| { |
| "paper_id": "O09-1017", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:11:21.926635Z" |
| }, |
| "title": "Dialogue Act Detection Using Sentence Structure and Partial Pattern Trees", |
| "authors": [ |
| { |
| "first": "Wei-Bin", |
| "middle": [], |
| "last": "\u6881\u7dad\u5f6c\u3001\u856d\u80b2\u4e1e\u3001\u5433\u5b97\u61b2", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "National Cheng Kung University", |
| "location": {} |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Yu-Cheng", |
| "middle": [], |
| "last": "Liang", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "National Cheng Kung University", |
| "location": {} |
| }, |
| "email": "liang@csie.ncku.edu.tw" |
| }, |
| { |
| "first": "Chung-Hsien", |
| "middle": [], |
| "last": "Hsiao", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "National Cheng Kung University", |
| "location": {} |
| }, |
| "email": "ychsiao9@gmail.com" |
| }, |
| { |
| "first": "\u570b\u7acb\u6210\u529f\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "National Cheng Kung University", |
| "location": {} |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "This paper presents a dialogue act detection approach using sentence structures and partial pattern trees to generate candidate sentences (CSs). A syntactic parser is utilized to convert the CSs to sentence grammar rules (SRs). To avoid the confusion between dialogue intentions, the K-means algorithm is adopted to cluster the sentence structures of the same dialogue intention based on the SRs. Finally, the relationship between these SRs and the intentions is modeled by a latent dialogue act matrix. Moreover, for the application to a travel information dialogue system, optimal dialogue strategies are trained using the partially observable Markov decision process (POMDP) for robust dialogue management. In evaluation, compared to the semantic slot-based method which achieves 48.1% dialogue act detection accuracy, the proposed approach can achieve 81.9% accuracy, with 33.3% improvement.", |
| "pdf_parse": { |
| "paper_id": "O09-1017", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "This paper presents a dialogue act detection approach using sentence structures and partial pattern trees to generate candidate sentences (CSs). A syntactic parser is utilized to convert the CSs to sentence grammar rules (SRs). To avoid the confusion between dialogue intentions, the K-means algorithm is adopted to cluster the sentence structures of the same dialogue intention based on the SRs. Finally, the relationship between these SRs and the intentions is modeled by a latent dialogue act matrix. Moreover, for the application to a travel information dialogue system, optimal dialogue strategies are trained using the partially observable Markov decision process (POMDP) for robust dialogue management. In evaluation, compared to the semantic slot-based method which achieves 48.1% dialogue act detection accuracy, the proposed approach can achieve 81.9% accuracy, with 33.3% improvement.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "(AT&T)\u7684\u7dda\u4e0a\u670d\u52d9\u7cfb\u7d71 [3] \u3001\u65e5\u672c\u570b\u5bb6\u8cc7\u8a0a\u901a\u4fe1\u79d1\u6280\u7814\u7a76\u6a5f\u69cb(NICT)\u7684\u65c5\u904a\u5c0e\u89bd\u7cfb\u7d71 [4] \uff0c \u4ee5\u53ca Philips \u516c\u53f8\u6240\u958b\u767c\u7684\u706b\u8eca\u6642\u523b\u7968\u50f9\u67e5\u8a62\u7cfb\u7d71 [5] \u3002\u5728\u570b\u5167\u65b9\u9762\uff0c\u53f0\u5927\u6709\u9280\u884c\u96fb\u8a71\u67e5 \u8a62\u7cfb\u7d71 [6] \u3001\u4ea4\u5927\u6709\u6c7d\u8eca\u5c0e\u89bd\u7cfb\u7d71 [7] \u3001\u5de5\u7814\u9662\u5247\u6709\u667a\u6167\u578b\u7e3d\u6a5f\u3001\u6c23\u8c61\u67e5\u8a62\u7cfb\u7d71\u7684\u5be6\u73fe [8] \uff0c \u548c\u6210\u5927\u667a\u6167\u578b\u91ab\u7642\u670d\u52d9\u5c0d\u8a71\u7cfb\u7d71 [9] \u3002\u5728\u610f\u5716\u5075\u6e2c\u90e8\u4efd\u7684\u76f8\u95dc\u7814\u7a76\uff0cChoi \u7b49\u5b78\u8005 [10] \u5c07\u5c0d \u8a71\u8a9e\u6599\u4e2d\u6bcf\u53e5\u8a71\u6a19\u8a18\u5176\u610f\u5716\uff0c\u53ca\u5c0d\u6574\u500b\u5c0d\u8a71\u904e\u7a0b\u6a19\u8a18\u5176\u70ba\u5c0d\u8a71\u904e\u7a0b\u7684\u958b\u59cb\u3001\u7d50\u675f\u3001\u6b63\u5728 \u5c0d\u8a71\u7b49\uff0c\u518d\u4f7f\u7528\u6a5f\u5668\u5b78\u7fd2(machine learning)\u7684\u65b9\u6cd5\u4f86\u5efa\u7acb\u5176\u6a21\u578b\u4ee5\u5224\u65b7\u610f\u5716\uff0c\u4f46\u9019\u4e00\u90e8 \u5206\u7684\u65b9\u6cd5\u5c0d\u65bc\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u9020\u6210\u7cfb\u7d71\u56de\u61c9\u932f\u8aa4\u7684\u554f\u984c\u4e26\u672a\u8003\u616e\u3002\u800c\u5728\u5c0d\u8a71\u7ba1\u7406\u90e8\u4efd\uff0c\u76ee \u524d\u7684\u7814\u7a76\u6709\u61c9\u7528\u6709\u9650\u72c0\u614b\u6a5f(finite state machine) [4] \u8207\u90e8\u4efd\u89c0\u5bdf\u99ac\u53ef\u592b\u6c7a\u5b9a\u7a0b\u5e8f(partial observation Markov decision process, POMDP) [11] ", |
| "cite_spans": [ |
| { |
| "start": 14, |
| "end": 17, |
| "text": "[3]", |
| "ref_id": "BIBREF2" |
| }, |
| { |
| "start": 47, |
| "end": 50, |
| "text": "[4]", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 81, |
| "end": 84, |
| "text": "[5]", |
| "ref_id": "BIBREF5" |
| }, |
| { |
| "start": 105, |
| "end": 108, |
| "text": "[6]", |
| "ref_id": "BIBREF6" |
| }, |
| { |
| "start": 120, |
| "end": 123, |
| "text": "[7]", |
| "ref_id": null |
| }, |
| { |
| "start": 146, |
| "end": 149, |
| "text": "[8]", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 167, |
| "end": 170, |
| "text": "[9]", |
| "ref_id": "BIBREF9" |
| }, |
| { |
| "start": 194, |
| "end": 198, |
| "text": "[10]", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 364, |
| "end": 367, |
| "text": "[4]", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 433, |
| "end": 437, |
| "text": "[11]", |
| "ref_id": "BIBREF11" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\uf0e5 = \u2212 = k i i x 1 2 2 ) ( \u03c3 \u03bc \u03c7 (9) \u85c9\u6b64\u5224\u65b7\u8a9e\u97f3\u8fa8\u8b58\u7d50\u679c\u7684\u6bcf\u500b\u8a5e\u5f59\u8981\u63a5\u53d7\u6216\u62d2\u7d55\uff0c\u62d2\u7d55\u7684\u8a5e\u5f59\u6211\u5011\u66ff\u63db\u6210 Filler\u3002\u66ff\u63db \u6210 Filler \u7684\u539f\u56e0\u662f\u6211\u5011\u60f3\u4fdd\u6301\u53e5\u5b50\u539f\u672c\u7684\u53e5\u578b\uff0c\u4e14\u4e5f\u53ef\u4ee5\u5047\u8a2d\u70ba\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u6642\u53ef\u80fd\u6703 \u767c\u751f\u7684\u60c5\u6cc1\u3002", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "(1) Root \u2192 IP\u3001(2) IP \u2192 VP\u3001 (3) VP \u2192 ADVP VP\u3001(4) ADVP \u2192 AD \u7591\u554f\u8a5e\u3001(5) VP \u2192 VV \u8def\u7dda NP \u548c(6) NP \u2192 NN \u5730\u9ede\u3002\u6211\u5011\u4fbf\u4ee5\u9019\u516d\u689d\u898f\u5247\u4ee3\u8868\u9019\u53e5\u8a71\u3002 \u5716 3\uff1a\u5256\u6790\u5f97\u5230\u7684\u6587\u6cd5\u6a39\u7bc4\u4f8b\u3002 3.6 \u53e5\u578b\u898f\u5247\u6b78\u7d0d(Induction) \u5047\u8a2d\u5f9e\u6240\u6709\u8a9e\u6599\u4e2d\u5f97\u5230\u7684\u53e5\u578b\u898f\u5247\u53ef\u88ab\u8868\u793a\u70ba\u7dad\u5ea6\u70ba L \u7684\u898f\u5247\u5411\u91cf Rule\uff0c\u6bcf\u500b\u7dad\u5ea6\u5c0d \u61c9\u8457\u4e00\u689d\u53e5\u578b\u898f\u5247\u3002\u5247\u6b64\u898f\u5247\u5411\u91cf\u548c\u6240\u6709\u7684 DA \u53ef\u69cb\u6210\u4e00\u500b\u77e9\u9663\u4f86\u5efa\u7acb\u53e5\u578b\u898f\u5247\u8207\u610f\u5716 \u4e4b\u9593\u7684\u95dc\u4fc2\uff0c\u6b64\u95dc\u4fc2\u53ef\u5b9a\u7fa9\u70ba\uff1a\u3002 \uf0fa \uf0fa \uf0fa \uf0fa \uf0fa \uf0fb \uf0f9 \uf0ea \uf0ea \uf0ea \uf0ea \uf0ea \uf0eb \uf0e9 = \u00d7 Q L L L Q Q Q L , 2 , 1 , , 2 2 , 2 1 , 2 , 1 2 , 1 1 , 1 \u03c6 \u03c6 \u03c6 \u03c6 \u03c6 \u03c6 \u03c6 \u03c6 \u03c6 \uf04c \uf04d \uf04f \uf04d \uf04d \uf04c \uf04c \u03a6 (10) \u5176\u4e2d\u03a6 L\u00d7Q \u662f\u7dad\u5ea6\u70ba L\u00d7Q \u7684\u6587\u6cd5\u7d50\u69cb\u8cc7\u8a0a\u77e9\u9663\uff0cL \u4ee3\u8868\u8a13\u7df4\u8a9e\u6599\u6240\u6709\u53e5\u578b\u898f\u5247\u7684\u500b\u6578\uff0cQ \u4ee3\u8868\u610f\u5716\u7684\u7e3d\u6578\u3002\u77e9\u9663\u4e2d\u6bcf\u500b\u5143\u7d20 \u03c6 l,q \u4ee3\u8868\u8457\u7b2c l \u689d\u6587\u6cd5\u898f\u5247 Rule l \u5728\u7b2c q \u500b DA \u4e2d\u6240 \u4f54\u7684\u91cd\u8981\u6027\u3002\u56e0\u6b64\u672c\u7814\u7a76\u4e2d\u5b9a\u7fa9 \u03c6 l,q \u7684\u4f30\u8a08\u6cd5\u5982\u4e0b\uff1a ) | ( ) 1 ( , q l l q l DA Rule P \u03b5 \u03c6 \u2212 = (11) \u5176\u4e2d\uff0cP(Rule l | DA q )\u662f\u8a72\u689d\u898f\u5247\u4f54\u8a72\u53e5\u8a9e\u6cd5\u7d50\u69cb\u7684\u6bd4\u91cd\uff0c\u8a72\u9805\u53ef\u4ee5\u5beb\u70ba\uff1a \uf0e5 = k q k q l q l DA Rule C DA Rule C DA Rule P ) , ( ) , ( ) | ( (12) \u4e14 C(Rule l , DA q )\u8868\u793a\u53e5\u578b\u898f\u5247 Rule l \u51fa\u73fe\u5728 DA q \u4e2d\u7684\u6b21\u6578\u3002\u53e6\u5916\uff0c( 1-\u03b5 l ) \u662f\u5229\u7528\u91cf\u5ea6\u6587 \u5b57\u4e82\u5ea6 (Entropy) \u7684\u65b9\u6cd5\u4f86\u91cf\u5ea6\u67d0\u689d\u898f\u5247\u5728\u8a72\u8a9e\u6599\u4e2d\u662f\u5426\u5177\u6709\u9451\u5225\u6027\u4e26\u8ce6\u4e88\u8a72\u5143\u7d20\u7684 \u6b0a\u91cd\uff0c\u5247\u03b5 l \u53ef\u5b9a\u7fa9\u70ba\uff1a \uf0e5 \uf0e5 \uf0e5 = = = \u2212 = Q q Q i i l q l Q i i l q l l DA Rule C DA Rule C DA Rule C DA Rule C Q 1 1 1 ) , ( ) , ( log ) , ( ) , ( log 1 \u03b5", |
| "eq_num": "(13)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "3.8 \u6f5b\u5728\u5c0d\u8a71\u884c\u70ba\u77e9\u9663\u6a21\u578b(Latent DA Model, LDAM) ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u7d93\u7531\u90e8\u4efd\u6a23\u672c\u53e5\u8207\u586b\u5145\u5b57\u64f7\u53d6\u5f8c\uff0c\u6211\u5011\u5c07\u5f97\u5230\u7684\u53e5\u578b\u898f\u5247\u8207\u610f\u5716\u53e5\u578b\u5206\u985e\u5f8c\u7684\u985e\u5225\u5efa\u7acb \u95dc\u4fc2\u77e9\u9663\uff0c\u800c\u7522\u751f\u7684\u53e5\u578b\u898f\u5247\u6bd4\u539f\u672c\u8a13\u7df4\u7684\u6587\u5b57\u8a9e\u6599\u6240\u7522\u751f\u7684\u53e5\u578b\u898f\u5247\u5305\u542b\u66f4\u591a\u610f\u6db5\uff0c \u6240\u4ee5\u6211\u5011\u7a31\u6b64\u77e9\u9663\u70ba\u6f5b\u5728\u610f\u5716\u77e9\u9663\u4e14\u5b9a\u7fa9\u70ba\uff1a \uf0fa \uf0fa \uf0fa \uf0fa \uf0fa \uf0fb \uf0f9 \uf0ea \uf0ea \uf0ea \uf0ea \uf0ea \uf0eb \uf0e9 = \u00d7 M L L L M M M L , 2 , 1 , , 2 2 , 2 1 , 2 , 1 2 , 1 1 , 1 \u03bd \u03bd \u03bd \u03bd \u03bd \u03bd \u03bd \u03bd \u03bd \uf04c \uf04d \uf04f \uf04d \uf04d \uf04c \uf04c LDAM (17) \u5176\u4e2d\u610f\u5716\u77e9\u9663 LDAM L\u00d7M \u70ba\u4e00\u500b\u7dad\u5ea6\u70ba L\u00d7M \u7684\u6587\u6cd5\u7d50\u69cb\u8cc7\u8a0a\u77e9\u9663\uff0cL \u4ee3\u8868\u8a13\u7df4\u8a9e\u6599\u7d93\u904e \u8a9e\u53e5\u898f\u5247\u7522\u751f\u6b65\u9a5f\u5f8c\u6240\u6709\u7684\u53e5\u578b\u898f\u5247\u500b\u6578\uff0cM \u4ee3\u8868\u610f\u5716\u53e5\u578b\u7fa4\u805a\u5f8c\u7684\u985e\u5225\u7e3d\u6578\u3002\u77e9\u9663\u4e2d \u6bcf\u500b\u5143\u7d20 \u03bd lm \u4ee3\u8868\u8457\u7b2c l \u689d\u53e5\u578b\u898f\u5247\u5728\u7b2c m \u500b DA \u4e2d\u6240\u4f54\u7684\u91cd\u8981\u6027\u3002\u56e0\u6b64\u672c\u7814\u7a76\u4e2d\u5b9a\u7fa9 \u03bd lm \u7684\u4f30\u8a08\u6cd5\u5982\u4e0b\uff1a ) | ( ) 1 (", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Spoken Language Processing", |
| "authors": [ |
| { |
| "first": "X.-D", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "" |
| }, |
| { |
| "first": "Alex", |
| "middle": [], |
| "last": "Acero", |
| "suffix": "" |
| }, |
| { |
| "first": "H.-Wd", |
| "middle": [], |
| "last": "Hon", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "X.-D Huang, Alex Acero, H.-Wd Hon, \"Spoken Language Processing\", Prentice-Halln, Inc. 2001", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Citybrowser \u2161: A multimodal restaurant guide in Mandarin", |
| "authors": [ |
| { |
| "first": "Ji.-J", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| }, |
| { |
| "first": "Y.-S", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Seneff", |
| "suffix": "" |
| }, |
| { |
| "first": "Victor", |
| "middle": [], |
| "last": "Zue", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "Proc. International Chinese Spoken Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ji.-J. Liu, Y.-S. Xu, S. Seneff, and Victor Zue, \"Citybrowser \u2161: A multimodal restaurant guide in Mandarin\", in Proc. International Chinese Spoken Language Processing, 2008.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "How May I Help You?", |
| "authors": [], |
| "year": 2002, |
| "venue": "AT&T", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "AT&T(2002) How May I Help You? [Online].", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Available", |
| "authors": [], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Available: http://www.research.att.com/~algot/hmihy/", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Dialog Management using Weighted Finite-State Transducers", |
| "authors": [ |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Hori", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Ohtake", |
| "suffix": "" |
| }, |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Misu", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Kashioka", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Nakamura", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "C. Hori, K. Ohtake, T. Misu, H. Kashioka, S. Nakamura, \"Dialog Management using Weighted Finite-State Transducers\", Interspeech, 2008", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Dialogue in the RAILTEL Telephone-Based System", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Bennacef", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Devillers", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Rosset", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Lamel", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "Proc. of ICSKP'96", |
| "volume": "1", |
| "issue": "", |
| "pages": "550--553", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Bennacef, L. Devillers, S. Rosset, and L. Lamel, \"Dialogue in the RAILTEL Telephone-Based System\", in Proc. of ICSKP'96, vol. 1, pp. 550-553, 1996", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "A Multi-keyword Spotter for the Application of the TL Phone Directory Assistant Service", |
| "authors": [ |
| { |
| "first": "C.-J", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| }, |
| { |
| "first": "E.-F", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "" |
| }, |
| { |
| "first": "J.-K", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "Proc. Workshop on Distributed System Technologies & Applications", |
| "volume": "", |
| "issue": "", |
| "pages": "197--202", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "C.-J. Lee, E.-F. Huang, and J.-K. Chen, \"A Multi-keyword Spotter for the Application of the TL Phone Directory Assistant Service\", in Proc. Workshop on Distributed System Technologies & Applications, pp. 197-202, 1997", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "The Design of a Mandarin Chinese Spoken Dialogue System", |
| "authors": [ |
| { |
| "first": "T.-H", |
| "middle": [], |
| "last": "Chiang", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-M", |
| "middle": [], |
| "last": "Peng", |
| "suffix": "" |
| }, |
| { |
| "first": "Y.-C", |
| "middle": [], |
| "last": "Lin", |
| "suffix": "" |
| }, |
| { |
| "first": "H.-M", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "S.-C", |
| "middle": [], |
| "last": "Chieh", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Proc. COTEC'98", |
| "volume": "", |
| "issue": "", |
| "pages": "2--5", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T.-H. Chiang, C.-M. Peng, Y.-C. Lin, H.-M. Wang and S.-C. Chieh, \"The Design of a Mandarin Chinese Spoken Dialogue System\", in Proc. COTEC'98, Taipei 1998, pp.E2-5.1~E2-5.7", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Proc. ROCLING XV", |
| "authors": [ |
| { |
| "first": "\u8449\u745e\u5cf0", |
| "middle": [], |
| "last": "\u9673\u9298\u8ecd", |
| "suffix": "" |
| }, |
| { |
| "first": "\u5433\u5b97\u61b2", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| }, |
| { |
| "first": "\"", |
| "middle": [], |
| "last": "\u4ee5\u77e5\u8b58\u6982\u5ff5\u6a21\u578b\u70ba\u57fa\u790e\u4e4b\u591a\u4e3b\u984c\u5c0d\u8a71\u7ba1\u7406\u7cfb\u7d71", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "\u9673\u9298\u8ecd, \u8449\u745e\u5cf0, \u5433\u5b97\u61b2, \"\u4ee5\u77e5\u8b58\u6982\u5ff5\u6a21\u578b\u70ba\u57fa\u790e\u4e4b\u591a\u4e3b\u984c\u5c0d\u8a71\u7ba1\u7406\u7cfb\u7d71\", in Proc. ROCLING XV, Hsinchu, Taiwan, 2003.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "An Integrated Dialogue Analysis Model for Determining Speech Acts and Discourse Structures", |
| "authors": [ |
| { |
| "first": "W.-S", |
| "middle": [], |
| "last": "Choi", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Kim", |
| "suffix": "" |
| }, |
| { |
| "first": "J.-Y", |
| "middle": [], |
| "last": "Seo", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "W.-S. Choi, H. Kim, and J.-Y. Seo, \"An Integrated Dialogue Analysis Model for Determining Speech Acts and Discourse Structures,\" the Institute of Electronics, Information and Communication Engineers (IEICE), 2005", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Partially Observable Markov Decision Processes for Spoken Dialog Systems", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [ |
| "D" |
| ], |
| "last": "Williams", |
| "suffix": "" |
| }, |
| { |
| "first": "Steve", |
| "middle": [], |
| "last": "Young", |
| "suffix": "" |
| } |
| ], |
| "year": 2007, |
| "venue": "Computer Speech and Language", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. D. Williams, and Steve Young, \"Partially Observable Markov Decision Processes for Spoken Dialog Systems,\" Computer Speech and Language, 2007.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Speech Act for Dialogue Agents", |
| "authors": [ |
| { |
| "first": "David", |
| "middle": [ |
| "R" |
| ], |
| "last": "Traum", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "David R. Traum, \"Speech Act for Dialogue Agents,\" Kluwer Academic Publishers, 1999.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "MHMC Annotation of MHMC Travel Corpus", |
| "authors": [ |
| { |
| "first": "Y.-C", |
| "middle": [], |
| "last": "Xiao", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Y.-C. Xiao, \"MHMC Annotation of MHMC Travel Corpus,\" 2009. [Online]. Avaliable: http://chinese.csie.ncku.edu.tw/~liang/MHMC_Annotation_of_Travel_Corpus.pdf", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Word Predictability After Hesitations: A Corpus-Based Study", |
| "authors": [ |
| { |
| "first": "E", |
| "middle": [], |
| "last": "Shriberg", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Stolcke", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "Proc. International on Conference Spoken Language Processing (ICSLP)", |
| "volume": "", |
| "issue": "", |
| "pages": "1868--1871", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "E. Shriberg and A. Stolcke, \"Word Predictability After Hesitations: A Corpus-Based Study\", in Proc. International on Conference Spoken Language Processing (ICSLP), pp. 1868-1871, 1996.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Modeling Disfluencies in Conversational Speech", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Siu", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Ostendorf", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Gish", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "Proc. International on Conference Spoken Language Processing", |
| "volume": "1", |
| "issue": "", |
| "pages": "386--389", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M Siu, M. Ostendorf, and H. Gish, \"Modeling Disfluencies in Conversational Speech\" , in Proc. International on Conference Spoken Language Processing (ICSLP), vol 1, pp. 386-389, 1996.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Variable-Length Category N-gram Language Models", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [ |
| "R" |
| ], |
| "last": "Niesler", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [ |
| "C" |
| ], |
| "last": "Woodland", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Computer, Speech and Language", |
| "volume": "21", |
| "issue": "", |
| "pages": "1--26", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T.R. Niesler and P.C. Woodland, \"Variable-Length Category N-gram Language Models\", Computer, Speech and Language, vol. 21, pp. 1-26, 1999.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "Towards Building a Better Language Model for Switchboard: the POS Tagging Task", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [ |
| "S" |
| ], |
| "last": "Hamaker", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Proc. International Conference on Acoustics, Speech, and Signal Processing(ICASSP)", |
| "volume": "", |
| "issue": "", |
| "pages": "579--582", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. S. Hamaker, \"Towards Building a Better Language Model for Switchboard: the POS Tagging Task,\" in Proc. International Conference on Acoustics, Speech, and Signal Processing(ICASSP), pp. 579-582, 1999.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Confidence Measures for Large Vocabulary Continuous Speech Recognition", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Wessel", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Schluter", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Macherey", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Ney", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "IEEE Trans. on Speech and Audio Processing", |
| "volume": "9", |
| "issue": "3", |
| "pages": "288--298", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. Wessel, R. Schluter, K. Macherey, and H. Ney, \"Confidence Measures for Large Vocabulary Continuous Speech Recognition,\" IEEE Trans. on Speech and Audio Processing, vol. 9, no. 3, pp. 288-298, 2001", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Fast Exact Inference with a Factored Model for Natural Language Parsing", |
| "authors": [ |
| { |
| "first": "Dan", |
| "middle": [], |
| "last": "Klein", |
| "suffix": "" |
| }, |
| { |
| "first": "C", |
| "middle": [ |
| "D" |
| ], |
| "last": "Manning", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Advances in Neural Information Processing Systems", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Dan Klein, and C. D. Manning, \"Fast Exact Inference with a Factored Model for Natural Language Parsing,\" in Advances in Neural Information Processing Systems, 2003.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "Accurate Unlexicalized Parsing", |
| "authors": [ |
| { |
| "first": "Dan", |
| "middle": [], |
| "last": "Klein", |
| "suffix": "" |
| }, |
| { |
| "first": "C", |
| "middle": [ |
| "D" |
| ], |
| "last": "Manning", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Proc. the 41st Meeting of the Association for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "423--430", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Dan Klein, and C. D. Manning, \"Accurate Unlexicalized Parsing,\" in Proc. the 41st Meeting of the Association for Computational Linguistics, pp. 423-430, 2003.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "type_str": "figure", |
| "num": null, |
| "text": "\u5176\u4e2d P(Rule l |DA m )\u662f\u8a72\u689d\u898f\u5247\u4f54\u8a72\u53e5\u8a9e\u6cd5\u7d50\u69cb\u7684\u6bd4\u91cd\uff0c\u8a72\u9805\u53ef\u4ee5\u5beb\u70ba\uff1a \uf0e5 C(Rule l , DA m )\u8868\u793a\u53e5\u578b\u898f\u5247 l \u51fa\u73fe\u5728\u7b2c m \u500b\u610f\u5716\u53e5\u578b\u5206\u985e\u5f8c\u7684\u985e\u5225\u4e2d\u7684\u6b21\u6578\u3002\u53e6\u5916\uff0c (1-\u03b5 l )\u662f\u5229\u7528\u91cf\u5ea6\u6587\u5b57\u4e82\u5ea6 (Entropy) \u7684\u65b9\u6cd5\u4f86\u5ea6\u91cf\u8a72\u689d\u898f\u5247\u5728\u8a9e\u6599\u4e2d\u7684\u9451\u5225\u6027\uff0c\u7576\u4f5c\u77e9 \u9663\u4e2d\u8a72\u5143\u7d20\u7684\u6b0a\u91cd\uff0c\u03b5 l \u53ef\u5b9a\u7fa9\u70ba\uff1a \u5176\u4e2d\u5927\u5beb C \u70ba\u7531\u4e0a\u8ff0 K-means \u5206\u7fa4\u800c\u5f97\u6700\u7d42 DA \u6578\u91cf\uff0c\u800c DA C \u6240\u8868\u793a\u7684\u5c31\u662f LDAM \u7b2c c \u500b column \u7684\u5167\u5bb9\u3002\u6700\u5f8c\uff0c\u6211\u5011\u5f97\u5230 SLU \u5075\u6e2c\u8a9e\u7fa9\u6240\u9700\u8981\u7684\u6a21\u578b\u3002", |
| "uris": null |
| }, |
| "TABREF0": { |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td colspan=\"2\">\u7968\u50f9\u662f\u591a\u5c11\uff1f\u300d\u9019\u53e5\u8a71\u88ab\u8868\u9054\u51fa\u4f86\u7684 DA \u70ba\u8a62\u554f\u7968\u50f9\u3002\u56e0\u6b64\uff0c\u8a9e\u8a00\u5b78\u5bb6\u7a31\u6240\u6709\u985e\u578b\u7684\u6e9d \u4e09\u3001 \u7cfb\u7d71\u67b6\u69cb DA H \u7368\u7acb\uff0c\u5247\u6211\u5011\u53ef\u4ee5\u5f97\u5230\u5f0f\u5b50(5)\u3002\u6b64\u5916\uff0cP(DA C |W i )\u70ba\u7d93\u7531\u8c9d\u6c0f\u6c7a\u7b56\u6cd5\u5247(Bayes'</td></tr><tr><td colspan=\"2\">\u901a\u884c\u70ba\u70ba\"\u5c0d\u8a71\u884c\u70ba\"[12]\u3002\u5b8c\u6210\u8a9e\u6599\u6536\u96c6\u5f8c\uff0c\u4f9d\u64da\u7cfb\u7d71\u63d0\u4f9b\u7684\u4efb\u52d9(task)\uff0c\u6211\u5011\u5206\u6790\u5c0d\u8a71 decision rule)\u800c\u4f86\uff0c\u56e0\u6b64\u6211\u5011\u5c07\u5f0f\u5b50\u9032\u4e00\u6b65\u6539\u5beb\u70ba\u5f0f\u5b50(6)\uff1a ]\u4e2d\u986f\u793a\u51fa\uff0c\u4e00\u500b\u8a5e\u5f59\u51fa\u73fe\u5728\u9019\u4e9b\u4e0d\u6d41\u5229</td></tr><tr><td colspan=\"2\">\u4f86\u5be6\u73fe\u3002\u9019\u4e00\u90e8\u5206\u7684\u7814\u7a76\u4e3b\u984c\u70ba\u7cfb\u7d71\u8207 \u4f7f\u7528\u8005\u4e92\u52d5\u4e2d\uff0c\u5148\u5224\u65b7\u4f7f\u7528\u8005\u7684\u610f\u5716\uff0c\u518d\u5c0d\u6b64\u610f\u5716\u4f5c\u51fa\u6700\u9069\u7576\u7684\u56de\u61c9\u3002 \u8fd1\u5e7e\u5e74\u4f86\uff0c\u53e3\u8ff0\u8a9e\u8a00\u5c0d\u8a71\u7cfb\u7d71\u5df2\u7d93\u6709\u986f\u8457\u7684\u9032\u6b65\uff0c\u5c24\u5176\u662f\u5efa\u69cb\u65bc\u586b\u8868\u5f0f(slot-filling)\u7684\u8cc7 \u6599\u5eab\u67e5\u8a62\u65b9\u6cd5\u5df2\u9032\u5165\u61c9\u7528\u7684\u968e\u6bb5\u3002\u7136\u800c\uff0c\u56e0\u70ba\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u9020\u6210\u8868\u683c\u586b\u5165\u932f\u8aa4\uff0c\u9032\u4e00\u6b65 \u4f7f\u5f97\u81ea\u7136\u8a9e\u8a00\u7406\u89e3\u7522\u751f\u932f\u8aa4\u53ca\u8aa4\u5224\u4f7f\u7528\u8005\u7684\u610f\u5716\uff0c\u5c0e\u81f4\u7cfb\u7d71\u56de\u61c9\u932f\u8aa4\u3002\u9019\u985e\u578b\u7684\u554f\u984c\u5c1a \u672a\u6210\u529f\u5730\u89e3\u6c7a\u3002\u56e0\u6b64\uff0c\u5982\u4f55\u6709\u6548\u5730\u5728\u5177\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u7684\u689d\u4ef6\u4e0b\u4ecd\u80fd\u5f97\u5230\u597d\u7684\u610f\u5716\u5075\u6e2c\u7d50 \u610f\u5716\uff0c\u4f8b\u5982\u4f7f\u7528\u8005\u6b63\u5728\u67e5\u8a62\u9ad8\u9435\u6642\u523b\u8868\uff0c\u7cfb\u7d71\u537b\u9032\u5165\u65c5\u904a\u666f\u9ede\u7684\u76f8\u95dc\u8cc7\u8a0a\u67e5\u8a62\u529f\u80fd\u3002 \u679c\u662f\u6211\u5011\u7684\u7814\u7a76\u4e3b\u8981\u76ee\u6a19\u3002\u9664\u6b64\u4e4b\u5916\uff0c\u6211\u5011\u4ea6\u5e0c\u671b\u80fd\u5f97\u5230\u826f\u597d\u7684\u4eba\u6a5f\u4e92\u52d5\uff0c\u6240\u4ee5\u4e00\u500b\u6709 \u8a9e\u6599\u4e26\u4e14\u5728\u6bcf\u4e00\u7a2e\u7cfb\u7d71\u4efb\u52d9\u4e2d\u8a2d\u8a08\u4e86\u8a72\u4efb\u52d9\u6240\u542b\u62ec\u7684\u8868\u55ae(slot)\u503c\u53ca\u6bcf\u500b\u8868\u55ae\u53ef\u586b\u5165\u503c\uff0c \u5982\u8868 2 \u6240\u793a\u3002\u5728\u6211\u5011\u6240\u6536\u9304\u7684\u8a9e\u6599\u4e2d\uff0c\u53ef\u5206\u70ba\u4e09\u5927\u4efb\u52d9\uff0c\u5206\u5225\u70ba\u67e5\u8a62\u7cfb\u7d71\u670d\u52d9\u3001\u67e5\u8a62 \u666f\u9ede\u76f8\u95dc\u8cc7\u8a0a\u548c\u67e5\u8a62\u4ea4\u901a\u76f8\u95dc\u8cc7\u8a0a\uff0c\u5176\u4e2d\u7b2c j \u500b\u4efb\u52d9\u6240\u5305\u542b\u7684 DA \u6578\u8868\u793a\u70ba\uff1a Task \u7684 DA \u6578 Slot \u53ef\u586b\u5165\u503c\u6578 T \u5305\u542b\u7684 \u6578 1 (1) \u5176\u4e2d-1 \u662f\u56e0\u70ba\u672a\u586b\u503c\u53ef\u80fd\u5728\u5176\u4ed6\u4efb\u52d9\u4e2d\u6709\u586b\u503c\uff0c\u4f8b\u5982\uff1a\u4efb\u52d9 1 \u7684\"\u6211\u60f3\u67e5\u8a62\u9ad8\u9435\"\u548c\u4efb \u52d9 3 \u7684 \"\u8ddf\u6211\u8aaa\u9ad8\u9435\u7684\u6642\u523b\u8868\"\u3002 \u5176\u4ed6\u7684\u610f\u5716\u5305\u62ec\u6b61\u8fce\u3001\u7d50\u675f\u3001\u7121\u610f\u5716\uff0c\u7e3d\u5171 38 \u500b\u3002 \u7576\u5c0d\u8a71\u7cfb\u7d71\u5075\u6e2c\u51fa\u4f7f\u7528\u8005\u7684 DA \u5f8c\uff0c\u5c0d\u8a71\u7cfb\u7d71\u61c9\u4f5c\u51fa\u5408\u7406\u7cfb\u7d71\u56de\u61c9\u884c\u70ba\u4ee5\u9054\u5230\u548c\u4f7f\u7528\u8005 \u4e4b\u9593\u7684\u4e92\u52d5\u3002\u56e0\u6b64\uff0c\u6211\u5011\u6839\u64da DA \u7684\u5167\u5bb9\u6574\u7406\u51fa\u5982\u5716 2 \u6700\u53f3\u6b04\u4f4d\u6240\u793a\u7684\u7cfb\u7d71\u56de\u61c9\u3002\u5927 \u81f4\u53ef\u5206\u70ba\u7cfb\u7d71\u8a62\u554f\u672a\u586b\u503c\u7684\u8868\u683c\u8cc7\u8a0a\u548c\u56de\u7b54\u8cc7\u8a0a\u7684\u884c\u52d5\uff0c\u7e3d\u5171\u6709 20 \u7a2e [13]\u3002 \u8868 1\uff1a\u8a9e\u7fa9\u985e\u5225\u7bc4\u4f8b\u53ca\u5176\u5c0d\u61c9\u7684\u95dc\u9375\u8a5e\u5f59\u7bc4\u4f8b\u3002 \u672c\u8ad6\u6587\u7684\u7cfb\u7d71\u67b6\u69cb\u5982\u5716 1 \u6240\u793a\uff0c\u865b\u7dda\u4ee5\u4e0a\u70ba\u4f7f\u7528\u8005\u90e8\u4efd\uff0c\u5305\u62ec\u767c\u97f3 U \u548c\u63a5\u6536\u7cfb\u7d71\u56de\u61c9 \u8a0a\u606f\u6240\u7522\u751f\u7684\u8072\u97f3 U\u2032\uff1b\u800c\u865b\u7dda\u4ee5\u4e0b\u70ba\u5c0d\u8a71\u7cfb\u7d71\u90e8\u5206\uff0c\u4e3b\u8981\u5206\u70ba\u4e09\u500b\u90e8\u4efd\uff0c\u5305\u62ec\u8f38\u5165\u8655 \u7406(input processing) \u3001\u5c0d\u8a71\u7ba1\u7406(dialogue management, DM)\u548c\u8f38\u51fa\u8655\u7406(output processing)\u3002 \u5728\u8f38\u5165\u8655\u7406\u90e8\u5206\uff0c\u8a9e\u8005\u7684\u767c\u97f3 U \u7d93\u7531\u9ea5\u514b\u98a8\u50b3\u9001\u81f3\u81ea\u52d5\u8a9e\u97f3\u8fa8\u8b58\u5668(automatic speech DA* argmax max ( | ) ( | ) ( | U) C i C H i P DA W P DA DA P W = W \u7684\u8d05\u8a9e\u4e4b\u5f8c\u7684\u5e73\u5747\u6a5f\u7387\u6703\u5c0f\u65bc\u51fa\u73fe\u5728\u7121\u8d05\u8a9e\u4e4b\u5f8c\u7684\u60c5\u6cc1\u3002\u9019\u6307\u51fa\u6211\u5011\u5f88\u96e3\u53bb\u9810\u6e2c\uff0c\u7576\u4e00 (5) \u500b\u8a5e\u5f59\u51fa\u73fe\u5728\u591a\u9918\u8d05\u8a9e\u4e4b\u5f8c\u7684\u6a5f\u7387\u3002\u56e0\u6b64\u6839\u64da\u9019\u4e9b\u89c0\u5bdf\uff0c\u6211\u5011\u5c07\u7701\u7565\u6389\u8d05\u8a9e\u6216\u6c92\u88ab\u8fa8\u8b58 DA ) U | ( ) | ( ) ( ) ( ) | ( max argmax i H C i C C i W P DA DA P W P DA P DA W \u51fa\u4f86\u7684\u8a5e\u5f59\u800c\u7522\u751f\u7684\u53e5\u5b50\uff0c\u7a31\u4e4b\u70ba PP\u3002\u66f4\u9032\u4e00\u6b65\u63a2\u8a0e\uff0c\u90e8\u4efd\u6a23\u672c\u6a39\u4e00\u500b\u5f88\u91cd\u8981\u7684\u61c9\u7528\uff0c P W DA = \u5c31\u662f\u91dd\u5c0d\u66ff\u4ee3\u6027\u932f\u8aa4\u505a\u4fee\u6b63\uff0c\u800c\u4e4b\u524d\u7684\u7814\u7a76\u5c0d\u65bc\u932f\u8aa4\u7684\u56de\u5fa9(recovery)\u901a\u5e38\u90fd\u662f\u5229\u7528\u53e5 (6) \u578b\u898f\u5247\u5728\u773e\u591a\u7684\u5019\u9078\u53e5\u4e2d\u627e\u51fa\u6700\u7b26\u5408\u8a9e\u6cd5\u53e5\u578b\u7684\u53e5\u5b50[17][18]\u3002\u7136\u800c\u9019\u4e9b\u65b9\u6cd5\u6240\u7522\u751f\u7684 recognizer, ASR)\u4e26\u7522\u751f\u8fa8\u8b58\u7d50\u679c W\uff0c\u6b64\u4e00\u8fa8\u8b58\u7d50\u679c\u5c07\u9001\u81f3\u53e3\u8ff0\u8a9e\u8a00\u7406\u89e3(spoken language \u5176\u4e2d\uff0c P(W i )\u53ef\u88ab\u7701\u7565\uff0cP(DA C )\u5728\u672c\u8ad6\u6587\u5047\u8a2d\u70ba\u5747\u7b49\u4e8b\u524d\u6a5f\u7387(equal prior)\uff0c\u6240\u4ee5\u6700\u5f8c\u5f97 \u53e5\u5b50\u5728\u53e5\u578b\u4e0a\u96d6\u7136\u975e\u5e38\u7b26\u5408\uff0c\u4f46\u662f\u537b\u53ef\u80fd\u5728\u8a9e\u7fa9\u4e0a\u7684\u610f\u7fa9\u662f\u4e0d\u8db3\u7684\u3002\u6240\u4ee5\u91dd\u5c0d\u6b64\u9ede\u6211\u5011 understanding, SLU)\u55ae\u5143\u9032\u884c\u6f5b\u5728\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u800c\u5f97\u5230\u7b2c c \u985e\u5c0d\u8a71\u884c\u70ba DA C \uff0c\u800c SLU \u4e5f \u5230\u5f0f\u5b50\uff1a \u6240\u5b9a\u7fa9\u7684\u6bcf\u4e00\u53e5\u90e8\u5206\u6a23\u672c\u53e5\uff0c\u90fd\u9700\u4fdd\u7559\u539f\u53e5\u4e2d\u7684\u4e3b\u8981\u95dc\u9375\u8a5e\u4ee5\u7dad\u6301\u53e5\u5b50\u7684\u8a9e\u7fa9\uff0c\u7136\u800c\u529f \u662f\u672c\u8ad6\u6587\u7684\u6838\u5fc3\uff0c\u6211\u5011\u5c07\u9010\u4e00\u4ecb\u7d39 SLU \u7684\u8a13\u7df4\u65b9\u6cd5\u548c\u63a1\u7528\u7684\u6280\u8853\u3002\u7136\u5f8c\uff0cDA C \u5c07\u50b3\u9001 \u5230\u5c0d\u8a71\u7ba1\u7406\u4e26\u6839\u64da\u7531 POMDP \u6240\u8a13\u7df4\u800c\u5f97\u7684\u5c0d\u8a71\u7b56\u7565(strategy)\u548c\u5c0d\u8a71\u610f\u5716\u6b77\u53f2\u8a18\u9304 ) | ( ) | ( ) U | ( max argmax DA* H C C i i DA DA P DA W P W P W DA \u80fd\u6027\u8a5e\u5f59\u5247\u6709\u53ef\u80fd\u88ab\u7701\u7565\u3002\u56e0\u6b64\u6839\u64da\u4e0a\u8ff0\u89c0\u5bdf\uff0c\u6211\u5011\u63d0\u51fa\u4e86\u5229\u7528 PP \u4f86\u5efa\u7acb\u6548\u80fd\u66f4\u4f73\u7684 \u2248 (7) \u8a9e\u53e5\u6a21\u7d44\uff0c\u5728\u9019\u88e1\u6211\u5011\u5c07\u53e5\u5b50 Trans i \u8996\u70ba\u4e00\u9023\u4e32\u7684 OP \u8207\u4e00 MP \u7684\u7d44\u5408\uff0c\u8868\u793a\u6210\uff1a (dialogue act history) DA H \u4f86\u63a1\u53d6\u5408\u9069\u7684\u56de\u61c9(action) a t \u3002\u7576\u7cfb\u7d71\u505a\u51fa\u56de\u61c9\u5f8c\uff0c\u7cfb\u7d71\u5c07\u5f9e\u6211 \u5011\u8490\u96c6\u800c\u4f86\u7684\u65c5\u6709\u8cc7\u8a0a\u8cc7\u6599\u5eab(travel information database)\u4e2d\u67e5\u8a62\u5c0d\u61c9\u7684\u8cc7\u6599\u4e26\u8f38\u51fa\u6587\u5b57 \u90e8\u4efd\uff1bP(DA C |DA H )\u70ba\u5c0d\u8a71\u610f\u5716\u6b77\u53f2(dialogue history)\u6a5f\u7387\uff0c\u7528\u4f86\u9632\u6b62\u7cfb\u7d71\u8df3\u812b\u4f7f\u7528\u8005\u7684 \u4e0b\u6211\u5011\u5c07\u9010\u4e00\u4ecb\u7d39\u5404\u500b\u90e8\u4efd\u3002 \u8b58\u5b57\u4e32\u88ab\u5075\u6e2c\u70ba\u7b2c c \u500b DA \u7684\u5075\u6e2c\u6a5f\u7387(probability of DA detection)\uff0c\u5373\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u7684 Context t \u81f3\u8a9e\u97f3\u5408\u6210\u5668(text-to-speech synthesizer, TTS)\u7522\u751f\u8a9e\u97f3\u8cc7\u8a0a U\u2032\u50b3\u9054\u7d66\u4f7f\u7528\u8005\u3002\u4ee5 \u5176\u4e2d P(W i |U)\u70ba\u81ea\u52d5\u8a9e\u97f3\u8fa8\u8b58\u5668\u5f9e\u8a9e\u97f3 U \u6240\u5f97\u5230\u7684\u8fa8\u8b58\u5b57\u4e32 W \u7684\u6a5f\u7387\uff1bP(W i |DA C )\u70ba\u8fa8 } ,..., , , ,..., , { 1 2 1 i i NA i NB i i NB i i i NB i i i OP OP MP OP OP OP Trans + + = (8)</td></tr><tr><td colspan=\"2\">\u6548\u7684\u7cfb\u7d71\u56de\u61c9\u6a5f\u5236\u4ee5\u907f\u514d\u5c0d\u8a71\u767c\u6563\u4e4b\u7a98\u5883\uff0c\u8b93\u4f7f\u7528\u8005\u4e0d\u81f4\u65bc\u5c0d\u7cfb\u7d71\u7522\u751f\u6392\u65a5\u751a\u81f3\u53ad\u60e1\u9032</td></tr><tr><td>\u800c\u63d0\u9ad8\u5176\u53ef\u884c\u6027\uff0c\u4e5f\u662f\u6211\u5011\u6240\u8003\u91cf\u7684\u90e8\u4efd\u3002</td><td/></tr><tr><td colspan=\"2\">\u5728\u672c\u8ad6\u6587\u7684\u5176\u4ed6\u6bb5\u843d\u5b89\u6392\u5982\u4e0b\u3002\u7b2c\u4e8c\u7bc0\u63cf\u8ff0\u6211\u5011\u70ba\u4e86\u672c\u5be6\u9a57\u6240\u8490\u96c6\u7684\u65c5\u904a\u76f8\u95dc\u8cc7\u8a0a\u8a9e\u6599</td></tr><tr><td colspan=\"2\">\u53ca\u5176\u5c0d\u8a71\u3001\u8a9e\u7fa9\u985e\u5225\u3001\u5c0d\u8a71\u884c\u70ba\u548c\u5c0d\u8a71\u884c\u70ba\u6240\u5c0d\u61c9\u7684\u884c\u52d5\u6a19\u8a18\u3002\u7136\u5f8c\uff0c\u7b2c\u4e09\u7bc0\u4ecb\u7d39\u8ad6\u6587</td></tr><tr><td colspan=\"2\">\u6838\u5fc3\u7684\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u6a21\u578b\u8207\u5176\u8a13\u7df4\u65b9\u6cd5\u5c07\u88ab\u63cf\u8ff0\u3002\u4e0b\u4e00\u6bb5\u7684\u7b2c\u56db\u7bc0\u70ba\u5c0d\u8a71\u7ba1\u7406\u6c7a\u7b56\u7684\u8a13 \u7df4\u3002\u7b2c\u4e94\u7bc0\u7684\u5be6\u9a57\u8aaa\u660e\u4e86\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u5668\u4e2d\u8a9e\u97f3\u8fa8\u8b58\u5668\u5143\u4ef6\u7684\u8a13\u7df4\u3001\u6578\u7a2e\u5c0d\u8a71\u884c\u70ba\u884c\u70ba \u8868 2\uff1a\u7cfb\u7d71\u4efb\u52d9\u5206\u985e\u53ca\u5176\u8868\u55ae\u548c\u53ef\u586b\u5165\u503c\u4e4b\u5c0d\u7167\u8868\u7bc4\u4f8b\u3002</td></tr><tr><td>\u5075\u6e2c\u6bd4\u8f03\u5be6\u9a57\u548c\u76f8\u95dc\u7d71\u8a08\u8cc7\u6599\u3002\u6700\u5f8c\uff0c\u7d50\u8ad6\u8207\u672a\u4f86\u5c55\u671b\u5c07\u88ab\u8a0e\u8ad6\u65bc\u7b2c\u516d\u7bc0\u3002</td><td/></tr><tr><td>\u4e8c\u3001 \u8a9e\u6599\u6536\u96c6</td><td/></tr><tr><td colspan=\"2\">\u8868 3 \uff1aDA \u5217\u8868\u3001\u4f8b\u53e5\u548c\u5176\u5c0d\u61c9\u7684 Action\u3002 \u5716 1 \uff1a\u5c0d\u8a71\u7cfb\u7d71\u67b6\u69cb\u5716\u3002 3.1 \u53e3\u8ff0\u8a9e\u8a00\u7406\u89e3(SLU) \u5728\u6f5b\u5728\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u904e\u7a0b\u4e2d\uff0c\u7cfb\u7d71\u5fc5\u9808\u6839\u64da\u4f7f\u7528\u8005\u767c\u97f3 U \u548c\u5c0d\u8a71\u610f\u5716\u6b77\u53f2\u8a18\u9304 DA H \u4f86 \u5075\u6e2c\u6700\u4f73\u610f\u5716 DA*\uff0c\u5247\u6b64\u5075\u6e2c\u6cd5\u5247(detection criterion)\u5b9a\u7fa9\u70ba\u5f0f\u5b50\uff1a ) , U | ( argmax DA* H C DA DA P DA = (2) 2.1 \u5728\u4ee5\u586b\u8868(slot-filling)\u65b9\u5f0f\u70ba\u57fa\u790e\u7684\u5c0d\u8a71\u7cfb\u7d71\u4e2d\uff0c\u82e5\u95dc\u9375\u5b57\u8a5e\u904e\u65bc\u6563\u4e82\u5c07\u9593\u63a5\u5c0e\u81f4\u7cfb\u7d71\u6548 \u5176\u4e2d DA \u70ba\u6240\u6709\u53ef\u80fd\u7684 DA \u96c6\u5408\uff0cDA C \u70ba\u767c\u97f3 U \u88ab\u8fa8\u8b58\u70ba\u7b2c c \u985e DA\u3002\u7b2c i \u7a2e\u53ef\u80fd\u8fa8\u8b58</td></tr><tr><td colspan=\"2\">\u80fd\u4e0d\u5f70\u7684\u554f\u984c\uff0c\u76f8\u8f03\u4e4b\u4e0b\uff0c\u82e5\u5c07\u95dc\u9375\u5b57\u63d0\u5347\u5230\u8a9e\u7fa9\u985e\u5225\uff0c\u4e0d\u50c5\u53ef\u5728\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u4e0a\u53ef\u907f \u5b57\u4e32 W i \u70ba U \u7d93\u904e\u8fa8\u8b58\u5f8c\u6240\u5f97\u5230\u7684\u53ef\u80fd\u8fa8\u8b58\u7d50\u679c\u6587\u5b57\u3002\u7136\u800c\uff0c\u6211\u5011\u50c5\u53d6\u6700\u4f73\u7684\u8fa8\u8b58\u7d50\u679c</td></tr><tr><td colspan=\"2\">\u514d\u5075\u6e2c\u985e\u5225\u904e\u591a\u7684\u554f\u984c\uff0c\u5c0d\u65bc\u8a9e\u6599\u5eab\u7684\u64f4\u589e\u8207\u7dad\u8b77\u4e5f\u6709\u8f03\u826f\u597d\u7684\u7ba1\u7406\u3002\u56e0\u6b64\uff0c\u5728\u8a9e\u6599\u5eab \u7684\u8cc7\u6599\u5206\u6790\u904e\u7a0b\u4e2d\uff0c\u6211\u5011\u5c07\u6a19\u8a18\u70ba\u95dc\u9375\u5b57\u7684\u8a5e\u5f59\u4f5c\u9032\u4e00\u6b65\u7684\u6574\u7406\uff0c\u5373\u5982\u8868 1 \u4e2d\u6240\u5448\u73fe \u7684\u5167\u5bb9\u3002\u6700\u5f8c\uff0c\u6211\u5011\u8490\u96c6\u7684\u8a9e\u6599\u5eab\u7e3d\u5171\u5305\u542b 27 \u7a2e\u8a9e\u7fa9\u985e\u5225\u3002 W \uff0c\u56e0\u6b64\u5f0f\u5b50(2)\u6539\u5beb\u70ba\u5f0f\u5b50(3)\uff0c\u9032\u4e00\u6b65\u5c55\u958b\u70ba\u5f0f\u5b50(4)\uff1a * DA \uf0e5 = i W H i DA W DA P ) , U | , ( argmax C DA</td></tr><tr><td>2.3 \u5c0d\u8a71\u884c\u70ba(Dialogue Act)\u548c\u5176\u7cfb\u7d71\u56de\u61c9\u884c\u52d5(Action) ) , U | , ( max argmax H i C DA W DA P W DA \u2248</td><td>(3)</td></tr><tr><td colspan=\"2\">\u7576\u8a9e\u8005\u8aaa\u51fa\u4e86\u4e00\u53e5\u8a71\uff0c\u9019\u53e5\u8a71\u672c\u8eab\u7684\u6587\u5b57\u6709\u5176\u610f\u7fa9\uff0c\u4f46\u8a9e\u8005\u4e4b\u6240\u4ee5\u6703\u8aaa\u51fa\u9019\u53e5\u8a71\u6709\u5404\u7a2e ) , U | ( ) , U , | ( max argmax H i H i C DA W P DA W DA P W DA = (4) \u76ee\u7684\u3002\u9019\u6a23\u5c31\u53ef\u7528\u5c0d\u8a71\u4f86\u505a\u4e00\u500b\u884c\u52d5\uff0c\u9019\u5c31\u662f\u5c0d\u8a71\u884c\u70ba\u3002\u8209\u4f8b\u4f86\u8aaa\uff0c \u300c\u8acb\u554f\u5b89\u5e73\u53e4\u5821\u7684 \u5047\u8a2d\u8fa8\u8b58\u7d50\u679c W i \u8207\u8f38\u5165\u8a9e\u97f3 U \u4ee3\u8868\u76f8\u540c\u610f\u7fa9\u4e14\u8a9e\u97f3\u8fa8\u8b58\u7d50\u679c W i \u8207\u5c0d\u8a71\u610f\u5716\u6b77\u53f2\u8a18\u9304</td></tr></table>", |
| "type_str": "table", |
| "text": "\u8a9e\u6599\u9304\u88fd\u8207\u6a19\u8a18 \u5728\u5be6\u9a57\u5ba4\u74b0\u5883\u4e0b\uff0c\u4f7f\u7528 audio-technica AT9940 \u6578\u4f4d\u9304\u97f3\u9ea5\u514b\u98a8\uff0c\u4ee5 16 \u4f4d\u5143 16KHz \u7684\u53d6 \u6a23\u983b\u7387\u5c07\u767c\u97f3\u4eba\u7684\u8a9e\u6599\u9304\u65bc\u55ae\u8072\u9053\u3002\u9304\u97f3\u60c5\u5883\u70ba\uff0c\u767c\u97f3\u4eba\u9762\u5c0d\u96fb\u8166\u81ea\u884c\u8f38\u5165\u8b6f\u6587\u3001\u64cd\u4f5c \u9304\u97f3\u53ca\u8fa8\u8b58\u904e\u7a0b\uff0c\u9304\u97f3\u8a08\u5283\u8ca0\u8cac\u4eba\u64cd\u4f5c\u4fee\u6539\u56de\u61c9\u7684\u90e8\u5206\u3002\u7e3d\u5171\u6536\u9304\u5230 144 \u500b\u5c0d\u8a71\u56de\u5408 (dialogue turn)\uff0c\u7e3d\u6578\u70ba 1,586 \u53e5\u7684\u8a9e\u6599\u3002\u9304\u97f3\u5b8c\u6210\u5f8c\uff0c\u4ee5\u4eba\u5de5\u65b9\u5f0f\u9032\u884c\u5c0d\u8a71\u8a9e\u6599\u76f8\u95dc\u8cc7 \u6599\u6a19\u8a08(\u5982 Dialogue \u7de8\u865f\u548c Turn \u7de8\u865f)\u3001\u5c0d\u8a71\u884c\u70ba(dialogue act, DA)\u3002 2.2 \u8a9e\u7fa9\u985e\u5225(Semantic Class) \u5176\u4e2d NB i \u548c NA i \u5206\u5225\u70ba\u5728 MP \u4e4b\u524d\u8207 MP \u4e4b\u5f8c\u7684 OP \u6578\u3002\u6839\u64da\u4e0a\u8ff0\u5b9a\u7fa9\uff0cPP \u70ba\u5305\u542b MP" |
| }, |
| "TABREF1": { |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>\u5728\u53e5\u578b\u898f\u5247\u7522\u751f\u90e8\u5206\uff0c\u9996\u5148\uff0c\u6211\u5011\u9700\u8981\u4e00\u500b\u8a9e\u7fa9\u5256\u6790\u5668\u4f86\u8655\u7406\u8a13\u7df4\u8a9e\u53e5\uff0c\u4e26\u5efa\u7acb\u5c0d\u61c9\u7684</td></tr><tr><td>\u8a9e\u7fa9\u6a39\u72c0\u7d50\u69cb\uff0c\u5f97\u5230\u5176\u53e5\u578b\u898f\u5247\uff0c\u672c\u7814\u7a76\u5229\u7528\u53f2\u4e39\u4f5b\u5927\u5b78\u6240\u7814\u7a76\u958b\u767c\u7684\u5256\u6790\u5668\u4f86\u9054\u6210\u6b64</td></tr><tr><td>\u76ee\u7684\u3002\u53f2\u4e39\u4f5b\u7684\u5256\u6790\u5668[14]</td></tr></table>", |
| "type_str": "table", |
| "text": "\u662f\u57fa\u65bc PCFG (Probabilistic Context Free Grammar) \u7684\u89c0\u5ff5\u6240 \u5efa\u7acb\u800c\u6210\u7684\u5256\u6790\u5668\u3002\u6240\u8b02\u7684 PCFG \u662f\u4e00\u7a2e\u96a8\u6a5f\u8a9e\u8a00\u6a21\u578b (Stochastic Language Models, SLM)\uff0c\u800c SLM \u7684\u4e3b\u8981\u76ee\u7684\u4e4b\u4e00\u662f\u6839\u64da\u8a13\u7df4\u8a9e\u6599\u7684\u7d71\u8a08\u8cc7\u6599\u4f86\u63d0\u4f9b\u8db3\u5920\u7684\u6a5f\u7387\u8cc7\u8a0a\u4ee5\u904b \u7528\u5728\u8a9e\u97f3\u8fa8\u8b58\u7684\u69cb\u53e5\u8655\u7406\u4e0a\uff0c\u4e0d\u50c5\u80fd\u6709\u6548\u63d0\u9ad8\u7684\u8fa8\u8b58\u6b63\u78ba\u7387\uff0c\u66f4\u53ef\u85c9\u7531\u641c\u5c0b\u8def\u5f91\u7684\u9650\u5236\uff0c \u7bc0\u7701\u8a08\u7b97\u6642\u9593\uff0c\u800c\u61c9\u7528\u5728\u6587\u53e5\u5256\u6790\u4e0a\u5247\u80fd\u63d0\u4f9b\u6b63\u78ba\u6027\u8f03\u9ad8\u7684\u53e5\u6cd5\u7d50\u679c\u3002\u95dc\u65bc\u53f2\u4e39\u4f5b\u5256\u6790 \u5668\u4e3b\u8981\u7684\u6838\u5fc3\u6982\u5ff5\u53ef\u4ee5\u53c3\u8003\u6587\u737b[20][21]\u3002\u5716 2 \u6d41\u7a0b\u4e2d\u7684\u53e5\u578b\u898f\u5247\u7522\u751f\uff0c\u5728\u5256\u6790\u524d\uff0c\u6211 \u5011\u5148\u5229\u7528\u4e8b\u5148\u5b9a\u7fa9\u597d\u7684\u8a9e\u7fa9\u985e\u5225\u5c07\u8a9e\u53e5\u4e2d\u7684\u8a5e\u5f59\u66ff\u63db\u6210\u8a9e\u7fa9\u985e\u5225\uff0c\u518d\u900f\u904e\u53f2\u4e39\u4f5b\u5256\u6790\u5668 \u5f97\u5230\u5256\u6790\u7d50\u679c\u3002\u66ff\u63db\u7684\u76ee\u7684\u662f\u964d\u4f4e\u53e5\u578b\u898f\u5247\u7684\u8907\u96dc\u5ea6\uff0c\u8b93\u76f8\u540c\u8a9e\u7fa9\u7684\u8a5e\u5f59\u5c6c\u65bc\u540c\u4e00\u689d\u898f \u5247\uff0c\u4f8b\u5982\uff1a \u300cNP \u2192 NN \u5b89\u5e73\u53e4\u5821\u300d\u548c\u300cNP \u2192 NN \u5104\u8f09\u91d1\u57ce\u300d\u7686\u5c6c\u65bc\u300cNP \u2192 NN \u5730\u9ede\u300d \u9019\u689d\u898f\u5247\u3002\u5716 3 \u7bc4\u4f8b\u662f\u8a9e\u53e5\u300c\u600e\u9ebc\u53bb\u5b89\u5e73\u53e4\u5821\u300d\u7d93\u904e\u8a9e\u7fa9\u66ff\u63db\u70ba\u300c\u7591\u554f\u8a5e \u8def\u7dda \u5730\u9ede\u300d \uff0c \u7d93\u7531\u5256\u6790\u5668\u53ef\u5f97\u5230\u4e00\u9846\u6587\u6cd5\u6a39\uff0c\u5176\u5305\u542b\u7684\u53e5\u578b\u898f\u5247\u5305\u62ec\uff1a" |
| }, |
| "TABREF2": { |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>Trans i q \u03a6</td><td colspan=\"8\">1, 1 2, 2 , q q \u03c6 \u03b4 \u03c6 \u03b4 =(</td><td>, , \uf04b</td><td colspan=\"4\">, Lq L \u03c6 \u03b4</td><td>)</td><td/><td>(14)</td></tr><tr><td colspan=\"17\">\u5176\u4e2d \u03c6 l,q \u7528\u5230 Rule l \uff0c\u5247\u5176\u503c\u70ba 1\uff0c\u53cd\u4e4b\u70ba 0\u3002\u53e6\u5916\uff0c\u6211\u5011\u9078\u64c7\u4e00\u500b\u7279\u6b8a\u51fd\u5f0f\u4f5c\u70ba\u6700\u5927\u5316\u7fa4\u4e4b\u5167\u76f8</td></tr><tr><td colspan=\"12\">\u4f3c\u5ea6\uff0c\u6b64\u7279\u6b8a\u51fd\u5f0f\u8868\u793a\u70ba\uff1a</td><td/><td/><td/><td/></tr><tr><td colspan=\"10\">1 ( , , , )* argmax 2 K G G G = \uf04b</td><td colspan=\"2\">1 = K k \uf0e5</td><td>\u03a6</td><td colspan=\"2\">Trans i q</td><td>, \uf0e5 Trans q \u03a6</td><td>j</td><td>k G \u2208</td><td>Similarity</td><td>(</td><td>\u03a6</td><td>, qq i Trans Trans \u03a6</td><td>j</td><td>)</td><td>(15)</td></tr><tr><td colspan=\"2\">Similarity</td><td>(</td><td>\u03a6</td><td>Trans q</td><td>i</td><td>,</td><td>\u03a6</td><td colspan=\"2\">Trans q</td><td>j</td><td>)</td><td>=</td><td colspan=\"4\">i Trans Trans q q \u03a6 \u03a6</td><td>i</td><td>\u22c5 \u22c5</td><td>Trans q j Trans q \u03a6 \u03a6</td><td>j</td></tr></table>", |
| "type_str": "table", |
| "text": "3.7 \u8a9e\u53e5\u5206\u7fa4(Sentence Clustering) \u4f7f\u7528\u8005\u7684 DA \u53ef\u80fd\u4ee5\u4e0d\u540c\u8a9e\u53e5\u8868\u9054\uff0c\u4e0d\u540c\u8a9e\u53e5\u610f\u5473\u8457\u4ed6\u5011\u53ef\u80fd\u860a\u542b\u8457\u4e0d\u540c\u7684\u53e5\u578b\u898f\u5247\uff0c \u56e0\u6b64\uff0c\u540c\u4e00\u500b DA \u4e0b\u53ef\u80fd\u5305\u542b\u8457\u591a\u7a2e\u53e5\u578b\u898f\u5247\u5c0e\u81f4\u548c\u5176\u4ed6\u610f\u5716\u4e4b\u9593\u9020\u6210\u6df7\u6dc6\u3002\u4f8b\u5982\uff1a\u5169 \u6bb5\u5c6c\u65bc\u540c\u6a23 DA 1 \u7684\u8a9e\u97f3\u5206\u5225\u5305\u542b\u8a9e\u53e5\u898f\u5247{1,2,3}\u548c{4,5,6}\uff0c\u53e6\u6709\u4e00\u6bb5\u5c6c\u65bc DA 2 \u7684\u8a9e\u97f3\u5305 \u542b\u8a9e\u97f3\u898f\u5247{3,4,7}\uff0c\u5247\u6b64\u8a9e\u97f3\u53ef\u80fd\u6703\u88ab\u8aa4\u5224\u70ba DA 1 \u3002\u56e0\u6b64\uff0c\u70ba\u4e86\u907f\u514d\u610f\u5716\u4e4b\u9593\u7684\u6df7\u6dc6\uff0c \u8a9e\u53e5\u9700\u8981\u5206\u7fa4\u3002\u5c0d\u65bc\u50b3\u7d71\u7684\u5206\u7fa4\u65b9\u6cd5\uff0c\u5982 K-means \u6f14\u7b97\u6cd5\uff0c\u5fc5\u9808\u8a08\u7b97\u5404\u8cc7\u6599\u9ede\u548c centroid \u4e4b\u9593\u7684\u8ddd\u96e2\u3002\u7136\u800c\uff0c\u53e5\u578b\u898f\u5247\u7684 centroid \u4e0d\u5177\u6709\u4efb\u4f55\u7269\u7406\u610f\u7fa9\u3002\u56e0\u6b64\u70ba\u9069\u61c9\u672c\u8ad6\u6587\u7684\u9700 \u6c42\uff0c\u6211\u5011\u5148\u5c07\u5c6c\u65bc\u7b2c q \u500b DA \u7684\u7b2c i \u500b\u8b6f\u6587\u6a94 Trans i \u8868\u793a\u70ba\uff1a \u7684\u5b9a\u7fa9\u7b49\u540c\u65bc\u4e0a\u8ff0\u53e5\u578b\u898f\u5247\u6b78\u7d0d\u77e9\u9663 \u03a6 L\u00d7Q \u4e2d\u7684 \u03c6 l,q \u3002\u800c \u03b4 l \u6307\u51fa\u82e5 Trans i \u4f7f" |
| }, |
| "TABREF3": { |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>\u56db\u3001 \u5c0d\u8a71\u7ba1\u7406\u6c7a\u7b56 \u5bdf\u6a5f\u7387\uff0c\u53ca\u5b9a\u7fa9\u734e\u52f5\u51fd\u5f0f\u70ba\uff1a \u6240\u8b02\u5c0d\u8a71\u56de\u5408\u6b21\u6578\u5373\u662f\u4e00\u6b21\u5c0d\u8a71\u9700\u8981\u4f86\u56de\u591a\u5c11\u6b21(turns)\u3002\u5c31\u5982\u5716 6(b)\u6240\u793a\uff0c\u6bcf\u4e00\u6b21\u5c0d\u8a71</td></tr><tr><td>\u5c0d\u8a71\u7ba1\u7406\u662f\u57fa\u65bc\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u7d50\u679c\u800c\u63a1\u53d6\u9069\u7576\u56de\u61c9\u4ee5\u8207\u4f7f\u7528\u8005\u9032\u884c\u4e92\u52d5\uff0c\u800c\u9069\u7576\u56de\u61c9\u4ef0 \u8cf4\u65bc\u5c0d\u8a71\u7b56\u7565\u7684\u898f\u5283\u3002\u5728\u672c\u7814\u7a76\u4e2d\uff0c\u6211\u5011\u63a1\u7528 POMDP \u4f5c\u70ba\u5c0d\u8a71\u7b56\u7565\u898f\u5283\u7684\u5de5\u5177\u3002 POMDP \u5c07\u7cfb\u7d71\u6240\u8655\u7684\u72c0\u614b\u8996\u70ba\u96b1\u542b\u8b8a\u6578(hidden variable)\uff0c\u56e0\u6b64\u5fc5\u9808\u4f7f\u7528\u4e00\u500b\u76f8\u4fe1\u51fd\u6578 (belief function)\u4f86\u5047\u8a2d\u7cfb\u7d71\u6240\u8655\u7684\u72c0\u614b\u4e26\u5b9a\u7fa9\u4e86\u4e94\u500b\u8b8a\u6578\u503c\u7d44(tuples) {S,A,R,T,O}\uff0c\u5206 \u5225\u4ee3\u8868\u72c0\u614b\u7d44\u6210\u7684\u96c6\u5408 S \u4e26\u7528\u4e00\u500b\u76f8\u4fe1\u51fd\u6578(belief function) b \u4f86\u63a7\u5236(maintain)\uff0c\u5728\u672c\u7814 10 , 10 , 5 , 100 , 100 , if if r if if if + \uf0ec \u7684\u53e5\u6578\u5927\u90fd\u5206\u5e03\u5728 3~15 \u53e5\u4e4b\u9593\u3002\u800c 3~5 \u53e5\u4ee3\u8868\u8457\u4f7f\u7528\u8005\u53ea\u4f7f\u7528\u4e00\u9805\u4efb\u52d9\u4fbf\u7d50\u675f\u7cfb\u7d71\uff0c \u7cfb\u7d71\u63a1\u53d6\u6b63\u78ba\u56de\u61c9 \uf0ef \u2212 = \u2212 \uf0ed + \uf0ef \uf0ee \u7cfb\u7d71\u7d50\u675f \uf0ef \u7cfb\u7d71\u63a1\u53d6\u6b63\u78ba\u56de\u61c9\u6b61\u8fce\u767c\u751f\u5728\u958b\u59cb\u4e4b\u5916 \uf0ef \u2212 \u91cd\u8907\u8a62\u554f\u554f\u984c (27) \uf0ef \uf0ef \u7cfb\u7d71\u63a1\u53d6\u932f\u8aa4\u56de\u61c9 \u5176\u4ed6\u5247\u8868\u793a\u4f7f\u7528\u8005\u53ef\u80fd\u540c\u6642\u67e5\u8a62\u4e86\u597d\u5e7e\u9805\u8cc7\u8a0a\u624d\u7d50\u675f\u7cfb\u7d71\u3002</td></tr><tr><td>\u7a76\u4e2d\uff0c\u5b9a\u7fa9\u70ba\u524d\u6587\u6240\u63d0\u7684 DA\uff1b\u56de\u61c9\u4f7f\u7528\u8005\u7684\u65b9\u5f0f\u6240\u7d44\u6210\u7684\u96c6\u5408 A\uff1b\u734e\u52f5\u51fd\u6578 R(s,a)=r\uff0c \u5728\u89c0\u5bdf\u6a5f\u7387\uff0c\u5047\u8a2d\u89c0\u5bdf\u70ba\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u5075\u6e2c\u5f8c\u7684\u5047\u8a2d\u7d50\u679c\uff0c\u53ef\u8868\u793a\u70ba\uff1a</td></tr><tr><td>3.9 \u5c0d\u8a71\u884c\u70ba\u578b\u614b\u5075\u6e2c \u5728\u5c0d\u8a71\u884c\u70ba DA \u5075\u6e2c\u4e2d\uff0c P(W i |DA C )\u9805\u56e0\u70ba\u4e00\u500b\u53e5\u5b50\u5305\u542b\u8a9e\u7fa9\u6210\u5206\u53ca\u8a9e\u6cd5\u6210\u5206\uff0c\u6240\u4ee5\u6211 \u5011\u5c0d\u8a9e\u97f3\u8fa8\u8b58\u7684\u7d50\u679c\u9032\u4e00\u6b65\u62c6\u89e3\u70ba\uff1a ( | ) ( | ) ( C | ) c W c W c P W DA P DA P S DA \u2248 Rule (21) ( )| ) T W c W c W c DA P D A DA = \u00d7 Rule Rule Rule \uf067 (22) \u800c SC W \u70ba\u5c07\u8fa8\u8b58\u7d50\u679c W \u8f49\u63db\u70ba\u8a9e\u7fa9\u985e\u5225\u7684\u51fd\u5f0f\u3002\u5728\u8a9e\u7fa9\u6210\u5206\u5206\u6578\u7684\u8a08\u7b97\uff0c\u6211\u5011\u7d93\u7531\u7d71 \u8a08\u6587\u5b57\u8a9e\u6599\uff0c\u5f97\u5230\u6bcf\u500b\u5c0d\u8a71\u884c\u70ba DA \u51fa\u73fe\u6bcf\u500b\u610f\u5716\u985e\u5225\u7684\u6a5f\u7387\uff0c\u6578\u5b78\u5f0f\u53ef\u5beb\u70ba\uff1a j (SC | ) (SC ) n n w W c w W P DA P \u2208 = \u220f (23) \u5176\u4e2d ) (SC j n w P \u8868\u793a\u8fa8\u8b58\u5b57\u4e32 W \u7684\u7b2c n \u500b\u5b57 w n \u5c6c\u65bc\u7b2c j \u500b\u8a9e\u7fa9\u7684\u6a5f\u7387\u3002\u9019\u53ef\u4ee5\u5f9e\u8a9e\u6599\u5eab \u4e2d\u96e2\u7dda\u9810\u5148\u4f30\u6e2c\u800c\u5f97\u3002 3.10 \u5c0d\u8a71\u884c\u70ba\u6b77\u53f2\u8a18\u9304 \u5c0d\u8a71\u884c\u70ba\u6b77\u53f2\u8a18\u9304\u65b9\u9762\u7684\u76ee\u7684\u662f\u70ba\u907f\u514d\u4f7f\u7528\u8005\u5728\u67e5\u8a62\u5176\u4e2d\u4e00\u9805\u4efb\u52d9\u6642\uff0c\u5728\u5c1a\u672a\u5b8c\u6210\u4efb\u52d9 \u537b\u56e0\u70ba ASR \u932f\u8aa4\u9020\u6210\u5c0d\u8a71\u884c\u70ba\u8aa4\u5224\u800c\u8f49\u70ba\u8a62\u554f\u4f7f\u7528\u8005\u5176\u4ed6\u4efb\u52d9\u7684\u5167\u5bb9\u3002\u5047\u8a2d DA t \u5b9a\u7fa9 \u70ba\u76ee\u524d\u8a9e\u97f3\u6240\u5f97\u5230\u7684 DA\uff0c\u5373 DA C \uff0c\u800c\u904e\u53bb\u6b77\u53f2\u7d00\u9304 DA t t-1 \u5b9a\u7fa9\u70ba DA H \uff0c\u5018\u82e5\u6211\u5011\u5047\u8a2d \u5c0d\u8a71\u884c\u70ba\u53ea\u8207\u524d\u4e00\u500b\u5c0d\u8a71\u884c\u70ba\u6709\u95dc\uff0c\u5247\u6578\u5b78\u8868\u793a\u6cd5\u53ef\u5b9a\u7fa9\u70ba\uff1a 1 1 1 2 1 1 ( | ) ( | , , , ) ( | ) t-t C H t t t t P DA DA DA DA P DA DA DA DA P DA DA \u2212 \u2212 = = = = (24) \u5728\u8a13\u7df4\u5c0d\u8a71\u7b56\u7565\u4e4b\u524d\uff0c\u6211\u5011\u5fc5\u9808\u5b9a\u7fa9\u7cfb\u7d71\u7684\u72c0\u614b\u3001\u89c0\u5bdf\u53ca\u734e\u52f5\u51fd\u5f0f\u3002\u5728\u672c\u8ad6\u6587\u4e2d\uff0c\u5c0d\u8a71 \u5206\u6790\uff0c\u5206\u5225\u70ba\u5c0d\u8a71\u4e4b\u610f\u5716\u5206\u4f48\u5716\u548c\u5c0d\u8a71\u7684\u9577\u5ea6\u5206\u6790\u3002\u5728\u8868 3 \u4e2d\uff0c\u6211\u5011\u5b9a\u7fa9\u4e86 38 \u7a2e DA \u610f\u5716\u3002\u96d6\u7121\u6e1b\u5c11\u5c0d\u8a71\u56de\u5408\u6b21\u6578\u4f46\u537b\u4f7f\u5f97\u4f7f\u7528\u8005\u89ba\u5f97\u7cfb\u7d71\u56de\u61c9\u66f4\u70ba\u4eba\u6027\u5316\uff0c\u6240\u4ee5 POMDP \uf04c \u8868 \u793a \u5728 \u72c0 \u614b s \u63a1 \u53d6 \u56de \u61c9 \u4f7f \u7528 \u7684 \u65b9 \u5f0f a \uff0c \u7cfb \u7d71 \u6240 \u5f97 \u5230 \u7684 \u734e \u52f5 \u70ba r \uff1b \u8f49 \u79fb \u6a5f \u7387 ) , | ( ) , | ( ' ' ' ' a DA DA P a s o P s o \u2261 (28) T(s,a,s\u2032)=P(s t+1 =s\u2032|s t ,a t )\u70ba\u7cfb\u7d71\u5728\u6642\u9593\u9ede t\uff0c\u5728\u72c0\u614b s \u63a1\u53d6\u4f7f\u7528\u7684\u65b9\u5f0f a\uff0c\u800c\u5728\u6642\u9593\u9ede t+1\uff0c \u72c0\u614b\u5c07\u6703\u8b8a\u6210 s\u2032\u7684\u6a5f\u7387\uff1b\u89c0\u5bdf(observation)\u6240\u7d44\u6210\u7684\u96c6\u5408 O\uff0c\u63cf\u8ff0\u8457 POMDP \u80fd\u63a5\u6536\u7684 \u8a0a\u606f\u800c\u89c0\u5bdf\u6a5f\u7387 P(o\u2032|s\u2032,a)\u8868\u793a\u7cfb\u7d71\u5728\u6642\u9593\u9ede t \u63a1\u53d6\u4f7f\u7528\u7684\u65b9\u5f0f a\uff0c\u53ca\u5728\u6642\u9593\u9ede t+1 \u7cfb\u7d71 \u6240\u8655\u7684\u72c0\u614b s\u2032\uff0c\u6240\u89c0\u5bdf\u5230\u7684\u89c0\u5bdf\u7684\u6a5f\u7387\u3002\u7d9c\u5408\u4ee5\u4e0a\u8b8a\u6578\uff0cPOMDP \u61c9\u7528\u65bc\u672c\u7814\u7a76\u7684\u6982\u5ff5 \u5716\u5982\u5716 4 \u6240\u793a\u3002 \u5716 4\uff1a\u99ac\u53ef\u592b\u6c7a\u7b56\u7a0b\u5e8f\u6982\u5ff5\u5716\u3002 \u5728\u76f8\u4fe1\u51fd\u6578\u7684\u66f4\u65b0\u90e8\u4efd\uff0c\u6211\u5011\u5f15\u7528\u6587\u737b[11]\u6240\u63a8\u5c0e\u7684\u516c\u5f0f\u3002 \uf0e5 \u2208 \u22c5 = = S s s b a s s P a s o P k b a o s P s b ) ( ) , | ' ( ) , | ( ) , , | ( ) ' ( ' ' ' ' ' (25) \u5176\u4e2d ) ( ' ' s b \u70ba\u66f4\u65b0\u7684\u76f8\u4fe1\u51fd\u6578\uff0c ) , | ( ' ' a s o P \u70ba\u89c0\u5bdf\u6a5f\u7387\uff0c ) , | ( ' a s s P \u70ba\u8f49\u79fb\u6a5f\u7387\uff0c ) (s b \u70ba \u76f8\u4fe1\u51fd\u6578\u3002\u90e8\u5206\u89c0\u5bdf\u99ac\u53ef\u592b\u6c7a\u7b56\u7a0b\u5e8f\u7684\u6700\u4f73\u503c\u51fd\u6578(Optimal value function)\u70ba\uff1a ] )) ( ( ) ( ) , | ( ) , | ( ) ( ) , ( [ max ) ( ' ' , , ' ' ' ' ' * \uf0e5 \uf0e5 \u2208 \u2208 + = S s s s o A a s b V s b a s s p a s o p s b a s r b V \u03b3 (26) 4.2 \u5c0d\u8a71\u7b56\u7565\u5b78\u7fd2 \u500b\u5c0d\u8a71\u56de\u5408\uff0c\u7e3d\u6578\u70ba 1586 \u53e5\uff0c\u70ba\u4e86\u4e86\u89e3\u5c0d\u8a71\u7528\u53e5\u5206\u4f48\u60c5\u5f62\uff0c\u672c\u8ad6\u6587\u91dd\u5c0d\u8a9e\u6599\u505a\u4e86\u5169\u7a2e \u7528 POMDP \u5247\u53ef\u80fd\u7522\u751f\u8f03\u70ba\u6b63\u78ba\u7684\u56de\u61c9\uff0c\u4f8b\u5982\u4e0a\u8ff0\u4f8b\u5b50\u4e2d\uff0c\u7cfb\u7d71\u56de\u61c9\u5c07\u8b8a\u70ba\u8a62\u554f\u4f7f\u7528\u8005 \u5716 5\uff1a\u5c0d\u8a71\u7b56\u7565\u8a13\u7df4\u6d41\u7a0b\u5716\u3002 \u5047\u8a2d\u89c0\u5bdf\u8207\u4e0a\u4e00\u6b21\u7cfb\u7d71\u56de\u61c9\u7121\u95dc\uff0c\u6240\u4ee5\u89c0\u5bdf\u6a5f\u7387\u70ba\uff1a ' ' ' ' ( | , )(1 ) ( | , ) ( | ) (1 ( | U, )) DA 1 C C C H e r r o s o s err C H u P DA S DA p P DA DA a P DA DA p P DA DA \u2212 \uf0ec \uf0ef \u2245 = \uf0ed \u2212 \u22c5 \uf0ef \u2212 \uf0ee (29) \u5176\u4e2d P errc \u900f\u904e\u8a13\u7df4\u8a9e\u6599\u6c42\u5f97\u6bcf\u500b\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u7684\u6b63\u78ba\u7387\u3002\u6b64\u89c0\u5bdf\u6a5f\u7387\u5305\u542b\u8457\u5c0d\u8a71\u884c\u70ba \u578b\u614b\u5075\u6e2c\u5206\u6578 P(DA C |U, DA H )\u53ca\u6b64\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u5075\u6e2c\u7d50\u679c\u7684\u53ef\u4fe1\u5ea6(1-P errc )\u3002\u6700\u5f8c\u6211\u5011\u7d93 \u7531 POMDP \u8edf\u9ad4[22]\u8a13\u7df4\u6211\u5011\u7684\u5c0d\u8a71\u7b56\u7565\u3002 \u4e94\u3001 \u5be6\u9a57 5.1 \u81ea\u52d5\u8a9e\u97f3\u8fa8\u8b58\u5668(ASR)\u7684\u5efa\u7acb ASR \u7684\u57fa\u672c\u5efa\u7acb\u6b65\u9a5f\uff0c\u5305\u542b\u4e86\u8072\u5b78\u7279\u5fb5\u53c3\u6578\u7684\u8403\u53d6(feature extraction)\u3001\u8072\u5b78\u6a21\u578b\u8a13\u7df4 (acoustic model, AM)\u548c\u8a9e\u8a00\u6a21\u578b(language model)\u7684\u8a13\u7df4\u3002\u6211\u5011\u63a1\u7528\u528d\u6a4b\u5927\u5b78\u6240\u958b\u767c\u7684\u5de5 \u5177 HMM Tool Kit(HTK)\u4f86\u5efa\u7acb\u672c\u7814\u7a76\u7684 ASR\u3002\u70ba\u4e86\u5efa\u7acb\u4e00\u500b\u8f03\u70ba\u53ef\u9760\u7684 ASR\uff0c\u6211\u5011\u5148 \u5716 6\uff1a\u5c0d\u8a71\u610f\u5716\u5206\u4f48\u548c\u5c0d\u8a71\u56de\u5408\u6b21\u6578\u5206\u4f48\u3002 \u5716 7\uff1a\u5404\u65b9\u6cd5\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u6e96\u78ba\u7387\u6bd4\u8f03\uff1a(1)\u516d\u3001\u7d50\u8ad6\u8207\u672a\u4f86\u5c55\u671b \u672c\u8ad6\u6587\u63d0\u51fa\u4e86\u5229\u7528\u90e8\u4efd\u6a23\u672c\u53e5\u8207\u53e5\u578b\u898f\u5247\u5efa\u69cb\u51fa\u6f5b\u5728\u610f\u5716\u77e9\u9663\uff0c\u85c9\u6b64\u5224\u65b7\u5c0d\u8a71\u884c\u70ba\u578b\u614b\uff0c \u4e26\u6709\u6548\u6539\u5584\u56e0\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u9020\u6210\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u932f\u8aa4\u7684\u554f\u984c\u3002\u6b64\u5916\u672c\u8ad6\u6587\u4e5f\u52a0\u5165\u5c0d\u8a71\u6b77\u53f2 \u7684\u6982\u5ff5\uff0c\u8003\u616e\u5c0d\u8a71\u8a9e\u7fa9\u8108\u7d61\u4f86\u5e6b\u52a9\u5c0d\u8a71\u7406\u89e3\u3002\u70ba\u4e86\u589e\u52a0\u4eba\u6a5f\u4e4b\u9593\u7684\u4e92\u52d5\uff0c\u5728\u7cfb\u7d71\u6c7a\u7b56\u7ba1 \u7406\u65b9\u9762\u4e5f\u904b\u7528 POMDP \u4ee5\u6c42\u53d6\u6700\u4f73\u7b56\u7565\uff0c\u4f7f\u5f97\u7cfb\u7d71\u7522\u751f\u6700\u4f73\u56de\u61c9\uff0c\u6e1b\u5c11\u7cfb\u7d71\u8207\u4f7f\u7528\u8005\u4e4b \u9593\u7121\u6cd5\u5b8c\u6210\u4efb\u52d9\u7684\u60c5\u6cc1\u3002\u900f\u904e\u5be6\u9a57\u7684\u8b49\u660e\uff0c\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u7684\u65b9\u6cd5\u5728\u6f5b\u5728\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u4e0a \u5e73\u5747\u6e96\u78ba\u7387\u70ba 81.9%\uff0c\u76f8\u8f03\u65bc\u55ae\u7d14\u4f7f\u7528\u8868\u55ae\u586b\u683c\u65b9\u5f0f\u7684\u610f\u5716\u5075\u6e2c\u5e73\u5747\u6e96\u78ba\u7387 48.1%\uff0c\u4f7f \u7528\u672c\u65b9\u6cd5\u53ef\u63d0\u5347\u4e86 33.8%\u4e4b\u6e96\u78ba\u7387\uff0c\u672c\u8ad6\u6587\u6240\u63d0\u51fa\u7684\u65b9\u6cd5\u7684\u78ba\u662f\u6709\u6548\u7684\u65b9\u6cd5\u3002\u96d6\u7136\u672c\u8ad6 5.3 DA \u6709\u975e\u5e38\u986f\u8457\u7684\u6539\u5584\uff0c \u9019\u985e\u7684\u53e5\u5b50\u5982\u300c\u600e\u9ebc\u5230\u5b89\u5e73\u53e4\u5821\u300d\u5bb9\u6613\u5728\u8a55\u4f30\u65b9\u6cd5(1)\u548c(2)\u88ab\u5224\u5225\u70ba\u5176\u4ed6\u985e\u4f3c\u7684\u5c0d\u8a71\u884c \u70ba\u578b\u614b\uff0c\u5982\u300c\u706b\u8eca\u7121\u51fa\u767c\u5730\u6709\u76ee\u7684\u5730\u300dDA\u3002\u6700\u5f8c\uff0c\u8a55\u4f30\u65b9\u6cd5(4)\uff0c\u9019\u500b\u7d44\u5408\u65b9\u5f0f\u5373\u70ba\u8ad6 \u578b\u662f\u53ef\u884c\u7684\u3002\u53e6\u5916\uff0c\u67e5\u8a62\u5730\u9ede(2)\u3001\u67e5\u8a62\u8eca\u7ad9(3)\u548c\u7cfb\u7d71\u6b61\u8fce(36)\u56e0\u70ba\u554f\u984c\u672c\u8eab\u7c21\u55ae\uff0c\u4ee5\u81f4 \u65bc\u56db\u7a2e\u8a55\u4f30\u65b9\u6cd5\u7686\u6709\u76f8\u540c\u7684\u6548\u80fd\u3002\u5728\u67e5\u8a62\u4ea4\u901a\u65b9\u5f0f\u7684\u706b\u8eca\u548c\u9ad8\u9435\u90e8\u5206\uff0c\u56e0\u70ba\u672c\u8eab\u7684\u95dc\u9375 \u5b57\u8a5e\u91cd\u758a\u6027\u592a\u9ad8\uff0c\u4ee5\u81f4\u65bc\u6c92\u6709\u660e\u986f\u7684\u6539\u5584\u8868 4 \u70ba\u8a55\u4f30\u7531 POMDP \u8a13\u7df4\u800c\u5f97\u7684\u5c0d\u8a71\u7b56\u7565 \u6a21\u5f0f\u5c0d\u65bc\u6211\u5011\u6240\u8490\u96c6\u8a9e\u6599\u7684\u6548\u7528\uff0c\u6211\u5011\u53ef\u4ee5\u767c\u73fe POMDP \u7684\u78ba\u80fd\u964d\u4f4e\u5c0d\u8a71\u56de\u5408\u6b21\u6578\uff0c\u4f46 \u7531\u65bc\u6536\u9304\u7684\u8a9e\u6599\u8fa8\u8b58\u7d50\u679c\u9084\u7b97\u6b63\u78ba\u4e14\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u5224\u65b7\u5927\u90e8\u4efd\u4e5f\u662f\u6b63\u78ba\uff0c\u6240\u4ee5\u964d\u4f4e\u7684\u5c0d \u8a71\u56de\u5408\u6b21\u6578\u6c92\u6709\u986f\u8457\u7684\u63d0\u5347\u3002\u5728\u53e6\u4e00\u65b9\u9762\uff0c\u672c\u4f86\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u5075\u6e2c\u932f\u8aa4\u7684\u53e5\u5b50\uff0c\u82e5\u662f\u7528 \u4eba\u5de5\u8a02\u5b9a\u7684\u56de\u61c9\u53ef\u80fd\u6703\u7522\u751f\u5947\u602a\u7684\u56de\u61c9\uff0c\u4f8b\u5982\u4f7f\u7528\u8005\u8a62\u554f\u5730\u5740\u800c\u7cfb\u7d71\u537b\u56de\u7b54\u7968\u50f9\uff0c\u82e5\u4f7f \u6587\u6240\u63d0\u51fa\u7684\u65b9\u6cd5\u53ef\u9054\u5230\u4e0d\u932f\u7684\u6210\u6548\uff0c\u4f46\u4ecd\u6709\u4e0d\u5c11\u5730\u65b9\u6709\u5f85\u6539\u9032\uff0c\u4ee5\u4e0b\u6211\u5011\u5c07\u9010\u4e00\u8aaa\u660e\u53ef \u6539\u9032\u7684\u5730\u65b9\uff1a(1)\u5982\u4f55\u81ea\u52d5\u627e\u5c0b\u61c9\u7528\u9818\u57df\u7684\u5c0d\u8a71\u884c\u70ba\u578b\u614b\u4ee5\u6e1b\u5c11\u4eba\u70ba\u4ecb\u5165\uff0c\u70ba\u7406\u89e3\u7cfb\u7d71 \u53e6\u4e00\u500b\u7814\u7a76\u8b70\u984c\u3002(2)\u5728\u672c\u8ad6\u6587\u4e2d\u6240\u4f7f\u7528\u7684 POMDP\uff0c\u53ef\u4ee5\u9032\u4e00\u6b65\u5c07\u72c0\u614b\u5047\u8a2d\u70ba\u8a31\u591a\u4e0d\u540c \u503c\u57df\u7684\u96c6\u5408\uff0c\u4f7f\u7cfb\u7d71\u80fd\u66f4\u7d30\u7dfb\u5730\u8207\u4f7f\u7528\u8005\u4e92\u52d5\u3002(3)\u53ef\u4ee5\u5b9a\u7fa9\u734e\u52f5\u51fd\u6578\uff0c\u6839\u64da\u4e0d\u540c\u7684\u586b \u503c\u72c0\u6cc1\u4f5c\u4e0d\u540c\u7684\u734e\u61f2\u3002 \u63a1\u7528\u9ea5\u514b\u98a8\u8a9e\u6599\u5eab TCC300 \u9032\u884c\u7a2e\u5b50(seed) \u5c07\u6536\u96c6\u800c\u4f86\u7684\u8a9e\u6599\uff0c\u7d93\u904e\u6574\u7406\u6311\u9078\u51fa\u9069\u5408\u7684\u8a9e\u6599\u4f5c\u70ba\u8a13\u7df4\u8a9e\u6599\u4e4b\u7528\uff0c\u8a9e\u6599\u7e3d\u5171\u6709 144 \u6587\u4e2d LDAM \u6a21\u578b\u7684\u8a13\u7df4\u65b9\u5f0f\uff0c\u7d93\u5be6\u9a57\u8b49\u660e\uff0c\u5728\u6240\u6709\u7684\u65b9\u6cd5\u4e2d\uff0c\u6211\u5011\u6240\u5efa\u69cb\u7684 LDAM \u6a21 \u53c3\u8003\u6587\u737b</td></tr><tr><td>\u7b56\u7565\u5b78\u7fd2\u7684\u72c0\u614b\u548c\u72c0\u614b\u6240\u5c0d\u61c9\u7684\u56de\u61c9\u884c\u70ba\u70ba\u7b2c\u4e8c\u7bc0\u6240\u63d0\u5230\u7684 slot \u548c action \u7684\u90e8\u4efd\u3002\u800c\u5c0d \u800c\u5716 6(a)\u5448\u73fe\u51fa\u8a9e\u6599\u88e1\u5404\u7a2e DA \u5206\u4f48\u7684\u60c5\u5f62\u3002\u7531\u5206\u4f48\u5716\u53ef\u77e5\uff0c\u4f7f\u7528\u8005\u6703\u6839\u64da\u4ed6\u7684\u9700\u6c42\u67e5 \u78ba\u5be6\u6709\u5b83\u7684\u6548\u80fd\u3002</td></tr><tr><td>\u8a71\u7b56\u7565\u8a13\u7df4\u6d41\u7a0b\u5716\u5982\u5716 5 \u6240\u793a\u3002\u9996\u5148\uff0c\u5f9e\u6211\u5011\u6536\u9304\u7684\u8a9e\u6599\u4e2d\uff0c\u7531\u6587\u5b57\u7684\u6bcf\u500b\u5c0d\u8a71 \u8a62\u4ed6\u6240\u60f3\u8981\u7684\u8cc7\u8a0a\uff0c\u6240\u4ee5\u6bcf\u500b\u610f\u5716\u51fa\u73fe\u983b\u7387\u4e0d\u540c\uff0c\u800c\u300c\u7d50\u675f\u300d\u51fa\u73fe\u983b\u7387\u9060\u9ad8\u65bc\u5176\u4ed6\u610f\u5716</td></tr><tr><td>(dialogue)\u5f97\u5230\u8f49\u79fb(transition)\u6a5f\u7387\uff0c\u53ca\u5f9e\u6587\u5b57\u76f8\u5c0d\u61c9\u7684\u97f3\u6a94\u7d93\u7531\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u5f8c\u5f97\u5230\u89c0 \u8aaa\u660e\u4e86\u5728\u5c0d\u8a71\u7d50\u675f\u6642\u4f7f\u7528\u8005\u7fd2\u6163\u6027\u8aaa\u544a\u5225\u7528\u8a9e\u3002\u8aaa\u660e\u4e86\u8a9e\u6599\u88e1\u5c0d\u8a71\u56de\u5408\u6b21\u6578\u5206\u4f48\u7684\u60c5\u5f62\uff0c</td></tr></table>", |
| "type_str": "table", |
| "text": "\u5176\u4e2d Rule W \u70ba\u8a18\u9304\u8fa8\u8b58\u7d50\u679c W \u6240\u4f7f\u7528\u7684\u53e5\u578b\u898f\u5247\u3002\u5247 Rule W \u5c0d\u6f5b\u5728\u610f\u5716\u77e9\u9663\u4e2d\u7b2c c \u500b \u985e\u5225\u7684\u76f8\u4f3c\u5ea6\uff0c\u6211\u5011\u63a1\u7528 Cosine Measure \u4f86\u8a08\u7b97\uff0c\u5b9a\u7fa9\u70ba\uff1a AM \u8a13\u7df4\uff0c\u5305\u62ec 115 \u500b\u53f3\u76f8\u95dc(right-context dependent) initial \u6b21\u97f3\u7bc0\u548c 38 \u500b\u7368\u7acb(right-context independent) final \u6b21\u97f3\u7bc0\uff0c\u5206\u5225\u4f7f\u7528 3 \u500b\u548c 5 \u500b\u72c0\u614b(state)\uff0c\u6bcf\u500b\u72c0\u614b\u6700\u591a 32 \u500b mixtures\u3002\uff0c\u518d\u4ee5\u6211\u5011\u6240\u9304\u88fd\u800c\u4f86\u7684\u65c5\u904a\u76f8\u95dc \u8a9e\u6599\u9032\u884c\u8abf\u9069(adaptation)\u3002\u7279\u5fb5\u53c3\u6578\u70ba 39 \u70ba\u5ea6\u7684\u6885\u723e\u5012\u983b\u8b5c\u7279\u5fb5\u53c3\u6578(MFCC)\uff0c\u5176\u4e2d\u9810 \u5f37\u8abf\u4fc2\u6578\u70ba 0.97\uff0c\u5176\u9918\u76f8\u95dc\u53c3\u6578\u8a2d\u5b9a\u53ef\u53c3\u8003 HTK Book \u8aaa\u660e\u3002\u5728\u8a9e\u8a00\u6a21\u578b\u90e8\u5206\uff0c\u6211\u5011\u63a1 \u7528 TCC300 \u8a9e\u6599\u5eab\u4f86\u5efa\u7acb\u8a9e\u8a00\u6a21\u578b\u7684\u90e8\u5206\uff0c\u4e26\u5617\u8a66\u9032\u884c\u8abf\u9069\u3002\u7136\u800c\uff0c\u7d93\u7531\u5be6\u9a57\u767c\u73fe\uff0c\u8a9e \u8a00\u6a21\u578b\u5728\u672c\u7cfb\u7d71\u4f3c\u4e4e\u4f5c\u7528\u6027\u4e0d\u9ad8\uff0c\u9019\u662f\u56e0\u70ba\u6211\u5011\u6240\u6536\u9304\u7684\u65c5\u904a\u76f8\u95dc\u8a9e\u6599\u5167\u5bb9\u5c0d\u65bc TCC300 \u800c\u8a00\u5c6c\u65bc\u5fae\u91cf\uff0c\u56e0\u6b64\u8a31\u591a\u65c5\u904a\u666f\u9ede\u76f8\u95dc\u5b57\u8a5e\u7121\u8ad6\u5728 uni-gram \u548c bi-gram \u90fd\u53ea\u662f \u6975\u5c0f\u503c\u3002\u5c31\u8a9e\u97f3\u8fa8\u8b58\u5668\u800c\u8ad6\uff0c\u6211\u5011\u6240\u5efa\u7acb\u7684\u8a9e\u97f3\u8fa8\u8b58\u5668\u5c0d\u65bc\u6240\u8490\u96c6\u7684\u65c5\u904a\u76f8\u95dc\u8a9e\u6599\u5eab\u6709 \u9ad8\u9054 84.33%\u7684\u6b63\u78ba\u7387\u3002\u7d93\u8a9e\u8005\u8abf\u9069\u5f8c\uff0c\u66f4\u53ef\u9054 93.12%\u7684\u6b63\u78ba\u7387\u3002 5.2 \u5c0d\u8a71\u8a9e\u6599\u4e4b\u5206\u6790 \u7cfb\u7d71\u8a55\u4f30\u5206\u6790 \u7cfb\u7d71\u8a55\u4f30\u65b9\u9762\uff0c\u6211\u5011\u628a\u8a55\u4f30\u65b9\u5f0f\u5206\u70ba\u4e8c\u7a2e\uff0c\u7b2c\u4e00\u7a2e\u662f\u5c0d\u5404\u985e DA \u505a\u8a55\u4f30\uff0c\u89c0\u5bdf DA \u5075\u6e2c \u6a21\u7d44\u5c0d\u65bc\u6bcf\u985e DA \u7684\u904b\u4f5c\u6548\u80fd\u3002\u7b2c\u4e8c\u7a2e\u662f\u5c0d\u8a71\u56de\u5408(turn)\u6b21\u6578\u7684\u8a55\u4f30\uff0c\u6b64\u8a55\u4f30\u76ee\u7684\u5728\u65bc \u4e86\u89e3 POMDP \u5c0d\u65bc\u6574\u500b\u5c0d\u8a71\u6d41\u7a0b\u5176\u5f71\u97ff\u7d50\u679c\u3002\u8a55\u4f30\u8a9e\u53e5\u7e3d\u6578\u70ba 912 \u53e5\u3002\u8a55\u4f30 DA \u5075\u6e2c\u6a21 \u7d44\u6642\uff0c\u5206\u5225\u8003\u616e(1)\u50c5\u4f7f\u7528\u8a9e\u7fa9\u8868\u683c(semantic slot, SS)\uff0c\u4f9d\u64da\u4f7f\u7528\u8005\u586b\u5165\u7684 semantic slot \u5224\u65b7\u4f7f\u7528\u8005\u7684\u610f\u5716\uff0c(2)\u4f7f\u7528 SS \u548c\u53f2\u4e39\u4f5b\u5256\u6790\u5668(Stanford Parser, SP)\u5efa\u7acb\u6f5b\u5728\u610f\u5716\u77e9\u9663 LDAM \u7684\u5075\u6e2c\u65b9\u5f0f\uff0c(3)\u4f7f\u7528 SS\u3001SP \u548c\u90e8\u5206\u6a23\u672c\u6a39(PPT)\u5efa\u7acb LDAM \u7684\u5075\u6e2c\u65b9\u5f0f\u548c(4) \u4f7f\u7528 SS\u3001SP\u3001PPT \u548c\u53e5\u578b\u5206\u985e(sentence clustering, SC)\u5efa\u7acb LDAM \u7684\u5075\u6e2c\u65b9\u5f0f\u56db\u7a2e\u5c0d\u8a71 \u884c\u70ba\u6a21\u7d44\u7684\u8a55\u4f30\u65b9\u6cd5\u3002\u5404\u65b9\u6cd5\u7684\u5c0d\u8a71\u884c\u70ba\u5075\u6e2c\u6e96\u78ba\u7387\u5982\u5716 7\uff0c\u5e73\u5747\u6b63\u78ba\u7387\u5206\u5225\u70ba 49.6%\u3001 76.2%\u300181.6%\u548c 82.9%\u3002\u5716 7 \u6a19\u793a\u70ba DA \u7684\u6b04\u4f4d\u70ba 37 \u7a2e DA\uff0c\u7f3a\u5c11\u300c\u7121\u610f\u5716\u300dDA \u662f\u56e0 \u70ba\u7121\u6cd5\u6536\u96c6\u7121\u610f\u5716 DA \u7684\u53e5\u5b50\uff0c\u6545\u7121\u6cd5\u8a55\u4f30\u5176\u6e96\u78ba\u7387\uff0c\u7121\u610f\u5716 DA \u662f\u7528\u4f86\u7576\u7cfb\u7d71\u7121\u6cd5\u5224 \u65b7\u4f7f\u7528\u8005\u7684\u610f\u5716\u6642\u6240\u505a\u51fa\u7684\u56de\u61c9\u3002\u8a55\u4f30\u65b9\u6cd5(1)\u7684\u7d50\u679c\uff0c\u96d6\u7136\u5728\u67d0\u4e9b DA \u4e0a\u80fd\u6709\u53ef\u63a5\u53d7 \u7684\u8868\u73fe\uff0c\u4f46\u6211\u5011\u53ef\u4ee5\u767c\u73fe\u5c6c\u65bc\u4efb\u52d9 3 \u7684\u8868\u73fe\u51fa\u73fe\u843d\u5dee\uff0c\u56e0\u70ba DA \u4e2d\u6709\u5e7e\u7a2e\u5f7c\u6b64\u6703\u4e92\u76f8\u6df7 \u6dc6\uff0c\u9020\u6210\u5f88\u591a\u610f\u5716\u7684\u6e96\u78ba\u7387\u662f 0%\uff0c\u4f8b\u5982\u8a62\u554f\u4ea4\u901a\u65b9\u5f0f\u7684 DA\u3002\u4f7f\u7528\u8a55\u4f30\u65b9\u6cd5(2)\u7684\u7d50\u679c\uff0c \u76f8\u8f03\u65bc(1)\u6709\u660e\u986f\u7684\u6539\u5584\uff0c\u4f46\u4f9d\u7136\u7121\u6cd5\u89e3\u6c7a\u56e0\u70ba\u8a9e\u97f3\u8fa8\u8b58\u932f\u8aa4\u800c\u9020\u6210 DA \u5075\u6e2c\u7684\u554f\u984c\u3002 \u4f7f\u7528\u8a55\u4f30\u65b9\u6cd5(3)\u7684\u7d50\u679c\uff0c\u5728\u6e2c\u8a66\u6642\uff0c\u8a9e\u53e5\u7684\u8fa8\u8b58\u7d50\u679c\u5fc5\u9808\u7d93\u904e PPT \u7522\u751f\u6a23\u672c\u5019\u9078\u64da\u4f86 \u9032\u884c\u610f\u5716\u5075\u6e2c\uff0c\u6211\u5011\u53ef\u4ee5\u770b\u5230\u6a19\u793a\u70ba\u300c\u7121\u51fa\u767c\u5730\u6709\u76ee\u7684\u5730\u300d\u7684 \u50c5\u4f7f\u7528 SS \u7684\u5075\u6e2c\uff0c(2)\u4f7f\u7528 SS \u548c SP \u5075\u6e2c\uff0c (3)\u4f7f\u7528 SS\u3001SP \u548c PPT \u7684\u5075\u6e2c\uff0c(4)\u4f7f\u7528 SS\u3001SP\u3001PPT \u548c SC \u7684\u5075\u6e2c\u3002 \u8868 4 \uff1a\u5c0d\u8a71\u56de\u5408\u6b21\u6578\u4e4b\u6bd4\u8f03\u3002\u4f7f\u7528 POMDP \u8a13\u7df4\u800c\u5f97\u7684\u7b56\u7565\u964d\u4f4e\u4e86\u5c0d\u8a71\u56de\u5408\u6578\u3002" |
| } |
| } |
| } |
| } |
