| { | |
| "paper_id": "O06-1019", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T08:07:01.062920Z" | |
| }, | |
| "title": "", | |
| "authors": [], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "", | |
| "pdf_parse": { | |
| "paper_id": "O06-1019", | |
| "_pdf_hash": "", | |
| "abstract": [], | |
| "body_text": [ | |
| { | |
| "text": "\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u53ef\u7531 ) B A, \u03a0, M, N, K, , S ( \u7b49\u4e03\u500b\u5143\u7d20\u4f86\u8868\u793a\uff0c\u5e95\u4e0b\u91dd\u5c0d\u6a21 \u578b\u76f8\u95dc\u7b26\u865f\u8207\u53c3\u6578\u505a\u8aaa\u660e\uff1a S \u8868\u793a\u6240\u6709\u72c0\u614b\u7684\u96c6\u5408\uff0c } ,..., , { 2 1 N s s s S = \u3002 K \u8868\u793a\u6240\u6709\u89c0\u6e2c\u7b26\u865f\u7684\u96c6\u5408\uff0c } ,..., , { 2 1 M k k k K = \u3002 N \u8868\u793a\u6a21\u578b\u4e2d\u6240\u6709\u72c0\u614b\u7684\u500b\u6578\u3002 M \u8868\u793a\u6a21\u578b\u4e2d\u6240\u6709\u89c0\u6e2c\u7b26\u865f\u7684\u6578\u76ee\u3002 ) ( i \u03c0 = \u03a0 \u4ee3\u8868\u72c0\u614b\u521d\u59cb\u7684\u6a5f\u7387\u5411\u91cf\uff0c ) ( 1 i i s q P = = \u03c0 \uff0c N 1 \u2264 \u2264 i \uff0c\u8868\u793a\u5728 1 = t \u6642\uff0c\u72c0\u614b\u70ba i s \u7684\u6a5f\u7387\uff0c\u4e14\u9700\u6eff\u8db3 1 = \u2211 i \u03c0 \u7684\u689d\u4ef6\u3002 ] [ A ij a = \u4ee3\u8868\u72c0\u614b\u8f49\u79fb\u6a5f\u7387\u77e9\u9663\uff0c ) | ( 1 i t j t ij s q s q P a = = = + \uff0c N , 1 \u2264 \u2264 j i \uff0c\u8868 \u793a\u5f9e\u72c0\u614b i s \u5230\u72c0\u614b j s \u7684\u6a5f\u7387\uff0c\u4e14\u6eff\u8db3 0 \u2265 ij a \u548c \u2211 = = N 1 j 1 ij a ] ) ( [ B k b j = \u4ee3\u8868\u89c0\u6e2c\u7b26\u865f\u77e9\u9663\uff0c ) | ( ) ( j t k t j s q v o P k b = = = \uff0c N 1 \u2264 \u2264 j \u548c M 1 \u2264 \u2264 k \uff0c\u8868\u793a\u5728\u72c0\u614b\u70ba j s \u6642\uff0c\u89c0\u6e2c\u7b26\u865f\u70ba k v \u7684\u6a5f\u7387\uff0c\u4e14\u6eff\u8db3 \u2211 = = K k j k b 1 1 ) ( \u3002 \u7d66\u5b9a\u8f38\u5165\u4e4b\u89c0\u5bdf\u5e8f\u5217 n o o o O L 2 1 = ( t o \u8868\u793a\u5728\u6642\u9593 t \u6240\u5c0d\u61c9\u7684\u89c0\u6e2c\u7b26\u865f\uff0c\u4e14\u6eff\u8db3 \u039a \u2208 t o )\u3002\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u76ee\u7684\u5c31\u662f\u8981\u9078\u51fa\u4e00\u500b\u5c0d\u61c9\u65bc\u89c0\u6e2c\u5e8f\u5217\u4e4b\u6700\u4f73\u7684\u72c0 \u614b\u5e8f\u5217 n q q q Q L 2 1 = ( t q \u8868\u793a\u5728\u6642\u9593 t \u6240\u5c0d\u61c9\u7684\u72c0\u614b\uff0c\u4e14\u6eff\u8db3 S \u2208 t q )\uff0c\u4e5f\u5c31\u662f\u627e\u51fa ) | ( 1 1 n n O Q P \u70ba\u6700\u5927\u6a5f\u7387\u503c\u6642\u7684\u72c0\u614b\u5e8f\u5217\u3002 \u7531\u65bc\u5728\u99ac\u53ef\u592b\u57fa\u672c\u5047\u8a2d\u4e0b\uff0c\u7b2c 1 + t \u7684\u6642\u9593\u72c0\u614b\u53ea\u548c\u7b2c t \u7684\u6642\u9593\u72c0\u614b\u6709\u95dc\uff0c\u8207\u5176 \u4ed6\u4efb\u4f55\u4ee5\u524d\u7684\u6642\u9593\u72c0\u614b\u7121\u95dc\uff0c\u5373 } | { } ,..., , | { 1 2 1 1 t k t t k t q s q P q q q s q P = = = + + \uff0c\u4e14\u96a8\u6a5f \u904e\u7a0b\u4e2d\u7684\u6a5f\u7387\u8f49\u79fb\u4e0d\u96a8\u6642\u9593\u6539\u8b8a\uff0c\u56e0\u6b64 ) | ( 1 1 n n O Q P \u7684\u8a08\u7b97\u53ef\u7c21\u5316\u6210\uff1a \u220f \u220f \u220f = \u2212 = = \u2212 + = = n t t q n t q q q n t t t t t n n o B A q o P q q P O Q P t t t 1 1 1 , 1 1 1 1 ) ( ) | ( ) | ( ) , ( 1 1 \u03c0 \u800c\u53d6\u5f97\u6b64\u6700\u5927\u503c\u7684\u72c0\u614b\u5e8f\u5217 n Q 1 \uff0c\u5247\u662f\u4f7f\u7528\u7dad\u7279\u6bd4(Viterbi)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "i i i i i o q o q o f \u22c5 = , ) , ( i i q o , \u4ee3\u8868\u67d0\u500b\u89c0\u6e2c\u7b26\u865f\u4ee5\u53ca\u5176\u5c0d\u61c9\u7684\u72c0\u614b\uff0c\u65b0\u7684\u72c0\u614b\u7b26\u865f\u7d93\u904e\u7279\u88fd\u5316\u7684\u904e\u7a0b \u4e2d\uff0c\u7531\u539f\u4f86\u7684\u89c0\u6e2c\u7b26\u865f\u52a0\u4e0a\u539f\u4f86\u72c0\u614b\u4f86\u7522\u751f\uff0c\u6b64\u7279\u88fd\u5316\u7684\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u53c8\u7a31 \u70ba\u300c\u7279\u88fd\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u300d (Specialized HMM) \u3002\u800c\u5982\u679c\u4e0d\u5c07\u6240\u6709\u7684\u89c0\u6e2c\u7b26\u865f \u6240\u5c0d\u61c9\u7684\u72c0\u614b\u90fd\u505a\u7279\u88fd\u5316\uff0c\u800c\u662f\u53ea\u5728\u7279\u5b9a\u7684\u89c0\u6e2c\u7b26\u865f\u4e0b\uff0c\u624d\u505a\u7279\u88fd\u5316\u7684\u904e\u7a0b\u5247\u7a31 \u70ba\u300c\u8a5e\u5f59\u5f0f\u7684\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u300d (Lexicalized HMM) \uff0c\u6b64\u904e\u7a0b\u5c6c\u65bc\u7279\u88fd\u5316\u904e\u7a0b \u7684\u4e00\u7a2e\u7279\u4f8b\uff0c\u53c8\u88ab\u7a31\u70ba\u8a5e\u5f59\u5316(lexicalization) \uff0c\u6b64\u904e\u7a0b\u4ee5\u5e95\u4e0b\u5f0f\u5b50\u4f86\u8aaa\u660e\uff1a if , if . , ) , ( \uf8f4 \uf8f3 \uf8f4 \uf8f2 \uf8f1 \u2209 \u2208 = W o q o W o o q o q o f i i i i i i i i i \u5176\u4e2dW \u70ba\u7279\u88fd\u8a5e(specialized words) \uff0c\u53ea\u6709\u5c6c\u65bc\u7279\u88fd\u8a5e\u7684\u89c0\u6e2c\u7b26\u865f\u624d\u6703\u505a\u7279\u88fd\u5316", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u4e4b\u72c0\u614b\u300cB\u300d \u3002\u56e0\u6b64\u5728\u65b0\u7684\u8a13\u7df4\u8cc7\u6599\u4e2d\uff0c\u72c0\u614b\u7b26\u865f\u88ab\u5ef6\u4f38\u4e86\u3002 \u8868 4 \u7279\u88fd\u8a5e\u96c6\u5408\uff5b\u751f-E-B, \u8d77-B-B\uff5d\u505a\u8a5e\u5f59\u5316\u7522\u751f\u65b0\u7684\u72c0\u614b \u89c0\u6e2c\u7b26\u865f \u539f\u4f86\u7684\u72c0\u614b \u65b0\u7684\u72c0\u614b \u7814-B-B B B \u7a76-I-E E E \u751f-E-B B B-\u751f-E-B \u547d-S-E E E \u8d77-B-B B B-\u8d77-B-B \u6e90-E-E E", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Combination of Machine Learning Methods for Optimum Chinese Word Segmentation", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Asahara", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Fukuoka", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Azuma", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "L" | |
| ], | |
| "last": "Goh", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Watanabe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Matsumoto", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Tsuzuki", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "134--137", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Asahara, K. Fukuoka, A. Azuma, C. L. Goh, Y. Watanabe, Y. Matsumoto, T. Tsuzuki. \"Combination of Machine Learning Methods for Optimum Chinese Word Segmentation,\" In Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing, pp. 134-137, 2005", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Combining Segmenter and Chunker for Chinese Word Segmentation", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Asahara", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "L" | |
| ], | |
| "last": "Goh", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Matsumoto", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of Second SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "144--147", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Asahara, C. L. Goh, X. Wang and Y. Matsumoto. \"Combining Segmenter and Chunker for Chinese Word Segmentation,\" In Proceedings of Second SIGHAN Workshop on Chinese Language Processing, pp. 144-147, 2003", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Unknown Word Detection for Chinese By a Corpus-based Learning Method", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Bai", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Proceedings of ROCLING X", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "159--174", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "K. J. Chen and M. H. Bai. \"Unknown Word Detection for Chinese By a Corpus-based Learning Method,\" In Proceedings of ROCLING X, pp. 159-174, 1997", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Word Identification for Mandarin Chinese Sentences", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1992, | |
| "venue": "Proceedings COLING '92", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "101--105", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "K. J. Chen and S. H. Liu. \"Word Identification for Mandarin Chinese Sentences,\" Proceedings COLING '92, pp. 101-105, 1992", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Unknown Word Extraction for Chinese Documents", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [ | |
| "Y" | |
| ], | |
| "last": "Ma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of COLING 2002", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "169--175", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "K. J. Chen and W. Y. Ma. \"Unknown Word Extraction for Chinese Documents,\" In Proceedings of COLING 2002, pp. 169-175, 2002", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Chinese Word Segmentation by Classification of Characters", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "L" | |
| ], | |
| "last": "Goh", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Asahara", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Matsumoto", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "International Journal of Computational Linguistics and Chinese Language Processing", | |
| "volume": "10", | |
| "issue": "3", | |
| "pages": "381--396", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C. L. Goh, M. Asahara and Y. Matsumoto. \"Chinese Word Segmentation by Classification of Characters,\" International Journal of Computational Linguistics and Chinese Language Processing Vol. 10, No. 3, pp. 381-396, 2005", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "HMM Specialization with Selective Lexicalization", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Kim", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "Z" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Rim", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "Proceedings of the joinSIGDAT Conference on Empirical Methods in Natural Lan-guage Processing and Very Large Corpora(EMNLP-VLC-99)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "121--127", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. D. Kim, S. Z. Lee and H. C. Rim. \"HMM Specialization with Selective Lexicalization,\" In Proceedings of the joinSIGDAT Conference on Empirical Methods in Natural Lan-guage Processing and Very Large Corpora(EMNLP-VLC-99), pp. 121-127, 1999", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Lexicalized Hidden Markov Models for Part-of-Speech Tagging", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "Z" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "I" | |
| ], | |
| "last": "Tsujii", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Rim", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2000, | |
| "venue": "Proceedings of 18th International Conference on Computa-tional Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "481--787", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "S. Z. Lee, J. I. Tsujii and H. C. Rim. \"Lexicalized Hidden Markov Models for Part-of-Speech Tagging,\" In Proceedings of 18th International Conference on Computa-tional Linguistics, Saarbrucken, Germany, pp.481-787, 2000", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Unsupervised Training for Overlapping Ambiguity Resolution in Chinese Word Segmentation", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "F" | |
| ], | |
| "last": "Gao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "F" | |
| ], | |
| "last": "Li", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of Second SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--7", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Li, J. F. Gao, C. N. Huang and J. F. Li. \"Unsupervised Training for Overlapping Ambiguity Resolution in Chinese Word Segmentation,\" In Proceedings of Second SIGHAN Workshop on Chinese Language Processing, pp. 1-7, 2003", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Perceptron Learning for Chinese Word Segmentation", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "Y" | |
| ], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Miao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Bontcheva", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Cunningham", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "154--157", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Y. Y. Li, C. J. Miao, K. Bontcheva and H. Cunningham. \"Perceptron Learning for Chinese Word Segmentation,\" In Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing, pp. 154-157, 2005", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Towards a Hybrid Model for Chinese Word Segmentation", | |
| "authors": [ | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Lu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "189--192", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "X. Lu. \"Towards a Hybrid Model for Chinese Word Segmentation,\" In Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing, pp. 189-192, 2005", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Covering Ambiguity Resolution in Chinese Word Segmentation Based on Contextual Information", | |
| "authors": [ | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Luo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Sun", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Tsou", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of COLING 2002", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "598--604", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "X. Luo, M. Sun and B. K. Tsou. \"Covering Ambiguity Resolution in Chinese Word Segmentation Based on Contextual Information,\" In Proceedings of COLING 2002, pp. 598-604, 2002", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "A Bottom-up Merging Algorithm for Chinese Unknown Word Extraction", | |
| "authors": [ | |
| { | |
| "first": "W", | |
| "middle": [ | |
| "Y" | |
| ], | |
| "last": "Ma", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of Second SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "31--38", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "W. Y. Ma and K. J. Chen. \"A Bottom-up Merging Algorithm for Chinese Unknown Word Extraction,\" In Proceedings of Second SIGHAN Workshop on Chinese Language Processing, pp. 31-38, 2003", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Foundation of Statistical Natural Language Processing", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Manning", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Schutze", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "317--380", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C. D. Manning and H. Schutze. \"Foundation of Statistical Natural Language Processing,\" Chapter 9-10. pp. 317-380, 1999", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Shallow Parsing using Specialized HMMs", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Molina", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Pla", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Journal of Machine Learning Research", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "595--613", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A. Molina and F. Pla. \"Shallow Parsing using Specialized HMMs,\" Journal of Machine Learning Research 2, pp. 595-613, 2002", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "A Hidden Markov Model Approach to Word Sense Disambiguation", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Molina", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Pla", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "E", | |
| "middle": [], | |
| "last": "Segarra", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of the VIII Conferencia Iberoamericana de Inteligencia Artificial", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--9", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A. Molina, F. Pla and E. Segarra. \"A Hidden Markov Model Approach to Word Sense Disambiguation,\" In Proceedings of the VIII Conferencia Iberoamericana de Inteligencia Artificial, IBERAMIA 2002, pp. 1-9, 2002", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Improving Part-of-Speech Tagging using Lexicalized HMMs", | |
| "authors": [ | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Pla", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Molina", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Natural Language Engineering", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "167--189", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "F. Pla and A. Molina. \"Improving Part-of-Speech Tagging using Lexicalized HMMs,\" Natural Language Engineering, pp. 167-189, 2004", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Part-of-Speech Tagging with Lexicalized HMM", | |
| "authors": [ | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Pla", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Molina", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2001, | |
| "venue": "proceedings of International Conference on Recent Advances in Natural Language Processing(RANLP2001)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "F. Pla and A. Molina. \"Part-of-Speech Tagging with Lexicalized HMM,\" In proceedings of International Conference on Recent Advances in Natural Language Processing(RANLP2001), 2001", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition", | |
| "authors": [ | |
| { | |
| "first": "L", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Rabiner", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1989, | |
| "venue": "Proceedings of the IEEE", | |
| "volume": "77", | |
| "issue": "", | |
| "pages": "257--286", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "L. R. Rabiner. \"A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition,\" Proceedings of the IEEE, Vol.77, No.22, pp. 257-286, 1989", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "A Conditional Random Field Word Segmenter for Sighan Bakeoff", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Tseng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Chang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Andrew", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Jurafsky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Manning", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "168--171", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "H. H. Tseng, P. H. Chang, G. Andrew, D. Jurafsky, and C. Manning. \"A Conditional Random Field Word Segmenter for Sighan Bakeoff 2005,\" In Proceedings of Fourth SIGHAN Workshop on Chinese Language Processing, pp. 168-171, 2005", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "A General and Multi-lingual Phrase Chunking Model Based on Masking Method", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Wu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Chang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Lee", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Computer Science (LNCS): Computational Linguistics and Intelligent Text Processing", | |
| "volume": "3878", | |
| "issue": "", | |
| "pages": "144--155", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Y. C. Wu, C. H. Chang and Y. S. Lee, \"A General and Multi-lingual Phrase Chunking Model Based on Masking Method,\" Lecture Notes in Computer Science (LNCS): Computational Linguistics and Intelligent Text Processing, Vol. 3878, pp. 144-155, 2006", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Chinese Word Segmentation as Character Tagging", | |
| "authors": [ | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Xue", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "International Journal of Computational Linguistics and Chinese", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "29--48", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "N. Xue. \"Chinese Word Segmentation as Character Tagging,\" International Journal of Computational Linguistics and Chinese, pp. 29-48, 2003", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "Chinese Word Segmentation as LMR Tagging", | |
| "authors": [ | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Xue", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Shen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of Second SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "176--179", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "N. Xue and L. Shen. \"Chinese Word Segmentation as LMR Tagging,\" In Proceedings of Second SIGHAN Workshop on Chinese Language Processing, pp. 176-179, 2003", | |
| "links": null | |
| }, | |
| "BIBREF23": { | |
| "ref_id": "b23", | |
| "title": "Automatic Recognition of Chinese Unknown Words Based on Roles Tagging", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "P" | |
| ], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Q", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "X", | |
| "middle": [ | |
| "Q" | |
| ], | |
| "last": "Cheng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of First SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "71--77", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "H. P. Zhang, Q. Liu, H. Zhang and X. Q. Cheng. \"Automatic Recognition of Chinese Unknown Words Based on Roles Tagging,\" In Proceedings of First SIGHAN Workshop on Chinese Language Processing, pp. 71-77, 2002", | |
| "links": null | |
| }, | |
| "BIBREF24": { | |
| "ref_id": "b24", | |
| "title": "HHMM-based Chinese Lexical Analyzer ICTCLAS", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "P" | |
| ], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Yu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [ | |
| "Y" | |
| ], | |
| "last": "Xiong", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Q", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of Second SIGHAN Workshop on Chinese Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "187--187", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "H. P. Zhang, H. K. Yu, D. Y. Xiong and Q. Liu. \"HHMM-based Chinese Lexical Analyzer ICTCLAS,\" In Proceedings of Second SIGHAN Workshop on Chinese Language Processing, pp. 187-187, 2003", | |
| "links": null | |
| }, | |
| "BIBREF25": { | |
| "ref_id": "b25", | |
| "title": "Study on segmentation of ambiguous phrases with the combinatorial type", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Zheng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [ | |
| "F" | |
| ], | |
| "last": "Wu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "Collections of Papers on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. H. Zheng and F. F. Wu. \"Study on segmentation of ambiguous phrases with the combinatorial type,\" Collections of Papers on Computational Linguistics.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "num": null, | |
| "text": ")\u6539\u6210 M-HMM \u7684\u904e\u7a0b\uff0c\u4e3b\u8981\u662f\u5c07\u6b63\u5411\u9577\u8a5e\u512a \u5148(FMM)\u8207\u53cd\u5411\u9577\u8a5e\u512a\u5148(BMM)\u4e4b\u65b7\u8a5e\u7d50\u679c(\u5373\u6240\u5f97\u7684 BIES \u6a19\u7c64)\uff0c\u8207\u539f \u4f86\u7684\u300c\u5b57\u5143\u300d\u7d44\u6210\u7684\u65b0\u7684\u89c0\u6e2c\u7b26\u865f\uff0c\u5ef6\u4f38\u70ba\u300c\u5b57\u5143-FMM-BMM\u300d\u7b49\u4e09\u500b\u8cc7\u8a0a\u7d50 \u5408\u800c\u6210\u7684\u89c0\u6e2c\u5e8f\u5217\u3002\u8868 3 \u4e2d\u4ee5\u4e00\u500b\u4f8b\u5b50\u4f86\u91dd\u5c0d M-HMM \u8a13\u7df4\u4ee5\u53ca\u6e2c\u8a66\u904e\u7a0b\u505a\u500b \u8aaa\u660e\uff0c\u5728\u8a13\u7df4\u968e\u6bb5\u4e2d\uff0c\u539f\u59cb\u7684\u89c0\u6e2c\u7b26\u865f\u5e8f\u5217\u70ba\u300c\u7814\u3001\u7a76\u3001\u751f\u3001\u547d\u3001\u8d77\u3001\u6e90\u300d \uff0c\u52a0 \u5165\u4e86\u9577\u8a5e\u512a\u5148\u6cd5\u7684\u8cc7\u8a0a\u5f8c\uff0c\u65b0\u7684\u89c0\u6e2c\u7b26\u865f\u5e8f\u5217\u4fbf\u88ab\u8f49\u63db\u6210\u300c\u7814-B-B\u3001\u7a76-I-E\u3001\u751f -E-B\u3001\u547d-S-E\u3001\u8d77-B-B\u3001\u6e90-E-E\u300d \u3002\u9019\u4e9b\u4e2d\u6587\u5b57\u5143\u65c1\u7684 B\u3001I\u3001E\u3001S \u6a19\u7c64\u5373\u662f\u7531\u6b63 \u5411\u9577\u8a5e\u512a\u5148\u8207\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u6240\u6a19\u793a\u7684\uff0c\u56e0\u6b64\u65b0\u7684\u89c0\u6e2c\u7b26\u865f\u7a2e\u985e\u76f8\u7576\u65bc\u589e\u52a0\u4e86 \u6211\u5011\u4ee5\u4e00\u500b\u4f8b\u5b50\u4f86\u505a\u8aaa\u660e\uff0c\u5982\u8868 4\uff0c\u5047\u82e5\u89c0\u6e2c\u7b26\u865f\u300c\u751f-E-B\u300d \u3001 \u300c\u8d77-B-B\u300d\u662f \u5c6c\u65bc\u7279\u88fd\u8a5e\uff0c\u5247\u7d93\u904e\u8a5e\u5f59\u5316\u7684\u904e\u7a0b\u4e4b\u5f8c\uff0c\u89c0\u6e2c\u7b26\u865f \u300c\u751f-E-B\u300d \u4ee5\u53ca \u300c\u8d77-B-B\u300d \u6240\u5c0d\u61c9\u7684\u72c0\u614b\u5c31\u88ab\u8f49\u63db\u6210\u300cB-\u751f-E-B\u300d \u3001 \u300cB-\u8d77-B-B\u300d\u4e86\u3002\u5728\u89c0\u6e2c\u7b26\u865f \u300c\u751f-E-B\u300d \u4e2d\uff0c\u539f\u4f86\u7684\u72c0\u614b B \u4fbf\u88ab\u5206\u5272\u6210\u5169\u500b\u4e0d\u540c\u7684\u72c0\u614b\uff1a\u4e00\u500b\u662f\u7531\u89c0\u6e2c\u7b26\u865f\u300c\u751f-E-B\u300d \u6240\u5c6c\u7684\u72c0\u614b\u300cB-\u751f-E-B\u300d\u4ee5\u53ca\u5176\u4ed6\u672a\u5206\u5272\u7684\u89c0\u6e2c\u7b26\u865f(\u5982\u89c0\u6e2c\u7b26\u865f\u300c\u7814-B-B\u300d )", | |
| "type_str": "figure", | |
| "uris": null | |
| }, | |
| "TABREF0": { | |
| "content": "<table><tr><td>1. \u7dd2 \u7dd2 \u7dd2 \u7dd2\u8ad6 \u8ad6 \u8ad6 \u8ad6 \u53e6\u5916\uff0c \u300c\u672a\u77e5\u8a5e\u300d\u5247\u6307\u8fad\u5178\u4e2d\u672a\u6536\u9304\u7684\u8a5e\uff0c\u5305\u542b\u4e86\u4eba\u540d\u3001\u5730\u540d\u3001\u7d44\u7e54\u540d\u3001\u4eba 2. \u76f8\u95dc\u7814\u7a76 \u76f8\u95dc\u7814\u7a76 \u76f8\u95dc\u7814\u7a76 \u76f8\u95dc\u7814\u7a76 \u5176\u4ed6\u89e3\u6c7a\u672a\u77e5\u8a5e\u554f\u984c\u7684\u7814\u7a76\uff0c\u5982 Zhang \u7b49\u4eba [24] \u65bc 2002 \u5e74\u7684\u7814\u7a76\uff0c\u5247\u4f7f \u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u9700\u4ef0\u8cf4\u5176\u4ed6\u5916\u90e8\u8cc7\u6e90\u6216\u662f\u7d50\u5408\u5176\u4ed6\u7684\u5b78\u7fd2\u6f14\u7b97\u6cd5\uff0c\u624d\u53ef\u4ee5\u9054\u5230 \u76f4\u5230\u53e5\u5b50\u7684\u958b\u982d\u800c\u7d50\u675f\u3002 \u5728\u4e2d\u6587\u65b7\u8a5e\u7684\u554f\u984c\u4e0a\uff0c\u4e00\u65e6\u5c07\u6b32\u65b7\u8a5e\u5b57\u4e32\u4e2d\u7684\u6240\u6709\u5b57\u5143\u90fd\u5df2\u5206\u985e\u5b8c\u6210\uff0c\u5247\u4e5f\u8868\u793a</td></tr><tr><td>\u4e2d\u6587\u65b7\u8a5e\u5728\u4e2d\u6587\u7684\u81ea\u7136\u8a9e\u8a00\u8655\u7406\u4e0a\uff0c\u662f\u975e\u5e38\u91cd\u8981\u7684\u524d\u7f6e\u8655\u7406\u5de5\u4f5c\u3002\u8a31\u591a\u4e2d\u6587 \u540d\u5730\u540d\u7d44\u7e54\u540d\u4e4b\u7e2e\u5beb\u3001\u884d\u751f\u8a5e\u3001\u8907\u5408\u8a5e\u3001\u6578\u5b57\u578b\u614b\u7b49\uff0c\u7531\u65bc\u4eba\u985e\u6240\u4f7f\u7528\u7684\u8a9e\u8a00\u6703 \u4e2d\u6587\u65b7\u8a5e\u7684\u7814\u7a76\u5df2\u6709\u76f8\u7576\u6b77\u53f2\uff0c\u4f46\u5728\u8fd1\u5e7e\u5e74\u4ecd\u9678\u7e8c\u65b0\u7684\u65b9\u6cd5\u63d0\u51fa\uff0c\u5e95\u4e0b\u6211\u5011 \u7528\u985e\u4f3c\u8a5e\u6027\u6a19\u793a(part-of-speech tagging)\u7684\u4f5c\u6cd5\uff0c\u7a31\u70ba\u300c\u89d2\u8272\u6a19\u793a\u300d (roles \u53ef\u63a5\u53d7\u7684\u65b7\u8a5e\u6548\u80fd\u3002 \u6b64\u5169\u7a2e\u4e0d\u540c\u7684\u9577\u8a5e\u512a\u5148\u65b7\u8a5e\u6cd5\uff0c\u7576\u65b7\u8a5e\u7684\u7d50\u679c\u4e0d\u540c\u6642\uff0c\u5247\u8868\u793a\u767c\u751f\u4ea4\u96c6\u578b\u6b67 \u5df2\u7d93\u65b7\u8a5e\u5b8c\u6210\uff0c\u4f8b\u5982\uff1a \u300c\u4eca\u5929\u662f\u91cd\u8981\u7684\u65e5\u5b50\u300d\u9019\u500b\u4e2d\u6587\u5b57\u4e32\uff0c\u5229\u7528\u5206\u985e\u554f\u984c\u5c07\u627e</td></tr><tr><td>\u7684\u81ea\u7136\u8a9e\u8a00\u76f8\u95dc\u7684\u9818\u57df\uff0c\u4f8b\u5982\uff1a\u554f\u7b54\u7cfb\u7d71\u3001\u81ea\u52d5\u6458\u8981\u3001\u6587\u4ef6\u6aa2\u7d22\u3001\u6a5f\u5668\u7ffb\u8b6f\u3001\u8a9e \u96a8\u8457\u793e\u6703\u4e0d\u65b7\u6539\u8b8a\uff0c\u800c\u6301\u7e8c\u5730\u5275\u9020\u51fa\u65b0\u7684\u7528\u8a9e\uff0c\u4e26\u4e14\u8a5e\u7684\u884d\u751f\u73fe\u8c61\u4e5f\u975e\u5e38\u5730\u666e \u5206\u5225\u5c31\u89e3\u6c7a\u6b67\u7fa9\u6027\u53ca\u672a\u77e5\u8a5e\u5169\u500b\u554f\u984c\u5206\u5225\u505a\u6587\u737b\u56de\u9867\u3002 tagging) \uff0c\u89d2\u8272\u6307\u7684\u662f\u5728\u672a\u77e5\u8a5e\u7684\u7d44\u6210\u6210\u5206\u3001\u4e0a\u4e0b\u6587\u4ee5\u53ca\u53e5\u5b50\u4e2d\u7684\u5176\u4ed6\u90e8\u5206\uff0c\u4e26 3. \u7cfb\u7d71\u67b6\u69cb \u7cfb\u7d71\u67b6\u69cb \u7cfb\u7d71\u67b6\u69cb \u7cfb\u7d71\u67b6\u69cb \u7fa9\uff0c\u5982\u8868 1 \u4e2d\u7684\u7b2c\u4e8c\u500b\u4f8b\u5b50\uff1a \u300c\u5373\u5c07\u4f86\u81e8\u6642\u300d\u5b57\u4e32\uff0c\u56e0\u70ba\u300c\u5c07\u300d\u53ef\u8207\u300c\u5373\u300d\u548c\u300c\u4f86\u300d \u51fa\u6bcf\u500b\u5b57\u5143\u6240\u5c0d\u61c9\u7684 BIES \u6a19\u7c64\uff0c\u5728\u6b64\u4f8b\u5b50\u4e2d\uff0c\u4e5f\u5c31\u662f\u300cBESBESBE\u300d \uff0c\u5247\u76f8\u7576</td></tr><tr><td>\u97f3\u8fa8\u8b58\u2026\u7b49\uff0c\u90fd\u9700\u8981\u5148\u8655\u7406\u4e2d\u6587\u65b7\u8a5e\uff0c\u53ef\u898b\u4e2d\u6587\u65b7\u8a5e\u662f\u500b\u76f8\u7576\u57fa\u790e\u4e14\u975e\u5e38\u91cd\u8981\u7684 \u904d\uff0c\u56e0\u6b64\u65b0\u8a5e\u6703\u4e0d\u65b7\u7684\u51fa\u73fe\uff0c\u8fad\u5178\u6c38\u9060\u7121\u6cd5\u56e0\u61c9\u65b0\u8a5e\u7522\u751f\u7684\u901f\u5ea6\uff0c\u6240\u4ee5\u6703\u51fa\u73fe\u672a \u9996\u5148\u5c31\u65b7\u8a5e\u6b67\u7fa9\u6027\u554f\u984c\uff0cM. Li \u7b49\u4eba [9] \u65bc 2003 \u5e74\u7684\u7814\u7a76\u4e2d\uff0c\u63d0\u51fa\u4e00\u7a2e\u975e \u4e14\u4f9d\u64da\u53e5\u5b50\u7684\u89d2\u8272\u5e8f\u5217\u4f86\u8fa8\u8b58\u51fa\u672a\u77e5\u8a5e\u3002\u5be6\u9a57\u90e8\u5206\u91dd\u5c0d\u4e2d\u570b\u4eba\u540d\u4ee5\u53ca\u5916\u570b\u7ffb\u8b6f\u540d \u6211\u5011\u63d0\u51fa\u7684\u7cfb\u7d71\u662f\u4ee5\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u4f86\u89e3\u6c7a\u4e2d\u6587\u65b7\u8a5e\u7684\u554f\u984c\uff0c\u4e26\u4e14\u900f\u904e\u5169 \u7d50\u5408\u6210\uff5b\u5373\u5c07\u3001\u5c07\u4f86\uff5d\u7b49\u4e0d\u540c\u7684\u8a5e\uff0c\u56e0\u6b64\u5c6c\u65bc\u4ea4\u96c6\u578b\u6b67\u7fa9\u5b57\u4e32\uff0c\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5 \u65bc\u662f\u5df2\u7d93\u65b7\u8a5e\u51fa\uff5b\u4eca\u5929\u3001\u662f\u3001\u91cd\u8981\u3001\u7684\u3001\u65e5\u5b50\uff5d\u7b49\u8a5e\u51fa\u4f86\u4e86\uff0c\u56e0\u6b64\u539f\u4f86\u7684\u4e2d\u6587\u5b57</td></tr><tr><td>\u5de5\u4f5c\u3002 \u77e5\u8a5e\u554f\u984c\uff0c\u65b7\u8a5e\u7cfb\u7d71\u5fc5\u9808\u80fd\u5920\u8655\u7406\u672a\u77e5\u8a5e\uff0c\u624d\u53ef\u63d0\u9ad8\u65b7\u8a5e\u7684\u6b63\u78ba\u6027\u3002 \u76e3\u7763\u5f0f(unsupervised)\u8a13\u7df4\u7684\u65b9\u6cd5\uff0c\u85c9\u7531\u8a13\u7df4 Na\u00efve Bayes \u5206\u985e\u5668\uff0c\u4f86\u89e3\u6c7a\u4e2d\u6587 \u7b49\u672a\u77e5\u8a5e\u505a\u6e2c\u8a66\uff0c\u4e26\u4e14\u9054\u5230\u4e0d\u932f\u7684\u6e96\u78ba\u7387\u4ee5\u53ca\u53ec\u56de\u7387\u3002 \u968e\u6bb5\u300c\u7279\u88fd\u5316\u300d (specialization)\u7684\u65b9\u5f0f\u4f86\u52a0\u5f37\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u65b7\u8a5e\u6548\u80fd\u3002\u7b2c \u6703\u65b7\u8a5e\u6210\u300c\u5373\u5c07\uff0f\u4f86\u81e8\uff0f\u6642\u300d \uff0c\u800c\u53cd\u5411\u9577\u8a5e\u512a\u5148\u5247\u65b7\u8a5e\u6210\u300c\u5373\uff0f\u5c07\u4f86\uff0f\u81e8\u6642\u300d \u3002 \u4e32\u4fbf\u53ef\u4ee5\u8f49\u63db\u6210\u300c\u4eca\u5929\uff0f\u662f\uff0f\u91cd\u8981\uff0f\u7684\uff0f\u65e5\u5b50\u300d\u7684\u65b7\u8a5e\u7d50\u679c\u3002</td></tr><tr><td>\u6240\u8b02\u7684\u300c\u4e2d\u6587\u65b7\u8a5e\u300d\u5c31\u662f\u5c07\u4e00\u9023\u4e32\u7684\u4e2d\u6587\u300c\u5b57\u4e32\u300d\u8f49\u63db\u6210\u300c\u8a5e\u4e32\u300d\u7684\u7d44\u5408\u3002 \u8fd1\u5e74\u4f86\u7684\u65b7\u8a5e\u7cfb\u7d71\u50be\u5411\u65bc\u6a5f\u5668\u5b78\u7fd2\u5f0f(machine learning-based)\u6f14\u7b97\u6cd5\u4f86\u89e3 \u65b7\u8a5e\u7684\u4ea4\u96c6\u578b\u6b67\u7fa9\u554f\u984c\uff0c\u5be6\u9a57\u7d50\u679c\u53ef\u9054\u5230 94.13% \u7684\u6e96\u78ba\u7387\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u89e3\u6c7a\u7d44 \u8fd1\u5e74\u4f86\u7684\u7814\u7a76\u4e3b\u8981\u8da8\u5411\u65bc\u6a5f\u5668\u5b78\u7fd2\u5f0f\u7684\u65b9\u6cd5\u4f86\u8655\u7406\u4e2d\u6587\u65b7\u8a5e\uff0c\u4f8b\u5982 \u4e00\u968e\u6bb5\u7279\u88fd\u5316\uff0c\u6211\u5011\u7d50\u5408\u4e86\u9577\u8a5e\u512a\u5148\u6cd5\u7684\u7d50\u679c\u4f86\u589e\u52a0\u89c0\u6e2c\u7b26\u865f\u7684\u8cc7\u8a0a\uff0c\u4ee5\u300c\u64f4\u5145 \u8868 1 \u9577\u8a5e\u512a\u5148\u6cd5\u7684\u4e0d\u540c\u8b8a\u5f62 \u8868 2 \u5b57\u5143\u300c\u4e2d\u300d\u53ef\u51fa\u73fe\u5728\u8a5e\u7684\u4efb\u4f55\u4f4d\u7f6e</td></tr><tr><td>\u4f8b\u5982\uff1a \u300c\u6211\u6628\u5929\u53bb\u53f0\u5317\u300d\u9019\u500b\u4e2d\u6587\u53e5\u5b50\uff0c\u900f\u904e\u4e2d\u6587\u65b7\u8a5e\u7684\u8655\u7406\u5f8c\u8b8a\u6210\u300c\u6211\uff0f\u6628\u5929 \u6c7a\u4e2d\u6587\u65b7\u8a5e\u7684\u554f\u984c\uff0c\u4f8b\u5982 Maximum Entropy (ME) [22]\u3001Support Vector Machine \u5408\u578b\u6b67\u7fa9\u6bd4\u89e3\u6c7a\u4ea4\u96c6\u578b\u6b67\u7fa9\u66f4\u52a0\u56f0\u96e3\uff0c\u4e3b\u8981\u7684\u539f\u56e0\u662f\uff0c\u8981\u89e3\u6c7a\u7d44\u5408\u578b\u6b67\u7fa9\u5247\u9700\u8981 Maximum Entropy(ME)[22] \u4ee5\u53ca Conditional Random Field(CRF)[20] \u7b49\uff0c \u89c0\u6e2c\u7b26\u865f\u300d \uff1b\u7b2c\u4e8c\u968e\u6bb5\u7279\u88fd\u5316\uff0c\u5247\u662f\u900f\u904e\u8a5e\u5f59\u5f0f(Lexicalized HMM)\u7684\u7279\u88fd\u5316\u904e \u4f8b\u53e5 \u6b63\u5411\u9577\u8a5e\u512a\u5148 \u6b63\u5411\u9577\u8a5e\u512a\u5148 \u6b63\u5411\u9577\u8a5e\u512a\u5148 \u6b63\u5411\u9577\u8a5e\u512a\u5148 \u53cd\u5411\u9577\u8a5e\u512a\u5148 \u53cd\u5411\u9577\u8a5e\u512a\u5148 \u53cd\u5411\u9577\u8a5e\u512a\u5148 \u53cd\u5411\u9577\u8a5e\u512a\u5148 B \u4e2d\u91ab</td></tr><tr><td>\uff0f\u53bb\uff0f\u53f0\u5317\u300d \uff0c\u4e5f\u5c31\u662f\u5c07\uff5b\u6211\u3001\u6628\u3001\u5929\u3001\u53bb\u3001\u53f0\u3001\u5317\uff5d\u5b57\u4e32\u8f49\u6210\uff5b\u6211\u3001\u6628\u5929\u3001\u53bb\u3001 (SVM) [2, 6]\u3001Transformation-Based Learning Algorithm (TBL) [11]\u3001Hidden Markov \u4f9d\u8cf4\u66f4\u591a\u7684\u5167\u6587\u8cc7\u8a0a\uff0c\u5982\u53e5\u6cd5\u5206\u6790(syntactic) \u3001\u8a9e\u610f\u5206\u6790(semantic) \u4ee5\u53ca\u524d \u9019\u4e9b\u7d71\u8a08\u5f0f\u7684\u5b78\u7fd2\u6f14\u7b97\u6cd5\u90fd\u662f\u8f49\u6210\u5b57\u5143\u5206\u985e\u554f\u984c(character classification)\u4f86\u8655 \u7a0b\uff0c\u4ee5\u300c\u64f4\u5145\u72c0\u614b\u7b26\u865f\u300d \u3002 \u5373\u5c07\u7562\u696d \u5373\u5c07\uff0f\u7562\u696d I \u570b\u6c11\u4e2d\u5b78 \u5373\u5c07\uff0f\u7562\u696d E \u96c6\u4e2d</td></tr><tr><td>\u53f0\u5317\uff5d\u7684\u8a5e\u4e32\u7d44\u5408\u3002\u50b3\u7d71\u4e0a\uff0c\u8655\u7406\u4e2d\u6587\u65b7\u8a5e\u6703\u9047\u5230\u7684\u554f\u984c\uff0c\u5927\u81f4\u53ef\u6b78\u7d0d\u70ba\u5169\u9ede\uff0c Model (HMM) [2, 11, 23, 25] \u7b49\u7b49\uff0c\u4e26\u4e14\u986f\u793a\u4e86\u4f7f\u7528\u6a5f\u5668\u5b78\u7fd2\u5f0f\u6f14\u7b97\u6cd5\u505a\u4e2d\u6587\u65b7 \u56e0\u5f8c\u679c\u7684\u8cc7\u8a0a (pragmatic information) \u7b49\uff0c\u624d\u80fd\u6b63\u78ba\u7684\u89e3\u6c7a\u9019\u985e\u7684\u6b67\u7fa9\u554f\u984c\u30021999 \u7406\u4e2d\u6587\u65b7\u8a5e\u554f\u984c\uff0c\u4e26\u4e14\u4f7f\u7528\u4e86\u6578\u7a2e\u985e\u4f3c\u7684\u7279\u5fb5\uff0c\u5982\u76ee\u524d\u5b57\u5143\u3001\u52a0\u4e0a\u524d\u5f8c\u5404\u4e00\u5b57\u5143\u3001 \u56e0\u6b64\u6211\u5011\u7684\u7cfb\u7d71\u67b6\u69cb\u4e3b\u8981\u53ef\u5206\u70ba\u5169\u500b\u90e8\u5206\uff0c\u7b2c\u4e00\u90e8\u4efd\uff1a\u6211\u5011\u7a31\u4e4b\u70ba \u5373\u5c07\u4f86\u81e8\u6642 \u5373\u5c07\uff0f\u4f86\u81e8\uff0f\u6642 \u5373\uff0f\u5c07\u4f86\uff0f\u81e8\u6642 S \u5728 \u8cc7\u6599\u5eab \u4e2d</td></tr><tr><td>\u4e00\u662f\u300c\u6b67\u7fa9\u6027\u300d (ambiguity)\u554f\u984c\uff0c\u4e8c\u662f\u300c\u672a\u77e5\u8a5e\u300d (unknown word)\u554f\u984c\u3002\u6b67\u7fa9 \u8a5e\uff0c\u78ba\u5be6\u53ef\u4ee5\u9054\u5230\u5f88\u9ad8\u7684\u65b7\u8a5e\u6e96\u78ba\u7387\u3002 \u52a0\u4e0a\u524d\u5f8c\u5404\u5169\u5b57\u5143\u7b49\uff0c\u4f86\u7576\u4f5c\u6a21\u578b\u7684\u5c6c\u6027\u3002\u800c C. L. Goh \u7b49\u4eba\u5247\u4f7f\u7528 Support \u300cM-HMM\u300d \uff0c\u4e5f\u5c31\u662f\u7d50\u5408\u9577\u8a5e\u512a\u5148\u6cd5\u65bc\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u4e2d\uff0c\u8b93\u8a13\u7df4\u4e4b\u6a21\u578b\u589e \u53e6\u5916\uff0c\u7531\u65bc\u9577\u8a5e\u512a\u5148\u6cd5\u5c6c\u65bc\u8fad\u5178\u6bd4\u5c0d\u5f0f\u65b7\u8a5e\u65b9\u6cd5\uff0c\u53ea\u6709\u5728\u8fad\u5178\u4e2d\u7684\u8a5e\u624d\u6709\u53ef 3.3. \u96b1\u85cf\u5f0f\u99ac\u53ef\u592b \u96b1\u85cf\u5f0f\u99ac\u53ef\u592b \u96b1\u85cf\u5f0f\u99ac\u53ef\u592b \u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b \u6a21\u578b \u6a21\u578b \u6a21\u578b</td></tr><tr><td>\u6027\u554f\u984c\u5373\u662f\u540c\u4e00\u500b\u4e2d\u6587\u5b57\u4e32\uff0c\u65bc\u4e0d\u540c\u7684\u6587\u7ae0\u7576\u4e2d\uff0c\u5b58\u5728\u4e0d\u540c\u7684\u65b7\u8a5e\u7d50\u679c\uff0c\u56e0\u6b64\u5bb9 \u672c\u7814\u7a76\u4f7f\u7528\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(Hidden Markov Model, HMM)\u4f86\u89e3\u6c7a\u4e2d\u6587 Vector Machine(SVM)[6] \u4f86\u89e3\u6c7a\u4e2d\u6587\u65b7\u8a5e\u7684\u554f\u984c\uff0c\u8a72\u7bc7\u7814\u7a76\u7d50\u5408\u8fad\u5178\u6bd4\u5c0d\u5f0f \u52a0\u65b7\u8a5e\u6b67\u7fa9\u6027\u8207\u672a\u77e5\u8a5e\u7684\u8cc7\u8a0a\uff0c\u85c9\u6b64\u4ee5\u6539\u5584\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u8655\u7406\u4e2d\u6587\u65b7\u8a5e\u7684\u6b63 \u80fd\u6b63\u78ba\u65b7\u51fa\uff0c\u6240\u4ee5\u7121\u6cd5\u89e3\u6c7a\u672a\u77e5\u8a5e\u554f\u984c\u3002\u7576\u9047\u5230\u672a\u77e5\u8a5e\u6642\uff0c\u6b63\u5411\u9577\u8a5e\u512a\u5148\u8207\u53cd\u5411 \u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u53ef\u4ee5\u8996\u70ba\u4e00\u500b\u96d9\u5c64\u7684\u96a8\u6a5f\u5e8f\u5217\uff0c\u5305\u542b\u4e86\u96b1\u85cf\u5c64\u7684\u72c0\u614b\u5e8f</td></tr><tr><td>\u6613\u9020\u6210\u65b7\u8a5e\u4e0a\u7684\u932f\u8aa4\u3002\u6b67\u7fa9\u578b\u614b\u5927\u81f4\u4e0a\u53ef\u4ee5\u5206\u70ba\u5169\u985e\uff1a \u65b7\u8a5e\u7684\u554f\u984c\u3002\u96d6\u7136\u5df2\u6709\u6578\u7bc7\u7814\u7a76\u540c\u6a23\u4f7f\u7528\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u4f86\u8655\u7406\u65b7\u8a5e\u554f\u984c [2, \u65b9\u6cd5-\u9577\u8a5e\u512a\u5148\u6cd5\uff0c\u5229\u7528\u9577\u8a5e\u512a\u5148\u6cd5\u7684\u6b67\u7fa9\u6027\u4ee5\u53ca\u672a\u77e5\u8a5e\u7684\u8cc7\u8a0a\uff0c\u4f86\u52a0\u5f37 SVM \u78ba\u6027\uff1b\u7b2c\u4e8c\u90e8\u5206\uff1a\u6211\u5011\u7a31\u4e4b\u70ba\u300cLexicalized M-HMM\u300d \uff0c\u9019\u90e8\u5206\u900f\u904e\u5169\u7a2e\u4e0d\u540c\u7684 \u9577\u8a5e\u512a\u5148\u90fd\u5c07\u65b7\u8a5e\u6210\u55ae\u4e00\u4e2d\u6587\u5b57\u5143\u3002\u4f8b\u5982\uff1a \u300c\u9d3b\u6d77\u8463\u4e8b\u9577\u90ed\u53f0\u9298\u300d\u5b57\u4e32\uff0c\u7531\u65bc\u8fad \u5217(state sequence)\u548c\u53ef\u89c0\u5bdf\u5c64\u7684\u89c0\u6e2c\u5e8f\u5217(observation sequence) \u3002\u96b1\u85cf\u5c64\u662f\u7121</td></tr><tr><td>\u4ea4\u96c6\u578b\u6b67\u7fa9(overlapping ambiguity) 11, 23, 25]\uff0c\u4f46\u4f7f\u7528\u50b3\u7d71\u7684\u4f5c\u6cd5\uff0c\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u5728\u89e3\u6c7a\u4e2d\u6587\u65b7\u8a5e\u7684\u554f\u984c\u4e0a\uff0c \u7684\u7279\u5fb5\u5c6c\u6027\u4ee5\u6539\u5584\u65b7\u8a5e\u6548\u80fd\u3002\u53e6\u5916\u4e5f\u6709\u4f7f\u7528\u611f\u77e5\u6a5f(Perceptron)[10] \u7684\u65b9\u6cd5\u505a \u6e96\u5247(criteria)\u4f86\u6c7a\u5b9a\u7279\u88fd\u8a5e(specialized words) \uff0c\u4e26\u4ee5\u5c6c\u65bc\u7279\u88fd\u8a5e\u4e4b\u89c0\u6e2c\u7b26\u865f \u5178\u4e2d\u672a\u6536\u9304\uff5b\u9d3b\u6d77\u3001\u90ed\u53f0\u9298\uff5d\u7b49\u8a5e\uff0c\u56e0\u6b64\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u8207\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u90fd\u540c \u6cd5\u76f4\u63a5\u89c0\u5bdf\u5f97\u5230\u7684\uff0c\u4f46\u53ef\u4ee5\u5f9e\u53e6\u4e00\u500b\u53ef\u89c0\u5bdf\u7684\u89c0\u6e2c\u5e8f\u5217\u4e4b\u96a8\u6a5f\u904e\u7a0b\u7684\u96c6\u5408\u89c0\u5bdf\u5f97</td></tr><tr><td>\u7121\u6cd5\u9054\u5230\u8f03\u597d\u7684\u65b7\u8a5e\u6548\u80fd(F-measure \u7d04 80%) \uff0c\u56e0\u6b64\u9019\u4e9b\u7814\u7a76 [2, 11, 23] \u4fbf\u7d50 \u5408\u4e86\u5176\u4ed6\u6a5f\u5668\u5b78\u7fd2\u6f14\u7b97\u6cd5\uff0c\u4ee5\u589e\u52a0\u65b7\u8a5e\u7684\u6548\u80fd\u3002\u6211\u5011\u7684\u7814\u7a76\u76ee\u7684\u662f\u5e0c\u671b\u53ea\u4f7f\u7528\u96b1 \u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7576\u6210\u4e3b\u8981\u7684\u6f14\u7b97\u6cd5\uff0c\u4e26\u4e14\u61c9\u7528\u300c\u7279\u88fd\u5316\u300d (specialization)\u7684\u6982\u5ff5 \u65b7\u8a5e\uff0c\u8a72\u7bc7\u7814\u7a76\u8a8d\u70ba Perceptron \u65b9\u6cd5\u96d6\u7136\u8207 SVM \u985e\u4f3c\uff0c\u4e0d\u904e\u6548\u80fd\u537b\u8f03 SVM \u5dee \u4e00\u4e9b\uff0c\u4f46\u7531\u65bc\u5176\u8a13\u7df4\u7684\u901f\u5ea6\u975e\u5e38\u5feb\uff0c\u56e0\u6b64\u4ed6\u5011\u7cfb\u7d71\u63d0\u51fa\u7684\u4e3b\u8981\u8ca2\u737b\u5c31\u662f\u4e00\u500b\u901f\u5ea6 \u5feb\u4e14\u6548\u80fd\u4e0d\u81f3\u65bc\u5dee\u592a\u591a\u7684\u65b7\u8a5e\u65b9\u6cd5\u3002 \u505a\u7279\u88fd\u5316\uff0c\u900f\u904e\u64f4\u5145\u72c0\u614b\u7b26\u865f\u800c\u518d\u6b21\u52a0\u5f37\u65b7\u8a5e\u6e96\u78ba\u7387\u3002 3.1. \u9577\u8a5e\u512a\u5148\u6cd5 \u9577\u8a5e\u512a\u5148\u6cd5 \u9577\u8a5e\u512a\u5148\u6cd5 \u9577\u8a5e\u512a\u5148\u6cd5 \u9577\u8a5e\u512a\u5148\u6cd5(Maximum Matching Algorithm, MM)\u662f\u6700\u7c21\u55ae\u4e5f\u6700\u5ee3\u6cdb\u4f7f\u7528\u7684 \u6a23\u6703\u65b7\u8a5e\u6210\u300c\u9d3b\uff0f\u6d77\uff0f\u8463\u4e8b\u9577\uff0f\u90ed\uff0f\u53f0\uff0f\u9298\u300d \u3002 3.2. BIES \u5206\u985e\u554f\u984c \u5206\u985e\u554f\u984c \u5206\u985e\u554f\u984c \u5206\u985e\u554f\u984c \u5229\u7528\u6a5f\u5668\u5b78\u7fd2\u5f0f\u6f14\u7b97\u6cd5\u4f86\u89e3\u4e2d\u6587\u65b7\u8a5e\u7684\u554f\u984c\u6642\uff0c\u4e00\u822c\u7684\u4f5c\u6cd5\u662f\u5c07\u4e2d\u6587\u65b7\u8a5e\u554f \u51fa\u3002\u56e0\u6b64\uff0c\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u662f\u4e00\u500b\u99ac\u53ef\u592b\u93c8\u7684\u6a5f\u7387\u51fd\u6578\uff0c\u7121\u6cd5\u76f4\u63a5\u89c0\u5bdf\u7684\u96b1\u85cf \u5c64\u5c31\u662f\u4e00\u500b\u6709\u9650\u72c0\u614b\u7684\u99ac\u53ef\u592b\u93c8\uff0c\u5176\u521d\u59cb\u7684\u72c0\u614b\u6a5f\u7387\u5206\u4f48\u4ee5\u53ca\u72c0\u614b\u4e4b\u9593\u7684\u8f49\u79fb\u6a5f \u7387\u7531\u72c0\u614b\u521d\u59cb\u6a5f\u7387\u5411\u91cf \u220f \u548c\u72c0\u614b\u8f49\u79fb\u6a5f\u7387\u77e9\u9663 A \u4f86\u6c7a\u5b9a\uff0c\u53e6\u5916\u9084\u9700\u5b9a\u7fa9\u89c0\u6e2c\u7b26 \u4ee4 x, y, z \u6240\u7d44\u6210\u7684\u5b57\u4e32\uff0c\u8fad\u5178\u4e2d\u7684\u8a5e\u542b\u6709\u300c\u4e0d\u3001\u4e0d\u53ef\u3001\u53ef\u4ee5\u300d \uff0c \u300c\u4e0d\u53ef\u4ee5\u300d\u6240\u7d44\u6210\u7684 \u4f86\u63d0\u5347\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u6e96\u78ba\u7387\u3002\u6211\u5011\u7684\u4f5c\u6cd5\u662f\u7d66\u4e88\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u66f4\u591a\u7684 \u6709\u4e9b\u7814\u7a76\u70ba\u4e86\u52a0\u5f37\u5b78\u7fd2\u6f14\u7b97\u6cd5\u7684\u65b7\u8a5e\u6548\u80fd\uff0c\u5247\u662f\u7d50\u5408\u4e86\u6578\u500b\u5b78\u7fd2\u6a21\u578b\uff0c\u63a1\u7528 \u8fad\u5178\u6bd4\u5c0d\u5f0f\u7684\u65b7\u8a5e\u65b9\u6cd5\uff0c\u5176\u65b7\u8a5e\u7684\u7b56\u7565\u70ba\u7531\u53e5\u5b50\u7684\u4e00\u7aef\u958b\u59cb\uff0c\u8a66\u8457\u6bd4\u5c0d\u51fa\u5728\u8fad\u5178 \u984c\u8f49\u63db\u6210\u5206\u985e\u7684\u554f\u984c\uff0c\u800c\u6700\u5e38\u88ab\u4f7f\u7528\u7684\u65b9\u6cd5\u5c31\u662f\u8f49\u63db\u6210\u5b57\u5143\u5206\u985e\u554f\u984c(character \u865f\u6a5f\u7387\u77e9\u9663 B \uff0c\u5132\u5b58\u5404\u500b\u89c0\u6e2c\u7b26\u865f\u5728\u4e0d\u540c\u7684\u72c0\u614b\u4e0b\u7684\u6a5f\u7387\u503c\u3002</td></tr><tr><td>\u5b57\u4e32\uff0c\u5728\u4e0b\u5217\u53e5\u5b50\u4e2d\uff0c\u56e0\u5176\u4e0a\u4e0b\u6587\u7684\u4e0d\u540c\u800c\u7522\u751f\u4e0d\u540c\u7684\u65b7\u8a5e\u7d50\u679c\uff1a \u300c\u4e0d\uff0f\u53ef \u8cc7\u8a0a\uff0c\u5728\u5b8c\u5168\u4e0d\u4fee\u6539\u6a21\u578b\u4e4b\u8a13\u7df4\u53ca\u6e2c\u8a66\u904e\u7a0b\u7684\u524d\u63d0\u4e0b\uff0c\u900f\u904e\u5169\u968e\u6bb5\u7279\u88fd\u5316\u7684\u65b9 \u6df7\u5408\u5f0f\u4f5c\u6cd5\u4f86\u8655\u7406\u65b7\u8a5e\u554f\u984c\uff0c\u5982 M. Asahara \u7b49\u4eba [1] \u4ee5\u53ca N. Xue \u7b49\u4eba [23] \u7684 \u4e2d\u6700\u9577\u7684\u8a5e\uff0c\u7576\u4f5c\u65b7\u8a5e\u7d50\u679c\uff0c\u63a5\u8457\u53bb\u9664\u6b64\u8a5e\u5f8c\uff0c\u5269\u4e0b\u7684\u90e8\u5206\u7e7c\u7e8c\u505a\u9577\u8a5e\u512a\u5148\u6cd5\u65b7 classification problem) \uff0c\u5c07\u6bcf\u500b\u5b57\u5143\u90fd\u7d66\u4e88\u5176\u5c0d\u61c9\u7684\u985e\u5225\uff0c\u900f\u904e\u5b57\u5143\u985e\u5225\u4f86\u505a\u5206</td></tr><tr><td>\u4ee5\uff0f\u5fd8\u8a18\u300d \u3001 \u300c\u4e0d\u53ef\uff0f\u4ee5\uff0f\u71df\u5229\uff0f\u70ba\uff0f\u76ee\u7684\u300d \u3002 \u5f0f\uff0c\u5206\u5225\u70ba\u64f4\u5145\u300c\u89c0\u6e2c\u7b26\u865f\u300d \uff0c\u4ee5\u53ca\u64f4\u5145\u300c\u72c0\u614b\u7b26\u865f\u300d\u7684\u65b9\u5f0f\uff0c\u5927\u5927\u5730\u6539\u5584\u4e86\u96b1 \u6599\u5eab\uff0c\u7522\u751f\u6240\u6709\u55ae\u4e00\u5b57\u5143\u4e4b\u5df2\u77e5\u8a5e\u7684\u5075\u6e2c\u898f\u5247\u3002\u6b64\u968e\u6bb5\u7684\u7814\u7a76\u53ea\u80fd\u5075\u6e2c\u51fa\u6240\u6709\u7684 \u7814\u7a76\uff0c\u70ba\u4e86\u52a0\u5f37 ME \u65b7\u8a5e\u7d50\u679c\uff0c\u9019\u5169\u7bc7\u7814\u7a76\u5247\u7d50\u5408\u4e86 SVM\u3001CRF [1] \u4ee5\u53ca TBL [23] \u8a5e\uff0c\u76f4\u5230\u53e5\u5b50\u7684\u53e6\u4e00\u7aef\u7d50\u675f\u70ba\u6b62\u3002\u4e00\u822c\u4f86\u8aaa\uff0c\u5982\u679c\u6240\u4f7f\u7528\u7684\u8fad\u5178\u5920\u5927\uff0c\u9577\u8a5e\u512a\u5148 \u985e\uff0c\u9019\u4e9b\u5b57\u5143\u7684\u985e\u5225\u7531\u51fa\u73fe\u5728\u4e2d\u6587\u8a5e\u7576\u4e2d\u7684\u7279\u5b9a\u4f4d\u7f6e\u4f86\u6c7a\u5b9a\uff0c\u4e00\u500b\u5b57\u5143\u7684\u4f4d\u7f6e\u53ef</td></tr><tr><td>\u7d44\u5408\u578b\u6b67\u7fa9(covering ambiguity) \u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u65b7\u8a5e\u6e96\u78ba\u6027\u3002 \u55ae\u4e00\u5b57\u5143\u7684\u7d50\u679c\uff0c\u4e26\u672a\u771f\u6b63\u5c07\u672a\u77e5\u8a5e\u64f7\u53d6\u51fa\u4f86\u30022002 \u5e74\u7684\u7814\u7a76 [5] \uff0c\u5247\u662f\u4f7f\u7528 \u7b49\u4f5c\u6cd5\uff0c\u4f7f\u7528\u6df7\u5408\u5f0f\u65b9\u6cd5\u7684\u7d50\u679c\u63d0\u5347\u65b7\u8a5e\u6e96\u78ba\u7387\u3002\u53e6\u5916\uff0c\u8a31\u591a\u7814\u7a76\u4e5f\u4f7f\u7528\u96b1\u85cf\u5f0f \u6cd5\u65b7\u8a5e\u53ef\u9054\u5230\u8d85\u904e 90 % \u4ee5\u4e0a\u7684\u65b7\u8a5e\u6e96\u78ba\u7387\u3002 \u4ee5\u5206\u70ba\u4f4d\u65bc\u8a5e\u7684\u958b\u59cb(beginning) \u3001\u4f4d\u65bc\u8a5e\u7684\u4e2d\u9593 (intermediate) \u3001\u4f4d\u65bc\u8a5e\u7684\u7d50</td></tr><tr><td>\u65bc\u7b2c\u4e00\u968e\u6bb5\u4e2d\uff0c\u70ba\u4e86\u64f4\u5145\u89c0\u6e2c\u7b26\u865f\uff0c\u6211\u5011\u4f7f\u7528\u6700\u7c21\u55ae\u4e5f\u6700\u5e38\u88ab\u4f7f\u7528\u7684\u8fad\u5178\u6bd4 \u5c0d\u5f0f\u65b7\u8a5e\u6f14\u7b97\u6cd5-\u300c\u9577\u8a5e\u512a\u5148\u6cd5\u300d (maximum matching algorithm) \uff0c\u4f86\u589e\u52a0\u984d\u5916 \u4eba\u5de5\u52a0\u4e0a\u4e00\u4e9b\u7d71\u8a08\u7684\u65b9\u6cd5\u4f86\u5efa\u7acb\u64f7\u53d6\u898f\u5247\uff0c\u5c07\u6240\u6709\u88ab\u5075\u6e2c\u51fa\u5c6c\u65bc\u672a\u77e5\u8a5e\u90e8\u5206\u7684\u55ae \u4e00\u5b57\u8a5e\uff0c\u900f\u904e\u64f7\u53d6\u898f\u5247\u4ee5\u5408\u4f75\u9019\u4e9b\u55ae\u4e00\u5b57\u8a5e\u800c\u6210\u70ba\u672a\u77e5\u8a5e\u3002\u5be6\u9a57\u4e2d\u6e2c\u8a66 1,160 \u500b \u99ac\u53ef\u592b\u6a21\u578b\u4f86\u8655\u7406\u65b7\u8a5e\u554f\u984c\u3002\u5982 HHMM [25]\u7cfb\u7d71\uff0c\u4fbf\u4f7f\u7528\u4e86\u4e94\u5c64\u7684\u96b1\u85cf\u99ac\u53ef\u592b \u6a21\u578b\uff0c\u6839\u64da\u4e0d\u540c\u7684\u76ee\u7684\u5404\u81ea\u8a13\u7df4\u51fa\u5404\u500b\u6a21\u578b\uff0c\u6700\u5f8c\u518d\u6574\u5408\u6210\u65b7\u8a5e\u7cfb\u7d71\u3002\u800c \u9577\u8a5e\u512a\u5148\u6cd5\u4f9d\u7167\u6bd4\u5c0d\u65b9\u5411\u7684\u4e0d\u540c\u53c8\u53ef\u5206\u70ba\u5169\u7a2e\u4e0d\u540c\u7684\u8b8a\u5f62\uff0c\u7b2c\u4e00\u7a2e\u662f\u300c\u6b63\u5411 \u9577\u8a5e\u512a\u5148\u6cd5\u300d (Forward Maximum Matching, FMM) \uff0c\u5373\u7531\u53e5\u5b50\u958b\u982d\u7684\u7b2c\u4e00\u500b\u5b57\u5143 \u5c3e(end)\u4ee5\u53ca\u7531\u55ae\u4e00\u5b57\u5143\u7d44\u6210\u7684\u8a5e(single-character)\u7b49\u56db\u7a2e\u985e\u5225\uff0c\u56e0\u6b64\u4e5f\u7a31\u70ba \u300cBIES \u5206\u985e\u554f\u984c\u300d \u3002 \u4ee4 x, y \u7b49\u5169\u500b\u4e0d\u540c\u7684\u8a5e\u6240\u7d44\u6210\uff0c\u56e0\u6b64 xy \u7a31\u70ba\u300c\u7d44\u5408\u578b\u6b67\u7fa9\u5b57\u4e32\u300d \u3002\u4f8b\u5982\uff1a \u300c\u624d\u80fd\u300d \u7684\u8cc7\u8a0a\u65bc\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u4e2d\uff0c\u4f7f\u5f97\u6a21\u578b\u64c1\u6709\u66f4\u591a\u7684\u65b7\u8a5e\u8cc7\u8a0a\u505a\u5b78\u7fd2\u3002\u7b2c\u4e8c\u968e\u6bb5 \u672a\u77e5\u8a5e\uff0c\u7d50\u679c\u9054\u5230 89 % \u7684\u64f7\u53d6\u6e96\u78ba\u7387\u3002\u53e6\u5916\u65bc 2003 \u5e74\u7684\u7814\u7a76 [13] \u4e2d\uff0c\u540c\u6a23\u505a HMM+SVM [2]\u3001HMM+TBL [11] \u7b49\u5169\u7bc7\u7814\u7a76\uff0c\u5247\u4f7f\u7528\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u65b7\u8a5e \u958b\u59cb\uff0c\u7531\u5de6\u800c\u53f3\u9010\u4e00\u6383\u7784\uff0c\u6bd4\u5c0d\u51fa\u5728\u8fad\u5178\u4e2d\u6700\u9577\u7684\u8a5e\uff0c\u4ee5\u7576\u4f5c\u65b7\u8a5e\u7684\u7d50\u679c\uff0c\u4e26\u76f4 \u7406\u8ad6\u4e0a\u4e2d\u6587\u5b57\u5143\u53ef\u4ee5\u5b58\u5728\u65bc\u4e2d\u6587\u8a5e\u7684\u4efb\u4f55\u4f4d\u7f6e\u4e0a\uff0c\u4f8b\u5982\u8868 2 \u7684\u4f8b\u5b50\uff0c\u5b57\u5143</td></tr><tr><td>\u4e8c\u500b\u5b57\u6240\u7d44\u6210\u7684\u5b57\u4e32\uff0c\u8fad\u5178\u4e2d\u7684\u8a5e\u6709\u300c\u624d\u3001\u80fd\u3001\u624d\u80fd\u300d \uff0c\u5728\u4e0b\u5217\u53e5\u5b50\u4e2d\u300c\u624d \u64f4\u5145\u72c0\u614b\u7b26\u865f\u7684\u65b9\u5f0f\uff0c\u6211\u5011\u5247\u4f7f\u7528\u8a5e\u5f59\u5f0f\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(Lexicalized HMM) \u64f7\u53d6\u672a\u77e5\u8a5e\u7684\u7814\u7a76\uff0c\u8a72\u7814\u7a76\u4e2d\u5c07\u6240\u6709\u7a2e\u985e\u7684\u672a\u77e5\u8a5e\u7684\u69cb\u8a5e\u65b9\u5f0f\u4ee5 \u7d50\u679c\u7576\u6210\u662f\u4e00\u500b\u5c6c\u6027\uff0c\u4e26\u5206\u5225\u4f7f\u7528 SVM \u4ee5\u53ca TBL \u4f86\u7576\u6210\u4e3b\u8981\u7684\u6f14\u7b97\u6cd5\u505a\u65b7\u8a5e\uff0c \u5230\u53e5\u5b50\u7684\u7d50\u5c3e\u800c\u7d50\u675f\u3002\u76f8\u53cd\u5730\uff0c\u53e6\u4e00\u7a2e\u9577\u8a5e\u512a\u5148\u6cd5\u7684\u8b8a\u5f62\u5247\u662f \u300c\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u300d \u300c\u4e2d\u300d\u53ef\u4ee5\u5b58\u5728\u65bc\u8a5e\u7684\u958b\u59cb(B) \u3001\u8a5e\u7684\u4e2d\u9593(I) \u3001\u8a5e\u7684\u7d50\u5c3e(E) \u3001\u4ee5\u53ca\u55ae\u4e00\u5b57</td></tr><tr><td>\u80fd\u300d\u7d44\u6210\u7684\u5b57\u4e32\uff0c\u5c07\u7522\u751f\u4e0d\u540c\u7684\u65b7\u8a5e\u7d50\u679c\uff1a \u300c\u4ed6\uff0f\u624d\u80fd\uff0f\u975e\u51e1\u300d \u3001 \u300c\u53ea\u6709\uff0f\u4ed6 \u7684\u6982\u5ff5\uff0c\u4e5f\u5c31\u662f\u53ea\u6839\u64da\u67d0\u4e9b\u7279\u88fd\u8a5e(specialized words)\u4f86\u505a\u7279\u88fd\u5316\uff0c\u5c07\u72c0\u614b\u505a context free grammar \u8868\u793a\u51fa\u4f86\uff0c\u4e26\u642d\u914d bottom-up merging algorithm \u4f86\u89e3\u6c7a\u5927\u90e8 \u4ee5\u9054\u5230\u8f03\u4f73\u7684\u65b7\u8a5e\u7d50\u679c\u3002\u6b64\u5169\u7bc7\u7814\u7a76\u65bc\u5be6\u9a57\u4e2d\u4e5f\u5217\u51fa\u53ea\u4f7f\u7528\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u505a (Backward Maximum Matching, BMM) \uff0c\u7531\u53e5\u5b50\u7684\u6700\u5f8c\u4e00\u500b\u5b57\u5143\u958b\u59cb\u6383\u7784\uff0c\u5f9e\u53f3 \u5143\u7684\u8a5e(S) \u3002\u6240\u4ee5 BIES \u5206\u985e\u6240\u8981\u89e3\u6c7a\u7684\u554f\u984c\u4e5f\u5c31\u662f\u6c7a\u5b9a\u6bcf\u500b\u5b57\u5143\u7684\u6b63\u78ba\u985e\u5225\u3002</td></tr><tr><td>\uff0f\u624d\uff0f\u80fd\uff0f\u52dd\u4efb\u300d \u3002 \u5ef6\u4f38\uff0c\u4f86\u63d0\u5347\u7cfb\u7d71\u65b7\u8a5e\u7684\u6548\u80fd\u3002 \u5206\u7d71\u8a08\u7279\u6027\u4f4e\u7684\u672a\u77e5\u8a5e\u64f7\u53d6\u554f\u984c\u3002\u5be6\u9a57\u6548\u80fd\u9054\u5230 75 % \u7684\u64f7\u53d6\u6e96\u78ba\u7387\u3002 \u65b7\u8a5e\u7684\u6548\u80fd\uff0c\u5176 F-measure \u7684\u7d50\u679c\u5206\u5225\u70ba 80.4 % \u4ee5\u53ca 81.4 %\u3002\u56e0\u6b64\uff0c\u6211\u5011\u767c\u73fe \u81f3\u5de6\u4f9d\u5e8f\u6bd4\u5c0d\u8fad\u5178\u4e2d\u7684\u8a5e\uff0c\u6bd4\u5c0d\u5230\u6700\u9577\u7684\u8a5e\u7576\u6210\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u7684\u65b7\u8a5e\u7d50\u679c\uff0c\u4e26</td></tr></table>", | |
| "type_str": "table", | |
| "text": "\u4ee3\u8868\u4e2d\u6587\u5b57\u5143\u6240\u7d44\u6210\u7684\u5b57\u4e32\uff0c\u82e5 x\u3001z\u3001xy \u8207 yz \u7686\u70ba\u8fad\u5178\u4e2d\u7684\u8a5e\uff0c \u5247 xyz \u7684\u7d44\u5408\uff0c\u65bc\u4e0d\u540c\u7684\u6587\u7ae0\u4e2d\uff0c\u53ef\u80fd\u6703\u88ab\u65b7\u8a5e\u6210 xy/z \u6216 x/yz \u7b49\u5169\u7a2e\u4e0d\u540c \u7684\u7d50\u679c\uff0c\u5247 xyz \u7a31\u70ba\u300c\u4ea4\u96c6\u578b\u6b67\u7fa9\u5b57\u4e32\u300d \u3002\u4f8b\u5982\uff1a \u300c\u4e0d\u53ef\u4ee5\u300d\u4e09\u500b\u4e2d\u6587\u5b57\u5143 \u4ee3\u8868\u4e2d\u6587\u5b57\u5143\u6240\u7d44\u6210\u7684\u5b57\u4e32\uff0c\u82e5 x\u3001y\u3001xy \u90fd\u662f\u8fad\u5178\u4e2d\u7684\u8a5e\uff0cxy \u7684\u7d44 \u5408\u4e2d\uff0c\u53ef\u5728\u4e0d\u540c\u7684\u6587\u7ae0\u4e2d\uff0c\u5206\u5225\u88ab\u65b7\u8a5e\u6210 xy \u6216 x/y\uff0c\u56e0\u70ba\u8a5e xy \u662f\u7531 x \u8207 y", | |
| "html": null, | |
| "num": null | |
| }, | |
| "TABREF2": { | |
| "content": "<table><tr><td>\u4e0d\u8ad6\u662f\u4f7f\u7528 SWF \u6216\u662f SEF \u6e96\u5247\u4f86\u9078\u53d6\u7279\u88fd\u8a5e\uff0c\u90fd\u9700\u8981\u6c7a\u5b9a\u4e00\u500b\u9580\u6abb\u503c \u4f7f\u7528\u5b57\u5143\u8cc7\u8a0a\u7576\u6210\u89c0\u6e2c\u7b26\u865f\u7684\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(HMM) \u3002\u9664\u4e86 M-HMM \u7684\u5be6 \u5728\u6b64\u5be6\u9a57\u7684\u7b2c\u4e00\u500b\u90e8\u5206\uff0c\u5206\u5272\u6bd4\u4f8b\u70ba 100/0\uff0c\u76f8\u7576\u65bc\u5be6\u9a57\u4e00\u6b67\u7fa9\u6027\u7684\u6548\u80fd\uff0c \u8f03\u5c0f\u7684\u8fad\u5178 Dict-dict(i)\uff0c\u4f86\u6a19\u793a M-HMM \u6240\u9700\u8981\u7684\u89c0\u6e2c\u7b26\u865f\u3002\u5728\u9019\u904e\u7a0b\u4e2d\uff0c\u6709 \u7684\u8cc7\u6599\uff0c\u53d6\u51fa\u9ad8\u983b\u7387\u7684\u8a5e\u505a\u7279\u88fd\u8a5e\u3002\u800c SEF \u70ba\u53d6\u9ad8\u6e2c\u8a66\u932f\u8aa4\u7387\u7684\u8a5e\u7576\u6210\u7279\u88fd\u8a5e\uff0c</td></tr><tr><td>(threshold) \uff0c\u6b64\u9580\u6abb\u503c\u662f\u6c7a\u5b9a\u7279\u88fd\u8a5e\u7684\u5927\u5c0f\uff0c\u6211\u5011\u6703\u65bc\u5be6\u9a57\u56db\u4e2d\u627e\u51fa\u6700\u4f73\u65b7\u8a5e \u9a57\u7d50\u679c\u4e4b\u5916\uff0c\u540c\u6642\u6211\u5011\u4e5f\u6bd4\u8f03\u53ea\u7d50\u5408\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u8cc7\u8a0a(FMM+HMM)\u4ee5\u53ca \u800c\u672a\u77e5\u8a5e\u5be6\u9a57\u7684\u90e8\u5206\uff0c\u5171\u5be6\u9a57 80/0\u300160/20\u300140/40\u300120/60 \u7b49\u5206\u5272\u6bd4\u4f8b\u7684\u7d50\u679c\uff0c \u4e9b\u5b57\u8a5e\u6703\u56e0\u70ba\u672a\u77e5\u8a5e\u7684\u95dc\u4fc2\uff0c\u6703\u88ab\u932f\u6a19\u6210\u55ae\u4e00\u5b57\u8a5e S\uff0c\u4f46\u5176\u72c0\u614b\u7b26\u865f\uff0c\u53ef\u4ee5\u8b93 \u56e0\u6b64\u6211\u5011\u5148\u5f9e 70% \u7684\u8cc7\u6599\u5efa\u7acb M-HMM \u6a21\u578b\uff0c\u4e26\u4e14\u65bc\u8abf\u6574\u8cc7\u6599\u4e2d\u505a\u6e2c\u8a66\uff0c\u6839\u64da</td></tr><tr><td>\u6548\u80fd\u7684\u9580\u6abb\u503c\u3002 \u53ea\u7d50\u5408\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u8cc7\u8a0a(BMM+HMM)\u7684\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u4e4b\u65b7\u8a5e\u6548\u80fd\u3002 \u7531\u672a\u77e5\u8a5e\u6240\u4f54\u7684\u6bd4\u4f8b\u4e4b\u4e0d\u540c\u4f86\u9a57\u8b49\u65b7\u8a5e\u6548\u80fd\uff0c\u800c\u6700\u5f8c\u4e00\u500b\u90e8\u5206\uff0c\u5206\u5272\u6bd4\u4f8b\u70ba HMM \u77e5\u9053\u6b63\u78ba\u7684\u6a19\u7c64\uff1b\u5982\u679c\u6a19\u793a\u7d50\u679c\u8207\u539f\u4f86\u76f8\u540c\u6642\uff0c\u5247\u53ef\u76f4\u63a5\u7701\u7565\uff0c\u4ee5\u907f\u514d\u5728 \u8abf\u6574\u8cc7\u6599\u4e2d\u9ad8\u6e2c\u8a66\u932f\u8aa4\u7387\u7684\u8a5e\u505a\u7279\u88fd\u8a5e\u3002\u53d6\u5f97 SWF \u8207 SEF \u4e4b\u7279\u88fd\u8a5e\u5f8c\uff0c\u63a5\u8457\u9a57</td></tr><tr><td>E \u6b64\u7279\u88fd\u5316\u904e\u7a0b\u4e5f\u5c07\u727d\u626f\u5230\u4e00\u500b\u554f\u984c\uff1a\u7531\u65bc\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u4e09\u500b\u4e3b\u8981\u53c3\u6578 \u90fd\u8207\u300c\u72c0\u614b\u7b26\u865f\u300d\u6709\u95dc\uff0c\u56e0\u6b64\u9019\u968e\u6bb5\u7684\u7279\u88fd\u5316\u904e\u7a0b\uff0c\u5c07\u589e\u52a0\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684 \u53c3\u6578\u5927\u5c0f\uff0c\u56e0\u6b64\u8a08\u7b97\u91cf\u4e5f\u5c31\u6703\u8ddf\u8457\u589e\u52a0\uff0c\u800c\u4e14\u904e\u591a\u7684\u7279\u88fd\u8a5e\u4e0d\u898b\u5f97\u80fd\u4e00\u76f4\u63d0\u5347\u6e96 \u78ba\u7387\u3002\u6240\u4ee5\u6211\u5011\u5fc5\u9808\u6839\u64da\u8a13\u7df4\u8cc7\u6599\u4f86\u6c7a\u5b9a\u7279\u88fd\u8a5e\u7684\u5927\u5c0f\u3002 4. \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57 \u65bc\u7cfb\u7d71\u5be6\u9a57\u4e2d\uff0c\u6211\u5011\u4f7f\u7528\u4e2d\u7814\u9662\u5e73\u8861\u8a9e\u6599\u5eab\u7b2c 3.1 \u7248\uff0c\u7576\u6210\u6211\u5011\u5be6\u9a57\u7684\u8cc7\u6599\u3002 \u6b64\u8a9e\u6599\u5eab\uff0c\u5171\u6709 575 \u842c\u8a5e\uff0c\u662f\u7b2c\u4e00\u500b\u5df2\u65b7\u597d\u7684\u8a5e\u4e26\u5e36\u6709\u8a5e\u985e\u6a19\u8a18\u7684\u73fe\u4ee3\u6f22\u8a9e\u8a9e\u6599 \u5eab\u3002\u6211\u5011\u5c07\u5176\u4e2d\u5df2\u65b7\u8a5e\u7684\u4e2d\u6587\u6587\u7ae0\u4f86\u7576\u6210\u6211\u5011\u7684\u5be6\u9a57\u5c0d\u8c61\uff0c\u4e26\u7528\u96a8\u6a5f\u7684\u65b9\u5f0f\u5206\u6210 \u5169\u500b\u90e8\u5206\uff0c\u53d6\u5176\u4e2d\u7684 80% \u7576\u4f5c\u8a13\u7df4\u8a9e\u6599\uff0c\u7528\u4f86\u8a13\u7df4\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u3002\u800c\u5269\u4e0b \u7684 20% \u5247\u7576\u6210\u6211\u5011\u7cfb\u7d71\u7684\u6e2c\u8a66\u8a9e\u6599\u3002\u65b7\u8a5e\u7684\u8a55\u4f30\u65b9\u5f0f\u5247\u662f\u4f7f\u7528\u6e96\u78ba\u7387 (Precision) \u3001\u53ec\u56de\u7387(Recall)\u4ee5\u53ca F-measure \u4f86\u9a57\u8b49\u65b7\u8a5e\u6548\u80fd\uff0c\u5206\u5225\u5b9a\u7fa9\u5982\u4e0b\uff1a \u7cfb\u7d71\u65b7\u8a5e\u7684\u7e3d\u8a5e\u6578 \u7cfb\u7d71\u6b63\u78ba\u65b7\u51fa\u7684\u8a5e\u6578 = Precision \u771f\u6b63\u7684\u8a5e\u6578 \u7cfb\u7d71\u6b63\u78ba\u65b7\u51fa\u7684\u8a5e\u6578 = Recall Recall Precision Recall Precision 2 measure F + = * * \u7531\u65bc\u6211\u5011\u7684\u7cfb\u7d71\u5206\u6210 M-HMM \u8207 Lexicalized HMM \u5169\u90e8\u5206\uff0c\u56e0\u6b64\u5728\u5be6\u9a57\u7684 \u90e8\u5206\uff0c\u6211\u5011\u4e5f\u7531\u6b64\u5169\u90e8\u5206\u4f86\u505a\u5be6\u9a57\u3002M-HMM \u5be6\u9a57\u7684\u90e8\u4efd\u70ba\u5be6\u9a57\u4e00\u3001\u4e8c\u3001\u4e09\uff1b\u800c Lexicalized HMM \u5be6\u9a57\u7684\u90e8\u5206\u5247\u70ba\u5be6\u9a57\u4e09\u8207\u5be6\u9a57\u56db\u3002 4.1. M-HMM \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57(\u5be6\u9a57\u4e00\u3001\u5be6\u9a57\u4e8c) (\u5be6\u9a57\u4e00\u3001\u5be6\u9a57\u4e8c) (\u5be6\u9a57\u4e00\u3001\u5be6\u9a57\u4e8c) (\u5be6\u9a57\u4e00\u3001\u5be6\u9a57\u4e8c) M-HMM \u5be6\u9a57\u7684\u90e8\u5206\uff0c\u4e3b\u8981\u9a57\u8b49\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7d50\u5408\u9577\u8a5e\u512a\u5148\u6cd5\u4e4b\u5f8c\uff0c\u5728 \u8868 5 \u5be6\u9a57\u4e00\uff1aM-HMM \u89e3\u6b67\u7fa9\u6027\u7684\u65b7\u8a5e\u6548\u80fd FMM BMM HMM FMM+HMM BMM+HMM FMM+BMM+HMM (M-HMM) Recall 0.936 0.939 0.812 0.944 0.947 0.957 Precision 0.956 0.959 0.811 0.962 0.965 0.976 F-measure 0.946 0.949 0.812 0.953 0.956 0.967 \u5be6\u9a57\u986f\u793a\uff0c\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u53ea\u4f7f\u7528\u5b57\u5143\u7684\u8cc7\u8a0a\u6642\uff0c\u5176\u65b7\u8a5e\u7d50\u679c\u53ea\u6709 0.81 \u5de6\u53f3\uff0c\u800c\u52a0\u5165\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u8207\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u4e4b\u5f8c\uff0c\u7cfb\u7d71\u7684\u65b7\u8a5e\u6548\u80fd F-measure \u7531 0.812 \u5927\u5e45\u5730\u63d0\u5347\u5230 0.967\uff0c\u4e26\u4e14\u65b7\u8a5e\u7d50\u679c\u4e5f\u52dd\u904e\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u8207\u53cd\u5411\u9577\u8a5e \u512a\u5148\u6cd5\u7b49\u5169\u7a2e\u57fa\u7dda\u4f5c\u6cd5\u3002\u56e0\u6b64\uff0c\u5be6\u9a57\u7d50\u679c\u8b49\u660e\u4e86\u9577\u8a5e\u512a\u5148\u6cd5\u6240\u63d0\u4f9b\u4e4b\u6b67\u7fa9\u6027\u8cc7\u8a0a \u7684\u78ba\u53ef\u63d0\u5347\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u6548\u80fd\u3002 \u5be6\u9a57\u4e8c\u4e3b\u8981\u662f\u9a57\u8b49\u9577\u8a5e\u512a\u5148\u6cd5\u6240\u4f7f\u7528\u7684\u8fad\u5178\uff0c\u5c0d M-HMM \u65b7\u8a5e\u7cfb\u7d71\u7684\u5f71\u97ff\uff0c \u4e5f\u5c31\u662f\u5be6\u9a57\u672a\u77e5\u8a5e\u7684\u65b7\u8a5e\u6548\u80fd\u3002\u7531\u65bc\u8fad\u5178\u662f\u7531\u8a13\u7df4\u8cc7\u6599\u7522\u751f\uff0c\u56e0\u6b64\u5be6\u9a57\u6642\u6211\u5011\u5c07 \u8a13\u7df4\u8cc7\u6599\u96a8\u6a5f\u5206\u5272\u6210\u5169\u90e8\u5206\uff1a\u8a13\u7df4\u96c6\u5408 1(set 1)\u4ee5\u53ca\u8a13\u7df4\u96c6\u5408 2(set 2) \uff0c\u8fad\u5178 \u53ea\u7531\u8a13\u7df4\u96c6\u5408 1 \u4f86\u7522\u751f\uff0c\u85c9\u7531\u8abf\u6574\u8a13\u7df4\u8cc7\u6599\u4e0d\u540c\u7684\u5206\u5272\u6bd4\u4f8b\uff0c\u4ee5\u7522\u751f\u51fa\u4e0d\u540c\u7684\u8fad \u5178\u6578\u91cf\uff0c\u5728\u76f8\u540c\u7684\u6e2c\u8a66\u8cc7\u6599\u4e0b\u4ee5\u9a57\u8b49\u5404\u81ea\u7684\u65b7\u8a5e\u6548\u80fd\u3002\u5be6\u9a57\u7d50\u679c\u5982\u8868 6 \u6240\u793a\u3002 \u5be6\u9a57\u56db\u662f\u6839\u64da Lexicalized M-HMM \u7684 SWF \u8207 SEF \u5169\u7a2e\u4e0d\u540c\u7684\u8a5e\u5f59\u5316\u7b56\u7565 \u8868 7 \u5be6\u9a57\u4e8c\uff1aM-HMM \u89e3\u672a\u77e5\u8a5e\u7684\u65b7\u8a5e\u6548\u80fd 0/80\uff0c\u4ee3\u8868\u5b8c\u5168\u4e0d\u5f9e\u8a13\u7df4\u8cc7\u6599\u4e2d\u5efa\u7acb\u8fad\u5178\uff0c\u4e5f\u5c31\u662f\u6e2c\u8a66\u8cc7\u6599\u4e2d\u6240\u6709\u7684\u8a5e\u90fd\u5c6c\u65bc\u672a \u77e5\u8a5e\uff0c\u4e26\u4e14\u5728\u8a13\u7df4\u7684\u904e\u7a0b\u4e2d\u5b8c\u5168\u6c92\u6709\u5f9e\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u6216\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5\u4e2d\u5f97\u5230 \u4efb\u4f55\u8cc7\u8a0a\uff0c\u53ea\u4f9d\u8cf4\u5b57\u5143\u7684\u8cc7\u8a0a\u505a\u65b7\u8a5e\u3002 \u5be6\u9a57\u4e8c\u7684\u7d50\u679c\u53ef\u5f97\u77e5\uff0c\u96a8\u8457\u589e\u52a0\u672a\u77e5\u8a5e\u7684\u8cc7\u8a0a\uff0c\u4e5f\u5c31\u662f\u5728\u6e1b\u5c11\u5b57\u5178\u7684\u8a5e\u6578\u7684 \u60c5\u6cc1\u4e0b\uff0cM-HMM \u7684\u65b7\u8a5e\u6548\u80fd\u8ddf\u8457\u6e1b\u4f4e\uff0c\u4f46\u662f\u964d\u4f4e\u7684\u5e45\u5ea6\u4e26\u4e0d\u5927\uff0c\u986f\u898b\u53ea\u8981\u6709\u57fa \u672c\u8a5e\u5f59\uff0c\u5373\u53ef\u63d0\u5347 HMM \u65b7\u8a5e\u6548\u80fd\uff0c\u4f46\u5c0d\u65bc\u672a\u77e5\u8a5e\u554f\u984c\uff0c\u4e26\u4e0d\u80fd\u6709\u6240\u505a\u70ba\uff0c\u56e0\u6b64 \u6211\u5011\u5c07\u65bc\u5be6\u9a57\u4e09\u8a2d\u8a08 Mask \u7684\u5be6\u9a57\u4f86\u89e3\u6c7a\u6b64\u4e00\u554f\u984c\u3002 4.2. Mask \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57(\u5be6\u9a57\u4e09) (\u5be6\u9a57\u4e09) (\u5be6\u9a57\u4e09) (\u5be6\u9a57\u4e09) \u7531\u65bc\u5be6\u9a57\u4e8c\u662f\u900f\u904e\u6e1b\u5c11\u8a13\u7df4\u8cc7\u6599\u4e2d\u7684\u8a5e\uff0c\u4f86\u5efa\u7acb\u9577\u8a5e\u512a\u5148\u6cd5\u6240\u9700\u4e4b\u8fad\u5178\u7684\u65b9 \u6cd5\u4ee5\u63d0\u4f9b\u672a\u77e5\u8a5e\u8cc7\u8a0a\uff0c\u4f46\u662f\u72a0\u7272\u4e86\u9577\u8a5e\u512a\u5148\u6cd5\u7684\u6b63\u78ba\u6027\u3002\u56e0\u6b64\u6211\u5011\u5f15\u7528 Mask \u7684 \u4f5c\u6cd5 [21]\uff0c\u5728\u4e0d\u72a7\u7272\u8a13\u7df4\u8cc7\u6599\u7684\u8a5e\u7684\u524d\u63d0\u4e0b\uff0c\u7522\u751f\u5177\u6709\u672a\u77e5\u8a5e\u8cc7\u8a0a\u7684\u8a13\u7df4\u8cc7\u6599\u3002 Mask \u7684\u6982\u5ff5\u662f\u8b93\u8a13\u7df4\u904e\u7a0b\u4e2d\u4e5f\u6709\u6a5f\u6703\u78b0\u5230\u672a\u77e5\u8a5e\uff0c\u4e5f\u5c31\u662f\u4eff\u9020\u6e2c\u8a66\u6642\u771f\u6b63\u7684\u60c5 \u5f62\uff0c\u5176\u4f5c\u6cd5\u5982\u4e0b\uff1a \u4e00\u500b\u72c0\u614b\u6240\u898b\u5230\u7684\u89c0\u6e2c\u7b26\u865f\u6a5f\u7387\u4e0d\u516c\u5e73\u7684\u589e\u52a0\uff0c\u5982\u6b64\u91cd\u8907 K \u6b21\u6700\u5f8c\u5c07\u6b64 K+1 \u500b \u8cc7\u6599\u5f62\u6210\u6574\u500b Mask \u7684\u8a13\u7df4\u8cc7\u6599\u3002 \u5be6\u9a57\u4e09\u70ba\u4f7f\u7528 Mask \u65b9\u6cd5\u6240\u505a\u7684 M-HMM \u7684\u5be6\u9a57\uff0c\u53d6 Mask K=2 \u81f3 K=10 \u4f86 \u9a57\u8b49\u7d50\u679c\uff0c\u800c K=1 \u8868\u793a\u4e0d\u505a\u5206\u5272\uff0c\u4e5f\u5c31\u662f\u6c92\u6709\u4f7f\u7528 Mask \u7684\u7d50\u679c\uff0c\u5be6\u9a57\u5982\u5716 2 \u6240 \u793a\u3002\u5be6\u9a57\u4e09\u7d50\u679c\u986f\u793a\uff0c\u4f7f\u7528 Mask \u7684\u65b9\u6cd5\u53ef\u63d0\u4f9b\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u66f4\u591a\u672a\u77e5\u8a5e\u8cc7 \u8a0a\uff0c\u4f7f\u5f97\u65b7\u8a5e\u6548\u80fd\u6709\u6240\u63d0\u5347\uff0c\u4e26\u4e14\u5728 K=2 \u6642\uff0c\u9054\u5230\u6700\u4f73\u7684\u65b7\u8a5e\u6548\u80fd(F-measure = 95.25%) \u3002 94.80 94.90 95.00 95.10 95.20 95.30 1 2 3 4 5 6 7 8 9 10 K F-measure (%) \u5716 2 \u5be6\u9a57\u4e09\uff1aMask K=1 \u81f3 K=10 \u7684\u5be6\u9a57\u7d50\u679c 4.3. Lexicalized M-HMM \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57 \u5be6\u9a57(\u5be6\u9a57\u56db\u3001\u4e94) (\u5be6\u9a57\u56db\u3001\u4e94) (\u5be6\u9a57\u56db\u3001\u4e94) \u8b49\u5728\u4e0d\u540c\u7684\u9580\u6abb\u503c\u4e0b\uff0c\u8abf\u6574\u8cc7\u6599\u7684\u65b7\u8a5e\u6548\u80fd\u3002\u5be6\u9a57\u6578\u64da\u5982\u5716 3 \u6240\u793a\u3002 \u5be6\u9a57\u7d50\u679c\u986f\u793a\uff0c\u6211\u5011\u4f7f\u7528 SWF \u8207 SEF \u5169\u7a2e\u4e0d\u540c\u7684\u8a5e\u5f59\u5316\u7b56\u7565\uff0c\u5728\u525b\u958b\u59cb\u53d6 \u8f03\u5c11\u7684\u8a5e\u7576\u7279\u88fd\u8a5e\u6642\uff0c\u5169\u8005\u5728\u8abf\u6574\u8cc7\u6599\u4e0b\u7684\u65b7\u8a5e\u6548\u80fd\u90fd\u6709\u986f\u8457\u7684\u4e0a\u5347\uff0c\u800c SWF \u5728\u53d6 292 \u500b\u8a5e(\u51fa\u73fe\u983b\u7387\u5927\u65bc 4800 \u6b21)\u6642\uff0cSEF \u53d6 173 \u500b\u8a5e(\u51fa\u73fe\u983b\u7387\u5927\u65bc 25 \u6b21)\u6642\uff0c\u65b7\u8a5e\u6548\u80fd\u9054\u5230\u6700\u4f73\u7d50\u679c\uff0c\u4e26\u4e14\u518d\u7e7c\u7e8c\u96a8\u8457\u7279\u88fd\u8a5e\u6578\u7684\u589e\u52a0\uff0c\u65b7\u8a5e\u7d50\u679c\u4fbf \u958b\u59cb\u5f80\u4e0b\u964d\uff0c\u9019\u662f\u56e0\u70ba\u72c0\u614b\u6578\u589e\u52a0\uff0c\u4f7f\u5f97\u6a21\u578b\u8a08\u7b97\u91cf\u589e\u52a0\u800c\u5c0e\u81f4\u6e96\u78ba\u7387\u4e0b\u964d\u4e4b\u7de3 \u6545\u3002 95 95.2 95.4 95.6 95.8 96 96.2 96.4 0 50 100 150 200 250 300 # Specialized Words F-measure (%) SEF SWF BMM HMM M-HMM SWF M-HMM SEF M-HMM Recall 0.925 0.928 0.812 0.947 0.958 0.963 Precision 0.928 0.930 0.811 0.958 0.962 0.964 F-measure 0.926 0.929 0.812 0.953 0.960 0.963 5. \u7d50\u8ad6 \u7d50\u8ad6 \u7d50\u8ad6 \u7d50\u8ad6 \u5728\u672c\u7bc7\u8ad6\u6587\u4e2d\uff0c\u6211\u5011\u61c9\u7528\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u4e4b\u7279\u88fd\u5316\u7684\u6982\u5ff5\u4f86\u63d0\u5347\u4e2d\u6587\u65b7\u8a5e \u7684\u6548\u80fd\uff0c\u6211\u5011\u7cfb\u7d71\u7684\u6700\u5927\u7684\u512a\u9ede\uff0c\u5c31\u662f\u5b8c\u5168\u4e0d\u9700\u8981\u5c0d\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u8a13\u7df4\u904e \u7a0b\u4ee5\u53ca\u6e2c\u8a66\u904e\u7a0b\u505a\u4efb\u4f55\u4fee\u6539\uff0c\u53ea\u9700\u5c07\u8a13\u7df4\u8cc7\u6599\u6839\u64da\u7279\u88fd\u5316\u51fd\u5f0f\u4f86\u505a\u8f49\u63db\u5373\u53ef\u3002\u6211 \u5011\u4f7f\u7528\u5169\u968e\u6bb5\u7684\u7279\u88fd\u5316\u904e\u7a0b\u9010\u6b65\u7684\u6539\u826f\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u7684\u65b7\u8a5e\u6548\u80fd\uff0c\u5728\u7b2c\u4e00\u968e \u6bb5\u4e2d\u7d50\u5408\u4e86\u9577\u8a5e\u512a\u5148\u6cd5\u7684\u8cc7\u8a0a\uff0c\u4f7f\u5f97\u89c0\u6e2c\u7b26\u865f\u589e\u52a0\u66f4\u591a\u7684\u8cc7\u8a0a\uff0c\u65bc\u5be6\u9a57\u7d50\u679c\u986f \u793a\uff0c\u7d50\u5408\u9577\u8a5e\u512a\u5148\u6cd5\u5728\u6c92\u6709\u672a\u77e5\u8a5e\u7684\u60c5\u6cc1\u4e0b\uff0c\u53ef\u4ee5\u5927\u5e45\u5730\u63d0\u5347\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b \u7684\u65b7\u8a5e\u6548\u80fd(F-measure: 0.812\u21920.967) \uff0c\u800c\u5728\u6709\u672a\u77e5\u8a5e\u7684\u60c5\u6cc1\u4e0b\uff0c\u5229\u7528 Mask \u65b9 \u5f0f\u4e5f\u4e9b\u5fae\u6539\u5584\u65b7\u8a5e\u6548\u80fd(F-measure: 0.948\u21920.953) \u3002\u800c\u7b2c\u4e8c\u968e\u6bb5\u4f7f\u7528\u8a5e\u5f59\u5f0f\u7684\u7279 \u88fd\u5316\u65b9\u5f0f\uff0c\u6311\u9078\u9ad8\u932f\u8aa4\u7684\u5b57\u5143\u4f7f\u5f97\u72c0\u614b\u589e\u52a0\uff0c\u5be6\u9a57\u4e5f\u8b49\u660e\u80fd\u518d\u6b21\u63d0\u5347\u65b7\u8a5e\u6548\u80fd (F-measure: 0.953\u21920.963) \uff0c\u5be6\u9a57\u4e2d\u767c\u73fe\u4f7f\u7528 SEF \u6e96\u5247\u7684\u7d50\u679c\u6703\u6bd4 SWF \u6e96\u5247 \u5716 3 FMM (\u5be6\u9a57\u56db\u3001\u4e94) \u4e0d\u4f46\u4f7f\u7528\u7684\u7279\u88fd\u8a5e\u8f03\u5c0f\u4e14\u53c8\u80fd\u9054\u5230\u66f4\u597d\u7684\u65b7\u8a5e\u7d50\u679c\u3002</td></tr><tr><td>\u7279\u88fd\u8a5e\u7684\u9078\u64c7\u65b9\u5f0f\uff0c\u6211\u5011\u662f\u4f7f\u7528\u5169\u7a2e\u4e0d\u540c\u7684\u6e96\u5247(criteria)\u4f86\u9078\u53d6\uff0c\u8aaa\u660e\u5982 \u89c0\u6e2c\u7b26\u865f\u4e2d\u52a0\u5165\u66f4\u591a\u8cc7\u8a0a\u4e4b\u524d\u8207\u52a0\u5165\u4e4b\u5f8c\u7684\u65b7\u8a5e\u6548\u80fd\u7684\u6bd4\u8f03\u3002\u7531\u65bc\u9577\u8a5e\u512a\u5148\u6cd5\u53ef \u4e0d\u542b\u672a\u77e5\u8a5e \u672a \u672a \u672a \u672a\u77e5\u8a5e(\u5be6\u9a57\u4e8c) \u77e5\u8a5e(\u5be6\u9a57\u4e8c) \u77e5\u8a5e(\u5be6\u9a57\u4e8c) \u77e5\u8a5e(\u5be6\u9a57\u4e8c) HMM \u8a13\u7df4\u8cc7\u6599\u6bd4\u4f8b (Set1/Set2) 100/0 80/0 60/20 40/40 20/60 \u8a8c\u8b1d \u8a8c\u8b1d \u8a8c\u8b1d \u8a8c\u8b1d \u4e0b\uff0c\u7528\u4f86\u8abf\u6574\u5404\u81ea\u4f7f\u7528\u7684\u7279\u88fd\u8a5e\u5927\u5c0f\uff0c\u4ee5\u627e\u51fa\u4f7f\u5f97\u6a21\u578b\u80fd\u6709\u6700\u4f73\u65b7\u8a5e\u6548\u80fd\u7684\u9580\u6abb 0/80 \u4e0b\uff1a \u4ee5\u63d0\u4f9b\u65b7\u8a5e\u6b67\u7fa9\u6027\u8207\u672a\u77e5\u8a5e\u7b49\u8cc7\u8a0a\uff0c\u56e0\u6b64\u9019\u90e8\u5206\u7684\u5be6\u9a57\uff0c\u6211\u5011\u662f\u5148\u9a57\u8b49\u6b67\u7fa9\u6027\u7684 \u8fad\u5178\u4e2d\u7684\u8a5e\u6578 145,608 132,273 116,428 96,780 69,446 0 \u503c(threshold) \u3002\u7531\u65bc\u9019\u500b\u5be6\u9a57\u662f\u7528\u4f86\u8abf\u6574\u7cfb\u7d71\u7528\u5230\u7684\u7279\u88fd\u8a5e\uff0c\u800c\u4e0d\u662f\u505a\u65b7\u8a5e\u6548 \u672c\u7814\u7a76\u7531\u570b\u79d1\u6703\u7de8\u865f NSC 94-2213-E-008-020 \u8d0a\u52a9\u3002</td></tr><tr><td>SWF: (the Words with High Frequency) \u5be6\u9a57\u4e00\u9a57\u8b49\u65b7\u8a5e\u6b67\u7fa9\u6027\u6548\u80fd\uff0c\u65b9\u6cd5\u662f\u53d6\u6240\u6709\u8a13\u7df4\u8cc7\u6599\u8207\u6e2c\u8a66\u8cc7\u6599\u4e2d\u7684\u6240\u6709 Recall 0.957 0.946 0.946 0.944 0.941 0.812 \u9996\u5148\u5c07\u8a13\u7df4\u8cc7\u6599\u5206\u5272\u6210 K \u500b\u90e8\u5206\uff0c\u4e26\u4e14\u6bcf\u500b\u90e8\u5206\u90fd\u5efa\u7acb\u5404\u81ea\u7684\u8fad\u5178\uff0c\u56e0\u6b64\u53ef \u90e8\u8cc7\u6599 80%) \u5206\u5272\u6210\u5169\u90e8\u5206\uff0c\u4f9d 7 \u6bd4 1 \u7684\u6bd4\u4f8b\u4f86\u5206\u5272 (\u5206\u5225\u4f54\u5168\u90e8\u8cc7\u6599\u7684 70% \u8207 \u53d6\u5728\u8a13\u7df4\u8cc7\u6599\u4e2d\u5c6c\u65bc\u6700\u9ad8\u983b\u7387\u7684\u89c0\u6e2c\u7b26\u865f\uff0c\u7576\u6210\u7279\u88fd\u8a5e\u3002 \u65b7\u8a5e\u6548\u80fd\uff0c\u518d\u9a57\u8b49\u672a\u77e5\u8a5e\u8cc7\u8a0a\u591a\u5be1\u4e4b\u65b7\u8a5e\u6548\u80fd\u7684\u6bd4\u8f03\u3002 Set2 \u4e2d\u7684\u672a\u77e5\u8a5e\u6578 0 0 17,418 45,212 103,990 All \u6e2c\u8a66\u8cc7\u6599\u4e2d\u7684\u672a\u77e5\u8a5e\u6578 0 14,415 17,323 22,524 34,573 All \u5716 1 Mask (K=3) \u8cc7\u6599\u5206\u5272\u8207\u5efa\u7acb\u8fad\u5178 \u80fd\u7684\u5be6\u9a57\uff0c\u56e0\u6b64\u6211\u5011\u53ea\u53d6\u300c\u8a13\u7df4\u8cc7\u6599\u300d\u4f86\u505a\u6b64\u5be6\u9a57\u3002\u6211\u5011\u5c07\u5168\u90e8\u8a13\u7df4\u8cc7\u6599(\u4f54\u5168 \u53c3\u8003\u6587\u737b \u53c3\u8003\u6587\u737b \u53c3\u8003\u6587\u737b \u53c3\u8003\u6587\u737b</td></tr><tr><td>SEF: (the Words with Tagging Error Frequency) \u8a5e\uff0c\u4f86\u7576\u6210\u9577\u8a5e\u512a\u5148\u6cd5\u6240\u4f7f\u7528\u7684\u8fad\u5178\u7684\u8a5e(\u5171\u6709 145,608 \u500b\u8a5e) \uff0c\u4f7f\u5f97\u5728\u6e2c\u8a66\u904e Precision 0.976 0.951 0.949 0.945 0.934 0.811 \u7522\u751f K+1 \u500b\u8fad\u5178\uff0c\u5982\u5716 1 \u6240\u793a\uff0c\u6b64 K+1 \u500b\u8cc7\u6599\u4fbf\u53ef\u5efa\u7acb K+1 \u500b\u8a13\u7df4\u8cc7\u6599\u3002\u6211 10%) \uff0c\u5176\u4e2d 70% \u7684\u8cc7\u6599 (\u8f49\u63db\u6210\u5177\u6709\u9577\u8a5e\u512a\u5148\u6cd5\u8cc7\u8a0a\u7684\u8cc7\u6599) \u7528\u4f86\u8a13\u7df4 M-HMM</td></tr><tr><td>F-measure \u53d6\u5177\u6709\u9ad8\u6e2c\u8a66\u932f\u8aa4\u7387 (\u6216\u7a31\u6a19\u793a\u932f\u8aa4\u7387) \u7684\u8a5e\uff0c\u7576\u6210\u7279\u88fd\u8a5e\u3002 0.967 0.948 0.948 0.945 0.937 \u7a0b\u4e2d\u4e0d\u6703\u51fa\u73fe\u672a\u77e5\u8a5e\u3002\u8868 5 \u70ba M-HMM \u89e3\u6b67\u7fa9\u6027\u7684\u65b7\u8a5e\u6548\u80fd\u3002\u5176\u4e2d\u5be6\u9a57\u7684\u57fa\u7dda 0.812 \u5011\u5148\u4ee5\u6240\u6709\u8fad\u5178\u7684\u806f\u96c6(Dict=dict1\uff0bdict2+dict3)\u4f86\u6a19\u793a M-HMM \u6240\u9700\u8981\u7684\u89c0\u6e2c\u7b26 \u6a21\u578b\uff0c\u800c\u5269\u4e0b\u7684 10% \u5247\u7576\u6210\u9a57\u8b49\u6548\u80fd\u7684\u8abf\u6574\u8cc7\u6599(tuning data) \u3002</td></tr><tr><td>(baseline)\u4f5c\u6cd5\u70ba\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5(FMM) \u3001\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5(BMM) \uff0c\u4ee5\u53ca\u53ea \u865f\uff0c\u9019\u4e5f\u76f8\u7576\u662f\u539f\u59cb\u7684\u8a13\u7df4\u8cc7\u6599\u3002\u63a5\u8457\u6bcf\u6b21\u906e\u4f4f\u4e00\u500b\u90e8\u4efd\u8fad\u5178\uff0c\u4e5f\u5c31\u662f\u7522\u751f\u4e00\u500b \u7531\u65bc SWF \u70ba\u53d6\u8a13\u7df4\u8cc7\u6599\u4e2d\u51fa\u73fe\u983b\u7387\u6700\u9ad8\u7684\u8a5e\u7576\u6210\u7279\u88fd\u8a5e\uff0c\u56e0\u6b64\u6211\u5011\u7d71\u8a08 70%</td></tr></table>", | |
| "type_str": "table", | |
| "text": "\u5be6\u9a57\u56db\uff1a\u5728\u4e0d\u540c\u7279\u88fd\u8a5e\u5927\u5c0f\u4e0b\uff0cSEF \u8207 SWF \u6e96\u5247\u5728\u8abf\u6574\u8cc7\u6599\u4e0b\u7684\u65b7\u8a5e\u6548\u80fd \u5be6\u9a57\u4e94\u5247\u662f\u6e2c\u8a66\u6700\u4f73\u7279\u88fd\u8a5e\u7d50\u679c\u7684 SWF \u4ee5\u53ca SEF \u6e96\u5247\u4e4b Lexicalized M-HMM \u65b7\u8a5e\u6548\u80fd\uff0c\u5be6\u9a57\u7684\u8a2d\u5b9a\u4f7f\u7528 Mask K=2 \u4e4b M-HMM \u7684\u8a2d\u5b9a\u4ee5\u53ca\u6700\u4f73 SWF \u8207 SEF \u7279\u88fd\u8a5e (SWF \u70ba\u53d6 292 \u500b\u8a5e\u4f5c\u70ba\u7279\u88fd\u8a5e\uff0c\u800c SEF \u5247\u53d6 173 \u8a5e\u4f5c\u70ba\u7279\u88fd\u8a5e) \u4f86\u505a\u6b64\u5be6\u9a57\uff0c\u4e26\u4e14\u4e5f\u8207\u6b63\u5411\u9577\u8a5e\u512a\u5148\u6cd5(FMM) \u3001\u53cd\u5411\u9577\u8a5e\u512a\u5148\u6cd5(BMM) \u3001\u53ea \u4f7f\u7528\u5b57\u5143\u8cc7\u8a0a\u4e4b\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b(HMM)\u7b49\u57fa\u7dda\u65b7\u8a5e\u4f5c\u6cd5\u53ca M-HMM \u7684\u7d50\u679c \u4f5c\u6bd4\u8f03\uff0c\u4ee5\u9a57\u8b49\u672c\u7cfb\u7d71\u5728\u72c0\u614b\u5ef6\u4f38\u524d\u8207\u5ef6\u4f38\u5f8c\u7684\u65b7\u8a5e\u6548\u80fd\u4f5c\u6bd4\u8f03\u3002\u5be6\u9a57\u7d50\u679c\u5982\u8868 8 \u6240\u793a\u3002\u5be6\u9a57\u7d50\u679c\u986f\u793a Lexicalized M-HMM \u4e0d\u8ad6\u4f7f\u7528 SWF \u6216 SEF \u6e96\u5247\uff0c\u5176\u65b7\u8a5e \u7d50\u679c\u90fd\u6bd4 M-HMM \u7684\u65b7\u8a5e\u6548\u80fd\u8f03\u597d\uff0cF-measure \u7531 0.953 \u63d0\u5347\u5230 0.960 \u8207 0.963\uff0c \u800c\u4e14\u4f7f\u7528 SEF \u6e96\u5247\u8207\u4f7f\u7528 SWF \u6e96\u5247\u76f8\u8f03\u4e4b\u4e0b\uff0cSEF \u4e0d\u4f46\u7279\u88fd\u8a5e\u8f03\u5c11\u4e14\u65b7\u8a5e\u6548\u80fd \u4e5f\u8f03\u597d\u3002 \u8868 8 \u5be6\u9a57\u4e94\uff1aLexicalized M-HMM \u7684\u65b7\u8a5e\u6548\u80fd", | |
| "html": null, | |
| "num": null | |
| } | |
| } | |
| } | |
| } |