15] null Turney P, Littman M. . Measuring praise and crit-i cism: Inference of semantic orientation from associa- tion[ J]. ACM Transactions on Information Systems, 2003,21 ︵4︶ :315-346. 100