Feature Engineering f or Chinese Semantic Role Labeling

Transcription

1 J OU RNAL OF CH IN ESE IN FORMA TION PROCESSIN G Vol. 21, No. 1 Jan., 2007 : (2007) (SWCL2006) (, ),,,,, (, ) :,, :,, Chinese Proposition Bank ( CPB),, F2Score % %,, : ; ; ; ; ; : TP391 : A Feature Engineering f or Chinese Semantic Role Labeling L IU Huai2jun, CHE Wan2xiang, L IU Ting ( Information Retrieval Laboratory, Haerbin Institute of Technology, Haerbin, Heilongjiang , China) Abstract : In the natural language processing field, researchers have experienced a growth of interest in semantic role labeling by applying statistical and machine2learning methods. Using rich features is the most important part of semantic parsing system. In this paper, some new effective features and combination features are proposed, such as next word of the constituent, predicate and phrase type combination, predicate class and path combination, and so on. And then we report the experiment s on the dataset from Chinese Proposition Bank ( CPB). After these new features used, the final system improves the F2Score f rom % to %. The result s show that the performance of the system has a statistically significant increase. Therefore it is very important to find better features for semantic role labeling. Key words : computer application ; Chinese information processing ; semantic parsing ; semantic role labeling ; feature engineering ; maximum entropy classifier 1,,,, :,, : : : ( , , ) : (1982 ),,,

2 ( Semantic Role Labeling, SRL ),,, [ Agent ] [ Tmp ] [ V ][ Passive ],,,, ( SVM ) [ 1 ], ( Maximum Entropy ) [2 ], SNo W ( Sparse Network of Winnows) [ 3 ],, 2 3, Chinese Proposition Bank ( CPB) [4 ] CPB Upenn Penn Chinese Treebank, Penn Chinese Treebank Penn Chinese Treebank Sinorama CPB 20, Arg025,Arg0, Arg1, ArgM, ( Secondary Tags), ArgM2LOC, ArgM2TMP [5 ] 1 CPB CPB Penn Chinese Treebank, Penn Chinese Treebank,, 760, , Chinese Proposition Bank (Constituent) ( Phrase) ( Word) (Dependency Relation),, 1,N P2SBJ,N P2TMP,V P, [6 ] :, ;, ;,,,,, 3. 2, (N E), Xue [7,8 ]

3 1 : 81,, :, Sun [ 9 ] Chinese) ( Head rules for : 2, V P VV2N P2OBJ 5. :, Xue [8 ] 6. : 2,N P2TMP N P2TMP V P V P VV 7. :, 2, N P2TMP N P2 TMP V P 2 8. :, 9. : 2 N P2OBJ np_np_v_n P2OBJ,,, CPB,,, 3. 3, 1. : CPB,, 2IO,2OBJ,2SBJ : Xue [ 10 ] Penn Chinese Treebank, : ( CP) ( IP) ( IP2Q) IP CP IP2Q : :, (V P), IP CP : CPB, Arg2 5 : Arg2,,Arg2, Arg2,,, 3. 4, CPB,,

4 CPB,,, Arg02PSR Arg02PSE,, Arg02PSR Arg02PSE,,,, ,, F2Score : F2Score = 2 3 Precision 3 Recall Precisio n + Recall ( Precision) ( Recall) : Precision = Recall =,, ;, 1 F2Score,, 2 ( 1), n = , 2 = 0. 10, Fb, Fn, 2 = ( nfb - nfn ) 2 nfn + ( nfb - nfn ) 2 n(1 - Fn) 2 (1) 2 = , Fn 1 Precision ( %) Recall ( %) F2Score ( %) ( ) + : : : : :

5 1 : 83 Precision ( %) Recall ( %) F2Score ( %) + : : : : : : , F2Score,,, ;,,,,,,, Arg2,, 4. 2, 2 2 Precision ( %) Recall ( %) F2Score ( %) :,,,, 1. 55, %, :,,,,,, ArgMs, Arg3, Arg4,,Penn Chinese Treebank,, (2NON E2) 3 : (2NON E2 3 21) (2NON E2 3 22), (NR ) 3 CPB, :, (Subject) Arg0, (Object) Arg1, : Arg1, Arg2, Arg2, Arg3, Arg4,, Arg2 : ,, 1 Arg2 ; 2 Arg2 ;

6 Arg2 5 ( Maximum Entropy), Chinese Proposition Bank (CPB),,,,, F2Score % % CPB,,,, : [1 ] S. Pradhan, K. Hacioglu, V. Krugler, et al. Support vector learning for semantic argument classification [J ]. Machine Learning Journal, 2005, vol. 60, no. 12 3, [2 ] N. Kwon, M. Fleischman, E. Hovy. Senseval auto2 matic labeling of semantic roles using Maximum Entro2 py models [ A ]. Senseval23 : Third International Work2 shop on the Evaluation of Systems for the Semantic A2 nalysis of Text [ C]. Barcelona, Spain : Association for Computational Linguistics, 2004, [3 ] P. Koomen, V. Punyakanok, D. Roth, et al. Gener2 alized Inference with Multiple Semantic Role Labeling Systems [ A ]. In : Proceedings of the Ninth Conference on Computational Natural Language Learning ( CoN2 LL22005) [ C]. Ann Arbor, Michigan : Association for Computational Linguistics, 2005, [4 ] N. Xue, M. Palmer. Annotating the Propositions in the Penn Chinese Treebank [ A ]. In : Proceedings of the Second SIGHAN Workshop on Chinese Language Processing [ C]. Sapporo, Japan : 2003, [5 ] M. Palmer, D. Gildea, P. Kingsbury. The Proposi2 tion Bank : An Annotated Corpus of Semantic Roles [J ]. Computational Linguistics, 2005, 31 ( 1 ), [6 ] V. Punyakanok, D. Roth, W. Yih. The Necessity of Syntactic Parsing for Semantic Role Labeling [ A ]. In : Proceedings of CoNLL204 [ C]. 2004, [7 ] N. Xue, M. Palmer. Calibrating features for semantic role labeling [ A ]. In : Proc. of the EMNL P22004 [ C]. Barcelona, Spain : [8 ] N. Xue, M. Palmer. Automatic semantic role labeling for Chinese verbs [ A ]. In : Proc. IJ CAI2005 [ C]. Ed2 inburgh, Scotland : [9 ] H. Sun and D. J urafsky. Shallow semantic parsing of Chinese [ A ]. In : Proceedings of NAACL 2004 [ C ]. Boston, USA : [10 ] N. Xue, F. Xia. The Bracketing Guidelines for the Penn Chinese Treebank [ D ], IRCS Report U2 niversity of Pennsylvania, Oct 2000.