| { |
| "paper_id": "O10-1003", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:06:40.818172Z" |
| }, |
| "title": "Exploiting Discriminative Language Models for Reranking Speech Recognition Hypotheses", |
| "authors": [ |
| { |
| "first": "Chia-Wen", |
| "middle": [], |
| "last": "\uf9c7\u5bb6\u598f", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u570b\uf9f7\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Shih-Hsiang", |
| "middle": [], |
| "last": "\uf9f4\u58eb\u7fd4", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u570b\uf9f7\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
| "middle": [], |
| "last": "Lin", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u9673\u67cf\u7433", |
| "middle": [], |
| "last": "Berlin", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "berlin@ntnu.edu.tw" |
| }, |
| { |
| "first": "Chen", |
| "middle": [], |
| "last": "\u570b\uf9f7\u81fa\u7063\u5e2b\u7bc4\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "", |
| "pdf_parse": { |
| "paper_id": "O10-1003", |
| "_pdf_hash": "", |
| "abstract": [], |
| "body_text": [ |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "x M \u689d\u6700 \u4f73\u8fa8\uf9fc\u5019\u9078\u8a5e\u5e8f\uf99c\u96c6\u5408\u70ba ) { } ( j i i , W x GEN = \uff0c\u5176\u4e2d j \u4ecb\u65bc 1 \u5230 M \u4e4b\u9593\u3002 (b) \u5c07\u8a13\uf996\u8a9e\uf9be\u8868\u793a\u6210 { } R i i W x , L L R i ( ) i x GEN \u7684\u96c6\u5408\uff0ci \u4ecb\u65bc 1 \u5230 \u4e4b\u9593\uff0c \u70ba\u8a13\uf996\u8a9e\uf9be\uf906 \uf969\uff1bW \u70ba \u4e2d\u6700\u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f\uf99c\uff1b\u800c\u6e2c\u8a66\u8a9e\uf9be\u5247\u8868\u793a\u6210{ } k y \u7684\u96c6\u5408\uff0ck \u4ecb \u65bc 1 \u5230 K \u4e4b\u9593\uff0c K \u70ba\u6e2c\u8a66\u8a9e\uf9be\u4e4b\uf906\uf969\u3002 ( ) (c) \u5c0d\u6bcf\u689d\u5019\u9078\u8a5e\u5e8f\uf99cW \u5b9a\u7fa9 j i, 1 + D \u500b\u7279\u5fb5 j i d W f , d D j i, \uff0c \u4ecb\u65bc 0 \u5230 \u4e4b\u9593\uff0c \u6bcf\u500b\u8a9e\u8a00\u7279\u5fb5\u7686\u70ba\u4e00\u500b\u5c07\u5019\u9078\u8a5e\u5e8f\uf99cW \u5c0d\u61c9\u5230\u5be6\uf969\u503c\u4e4b\u51fd\uf969\u3002 ( ) j i W f , 0 j i, N \u70ba\u57fa\u790e\u8a9e \u8a00\u7279\u5fb5\uff0c\u672c\uf941\u6587\u5b9a\u7fa9\u70ba\u4e09\uf99a\u8a5e\u8a9e\u8a00\u6a21\u578b\u8207\u8072\u5b78\u6a21\u578b\u4e58\u7a4d\u4e4b\u5c0d\uf969\u503c\uff0c\u800c\u5176\u9918\u7684\u8a9e\u8a00 \u7279\u5fb5\u5247\u53ef\u5b9a\u7fa9\u70ba\u5019\u9078\u8a5e\u5e8f\uf99c W \u4e2d\uff0c\u5404 \uf99a\u8a5e\u51fa\u73fe\u7684\u6b21\uf969\uff1b\u672c\uf941\u6587\u4f7f\u7528\u55ae\uf99a\u8a5e (Word Unigram)\u8207\u4e8c\uf99a\u8a5e(Word Bigram)\u51fa\u73fe\u7684\u6b21\uf969\u505a\u70ba\u5176\u9918\u7684\u8a9e\u8a00\u7279\u5fb5\u3002 (d) \u6bcf\u4e00\u8a9e\u8a00\u7279\u5fb5\u5b9a\u7fa9\u90fd\u6709\u5176\u5c0d\u61c9\u7684\u6b0a\u91cd\uf96b\uf969\u503c\uff0c\u70ba\u4e00 \u7dad\u7684\uf96b\uf969\u5411\uf97e [ D ] \u03bb \u03bb \u03bb ,..., 0 = j i W , \uff0c\u6bcf\u4e00\u8a9e\u8a00\u7279\u5fb5\u53ca\u5176\u5c0d\u61c9\u7684\u6b0a\u91cd\uf96b\uf969\u503c\u5982\u5716\u4e00\u6240\u793a\u610f\u3002\u5019\u9078\u8a5e\u5e8f\uf99c \u7684\u6392\u5e8f\u5206\uf969\u5247\u5b9a\u7fa9\u70ba\u7279\u5fb5\u6b0a\u91cd\u5411\uf97e \u03bb \u8207\u7279\u5fb5\u5411\uf97e ( ) j i W , f \u4e4b\u5167\u7a4d\uff1a 1 + D ( ) ( ) ( , 0 , , , \u2211 = = \u2022 = D d j i d d j i j i W f W W Score \u03bb f \u03bb \u03bb ) (1) \u5247\u6392\u5e8f\u5206\uf969\u6700\u9ad8\u7684\u5019\u9078\u8a5e\u5e8f\uf99c \u5373\u70ba\u91cd\u65b0\u6392\u5e8f\u7d50\u679c\uff1a * i W ( ) , max arg , 1 * \u03bb j i M j i W Score W \u2264 \u2264 = (2) \u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u7684\u8a13\uf996\u76ee\u6a19\uff0c\u5373\u662f\u6c42\u5f97\u4e00\u7d44\u6700\u4f73\u6b0a\u91cd\uf96b\uf969\u89e3\uff0c\u80fd\u4f7f\u6392\u5e8f\u5206\uf969\u6700\u9ad8 \u7684\u5019\u9078\u8a5e\u5e8f\uf99c \u8207\u5019\u9078\u8a5e\u5e8f\uf99c\u96c6\u5408 * i W ( ) i x GEN \u4e2d\u7684\u6700\u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f\uf99c \u76f8\u7b49\u3002\u800c \u5728\u6e2c\u8a66\u968e\u6bb5\uff0c\u5247\u662f\uf9dd\u7528\u8a13\uf996\u968e\u6bb5\u6642\u6c42\u5f97\u7684\u6700\u4f73\u6b0a\u91cd\u5411\uf97e\uff0c\u4f7f\u7528\u5f0f(1)\u7684\u8a55\u5206\u6a5f\u5236\uff0c \u5f9e\u6e2c\u8a66\u8a9e\uf906 \u7684\u5019\u9078\u8a5e\u5e8f\uf99c\u96c6\u5408 R i W k y ( ) k y GEN \u4e2d\u627e\u51fa\u5f97\u5206\u6700\u9ad8\u4e4b\u8a5e\u5e8f\uf99c \uff0c\u5c07\u5176\u4f5c \u70ba\u91cd\u65b0\u6392\u5e8f\u5f8c\u7684\u8f38\u51fa\u7d50\u679c\u3002\u4ee5\u4e0b\u5c07\u4ecb\u7d39\u5e7e\u7a2e\u5e38\u7528\u7684\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u3002 * k W (\u4e8c)\u3001\u4e00\u822c\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b 1\u3001\u611f\u77e5\u5668\u6f14\u7b97\u6cd5(Perceptron Algorithm) \u611f\u77e5\u5668[8]\u6700\u521d\u662f\u88ab\u61c9\u7528\u5728\u4eba\u5de5\uf9d0\u795e\u7d93\u7db2\uf937(Artificial Neural Networks)\uf9b4\u57df\uff0c\u5b83\u662f \u4e00\u7a2e\u4e8c\u5143\u5206\uf9d0\u5668\uff0c\u628a\u8f38\u5165 (\u5be6\uf969\u503c\u5411\uf97e)\u6620\u5c04\u5230\u8f38\u51fa\u503c ( \u03bd ) \u03bd f 0 if 1 \u23a7 > + \u22c5 b \u03bd w \u4e0a(\u4e8c\u5143\uf969)\uff0c\u5982\u5f0f(3) \u6240\u793a\uff1a ( ) else 0 \u23a9 \u23a8 = f \u03bd", |
| "eq_num": "(3" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "( ) ( ) ( 1 * , , 2 \u2211 = \u2212 = i i R i Perc W Score W Score F \u03bb \u03bb \u03bb ) 2 1 L (4) \u5247\u6700\u4f73\u6b0a\u91cd\u5411\uf97e \u5373\u70ba\u6eff\u8db3 * \u03bb ( ) \u03bb \u03bb \u03bb Perc F min arg * = * i R i * d \u7684\u6b0a\u91cd\u5411\uf97e\uff0c\u5373\u6700\u5c0f\u5316\u6240\u6709\u8a13\uf996 \u8a9e\uf906\u4e4b\u6700\u9ad8\u6392\u5e8f\u5206\uf969\u5019\u9078\u8a5e\u5e8f\uf99cW \u8207\u6700\u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f\uf99cW \u6392\u5e8f\u5206\uf969\u7684\u5e73\u65b9\u8aa4 \u5dee\u7e3d\u548c\u3002\u70ba\uf9ba\u6c42\u5f97 \u03bb \uff0c\u6211\u5011\u53ef\u63a1\u7528\u68af\ufa01\u4e0b\ufa09\u6cd5(Gradient Descent Method)\u5c07\u5f0f(4) \u5c0d\u6bcf\u4e00\u7dad\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969 \u03bb \u4f5c\u504f\u5fae\u5206\uff0c\u4ee5\u6c42\u5f97\u6bcf\u4e00\u7dad\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\u7684\uf901\u65b0\uf97e\uff0c\u5982 \u5f0f(5)\u6240\u793a\uff1a ( ) ( ) ( ) ( ) ( ) ( ) ( ) \u2212 \u2202 i i d R i d i R i d Perc W f W f W Score W Score F 1 * * , , \u03bb \u03bb \u03bb \u03bb ( ) \u03bb Perc F d \u2211 = \u2212 = \u2202 L (5) \u7136\u800c\uff0c\u7531\u65bc \u53ef\u80fd\u5b58\u5728\u8a31\u591a\u5c40\u90e8\u6700\u4f73\u89e3(Local Minimum Solutions)\uff0c\u5c0e\u81f4\u7121 \u6cd5\u4fdd\u8b49\u68af\ufa01\u4e0b\ufa09\u6cd5\u53ef\u6c42\u5f97\u5168\u57df\u6700\u4f73\u89e3(Global Minimum Solution)\u3002\u56e0\u6b64\uff0c\u611f\u77e5\u5668 \u6f14\u7b97\u6cd5\u5e38\u63a1\u53d6\u96a8\u6a5f\u8fd1\u4f3c\u6cd5(Stochastic Approximation)[3]\uff1b\u76f8\u5c0d\u65bc\u68af\ufa01\u4e0b\ufa09\u6cd5\u540c\u6642 \u5c0d\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a08\u7b97\u6b0a\u91cd\uf96b\uf969\uf901\u65b0\uf97e\uff0c\u96a8\u6a5f\u8fd1\u4f3c\u6cd5\u4f7f\u7528\u589e\uf97e\u5f0f(Incremental)\u7684\u65b9 \u6cd5\u8a08\u7b97\uf901\u65b0\uf97e\uff0c\u4e5f\u5c31\u662f\u5c07\u5f0f(4)\u5c0d\u6bcf\u4e00\uf906\u8a13\uf996\u8a9e\uf906\u7684\u6bcf\u4e00\u7dad\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969 \u03bb \u4f5c\u504f \u5fae\u5206\uff0c\u4ee5\u6c42\u5f97\u6bcf\u4e00\u8a13\uf996\u8a9e\uf906 \u5c0d\u6bcf\u4e00\u7dad\u7279\u5fb5\u7684\u6b0a\u91cd\uf96b\uf969\u8abf\u6574\uf97e\uff1a i x ( ) ( ) ( ) ( ) ( ) ( ) * * R R ( ) \u03bb , i F Perc ( ) , , i d i d i i W f W f W Score W Score \u2212 \u2212 \u03bb \u03bb (6) \u65bc\u662f\uff0c\u96a8\u6a5f\u8fd1\u4f3c\u6cd5\u53ef\u8996\u70ba\u6700\u4f73\u5316\u500b\u5225\u8a13\uf996\u8a9e\uf906\u7684\u6392\u5e8f\u6e1b\u640d\u51fd\uf969 \uff1a ( ) ( ) ( ) 2 1 , \u03bb \u03bb \u03bb i R i Perc W Score W Score i F \u2212 = ( ) (7) \u800c\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u5c0d\u6bcf\u4e00\uf906\u8a13\uf996\u8a9e\uf906\u7684\u6bcf\u4e00\u7dad\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\uf901\u65b0\u5f0f\u53ef\u8868\u793a\u6210\uff1a ( ) ( ) ( ) ( ) ( ) * * , , i d R i d i R i i d i d W f W f W Score W Score \u2212 \u2212 \u22c5 \u2212 = \u03bb \u03bb \u03b7 \u03bb \u03bb (8) \u5176\u4e2d\u03b7 \u70ba\u5b78\u7fd2\u6b65\u8abf\u5e38\uf969\u3002\u95dc\u65bc\u6b0a\u91cd\u8abf\u6574\uf97e\u4ea6\u6709\u5b78\u8005\u63d0\u51fa\u76f4\u63a5\u4ee5\u8a08\u7b97\u8a9e\u8a00\u7279\u5fb5\u5411\uf97e \u7684\u5dee\u503c\uf92d\uf901\u65b0\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969[4]\uff0c\u5373 ( ) ( ) ( ) * i d R i d i d i d W f W f \u2212 + = \u03bb", |
| "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": "\u03bb i d i d d W f W f R i d ...D d x T ...T t ...D d \u2212 \u22c5 + = = = i R i i W L i W x = = = = \u03bb \u5716\u4e8c\u3001\u611f\u77e5\u5668\u6f14\u7b97\u6cd5[12] \u53e6\u5916\uff0c\u6709\u5b78\u8005\u63d0\u51fa\u4ee5\u6bcf\u4e00\u8a13\uf996\u8a9e\uf906\u5728\u6bcf\u4e00\u6b21\u905e\u8ff4\u8a13\uf996\u5f8c\u5404\u81ea\u7684\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969 \u4e4b\u5e73\u5747\u503c\uff0c\u7576\u6210\u6700\u5f8c\u7684\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969[13]\uff1a ( ) ( ) ( ) L T T t L i Local i t d global avg d * 1 1 , _ \u2211\u2211 = = = \u03bb \u03bb (9) \u6b64\u65b9\u6cd5\u7a31\u4e4b\u70ba\u5e73\u5747\u5168\u57df\u7279\u5fb5\u6b0a\u91cd\u611f\u77e5\u5668\u6f14\u7b97\u6cd5(Averaged Perceptron Algorithm)\uff1b \u5c0d\u65bc\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\uff0c\u672c\uf941\u6587\u4f7f\u7528\u5f0f(9)\uf92d\u8868\u793a\u6700\u5f8c\u7684\u8a9e\u8a00\u6a21\u578b\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\u3002 2\u3001\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b(Global Conditional Log-Linear Model, GCLM) \u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u65e9\u5148\u5728\u81ea\u7136\u8a9e\u8a00\u8655\uf9e4\uf9b4\u57df[14, 15]\u88ab\u63d0\u51fa\uff0c\u4e26\u5728 2007 \uf98e \u88ab\u7b2c\u4e00\u6b21\u61c9\u7528\u5728\u8a9e\u97f3\u8fa8\uf9fc\u91cd\u65b0\u6392\u5e8f\u554f\u984c\u4e0a[4]\uff1b\u5b83\uf9dd\u7528\u6b0a\u91cd\uf96b\uf969\u5411\uf97e \u03bb \u53ca\u6bcf\u4e00\uf906\u8a9e \u97f3\u8a0a\u865f \u7684\u6240\u6709\u5019\u9078\u8a5e\u5e8f\uf99c i x ( ) i i,j x W GEN \u2208 \uf92d\u5b9a\u7fa9\u4e00\u500b\u689d\u4ef6\u5206\u4f48\uff1a ( ) ( ) ( ( )) 1 ( ) \u03bb \u03bb , exp , | , , j i i i j i W Score x Z x W p = \u03bb (10) ( \u5176\u4e2d ) ( ) \u2211 = M W Score x Z , exp , \u03bb \u03bb x ( ) ( ) = j j i i 1 , R i W \u4ee3\u8868\u6240\u6709\u5019\u9078\u8a5e\u5e8f\uf99c\u7684\u6307\uf969\u6392\u5e8f\u5206\uf969\u7e3d \u548c\uff0c\u70ba\u4e00\u6b63\u898f\u5316\u7684\u5e38\uf969\u3002\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u5e0c\u671b\u7d66\u5b9a\u4e00\u8a13\uf996\u8a9e\uf906 \uff0c\u5176\u6700 \u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f\uf99c \u7684\u5c0d\uf969\u689d\u4ef6\u6a5f\uf961\u80fd\u8d8a\u5927\u8d8a\u597d\u3002\u56e0\u6b64\uff0c\u6839\u64da\u5f0f(10)\u7684\u5b9a\u7fa9\uff0c\u5168 \u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u7684\u76ee\u6a19\u51fd\uf969\u53ef\u8868\u793a\u6210\uff1a i ( ( )) ( ) ( ) , exp log | log \u2211 \u2211 = = L R i L i R i GCLM W Score x W p F \u03bb \u03bb \u03bb ( ) , exp 1 1 , 1 \u2211 = = = i M j j i i W Score \u03bb", |
| "eq_num": "(" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u03c3 \u03bb \u2211 = = i M j j i GCLM W Score 2 , exp log \u03bb \u03bb \u03bb \u2212 = \u2211 L R i W Score F (12) \u5176\u4e2d\u4fc2\uf969 \u03c3 \u63a7\u5236\u5c0d\uf969\u6a5f\uf961\u9805\u8207\u9ad8\u65af\u4e8b\u524d\u6a5f\uf961\u9593\u7684\u76f8\u4e92\u5f71\u97ff\uff0c\u53ef\uf9dd\u7528\u767c\u5c55\u96c6 (Development Set)\u4f30\u7b97\u5176\u503c\u3002\u800c\u6700\u4f73\u6b0a\u91cd\uf96b\uf969\u9700\u7b26\u5408 \uff0c\u56e0 \u70ba \u70ba\u4e00\u51f8\u51fd\uf969(Convex Function)\uff0c\u6240\u4ee5\u53ef\u6c42\u5f97 ( ) \u03bb \u03bb \u03bb GCLM F max arg * = ( GCLM F ) \u03bb ( ) \u03bb GCLM F \u7684\u5168\u57df\u6700\u4f73\u89e3 (Globally Optimal Solution)\u3002\u70ba\u6b64\uff0c\u6211\u5011\u53ef\u4ee5\uf9dd\u7528\u68af\ufa01\u4e0b\ufa09\u6cd5\u5c07\u6b64\u76ee\u6a19\u51fd\uf969\u5c0d\u6bcf \u4e00\u7dad\u6b0a\u91cd\uf96b\uf969 d \u03bb \u504f\u5fae\u5206\u5f8c\u5373\u53ef\u6c42\u5f97\u6b0a\u91cd\uf96b\uf969\u5411\uf97e * \u03bb \u6bcf\u4e00\u7dad\u7684\u8abf\u6574\uf97e\uff1a ( ) ( ) ( ) ( ) ( ) ( ) ( ) \u2211 \u2211 \u2211 = = = \u2212 \u23a5 \u23a5 \u23a5 \u23a5 \u23a6 \u23a4 \u23a2 \u23a2 \u23a2 \u23a2 \u23a3 \u23a1 \u2212 = \u2202 \u2202 L i d M k k i d M j j i k i R i d GCLM W f W Score W Score W f F 1 2 1 , 1 , , , exp , exp \u03c3 \u03bb \u03bb \u03bb \u03bb \u03bb (13) \u65bc\u662f\u5c0d\u6bcf\u4e00\u7dad\u7279\u5fb5\u5206\u5225\uf901\u65b0\u5176\u6b0a\u91cd\uf96b\uf969\uff0c\u4ee5\u6c42\u5f97\u6700\u4f73\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\uff1a ( ) ( ) ( ) ( ) ( ) ( ) \u2211 \u2211 \u2211 = = = \u2212 \u23a5 \u23a5 \u23a5 \u23a5 \u23a6 \u23a4 \u23a2 \u23a2 \u23a2 \u23a2 \u23a3 \u23a1 \u2212 \u22c5 + = L i d k i d M k M j j i k i R i d d W f W Score W Score W f 1 2 , 1 1 , , , exp , exp \u03c3 \u03bb \u03b7 \u03bb \u03bb \u03bb \u03bb (14) (\u4e09)\u3001\u8003\u616e\u6a23\u672c\u6b0a\u91cd(Sample Weight)\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b 1\u3001\u6b0a\u91cd\u5f0f\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b(Weighted Global Conditional Log-Linear Model, WGCLM) \u6b0a\u91cd\u5f0f\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b[19]\u662f\u5c07\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u52a0\u5165\u6a23\u672c\u6b0a \u91cd\u4f5c\u5ef6\u4f38\u3002\u56e0\u6b64\uff0c\u5b83\u7684\u76ee\u6a19\u51fd\uf969\u5b9a\u7fa9\u70ba\uff1a ( ( )) ( ) ( ) ( ) \u2211 = = L i M R i WGCLM W Score F 1 , exp log \u03bb \u03bb j i, j i W i , , \u2211 = j j i W i W Score j i 1 , , , exp , \u03bb \u03c9 (15) \u6211\u5011\u53ef\u767c\u73fe\u8207\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\uf967\u540c\u9ede\u5728\u65bc\uff0c\u6bcf\u4e00\u689d\u5019\u9078\u8a5e\u5e8f\uf99cW \u6703\u56e0 \u5176\u6a23\u672c\u6b0a\u91cd \u03c9 \uf967\u540c\u800c\u6709\uf967\u540c\u7684\u91cd\u8981\u6027\u3002\u53e6\u5916\uff0c\u8207\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u76f8 \u540c\u7684\u662f\uff0c\u70ba\uf9ba\u907f\u514d\u5728\uf96b\uf969\u8abf\u6574\u904e\u7a0b\u4e2d\u767c\u751f\u904e\ufa01\u8a13\uf996\u73fe\u8c61\uff0c\u6211\u5011\u53ef\u52a0\u5165\uf9b2\u5747\u503c\u9ad8\u65af \u4e8b\u524d\u6a5f\uf961\u65bc\u5f0f(15)\u800c\u5f97\uff1a ( ( )) ( ) ( ) ( ) 2 2 2 , exp log \u03c3 \u03bb \u03bb \u03bb \u2212 = \u2211 L M R i WGCLM W Score F 1 1 , , , exp , \u03c9 \u03bb \u2211 = = i j j i W i W Score j i (16) ( ) \u03bb \u03bb \u03bb WGCLM F max arg * = \u56e0\u6b64\uff0c\u6700\u4f73\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\u9700\u7b26\u5408 \u3002\u540c\u6642\uff0c\u5f9e\u5f0f(16)\u6211\u5011\u53ef \u770b\u51fa\u6b0a\u91cd\u5f0f\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u7684\u76ee\u6a19\u51fd\uf969\u4e5f\u662f\u4e00\u51f8\u51fd\uf969\uff1b\u65bc\u662f\uff0c\u53ef\u4ee5\u6c42\u5f97 * \u03bb \u7684\u5168\u57df\u6700\u4f73\u89e3\u3002\u8207\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u76f8\u540c\u7684\u662f\uff0c\u900f\u904e\u68af\ufa01\u4e0b\ufa09\u6cd5\u5c07\u6b64 \u76ee\u6a19\u51fd\uf969\u5c0d\u6bcf\u4e00\u7dad\u6b0a\u91cd\uf96b\uf969 d \u03bb \u504f\u5fae\u5206\u5f8c\u5373\u53ef\u6c42\u5f97\u6b0a\u91cd\uf96b\uf969\u5411\uf97e \u03bb \u7684\u8abf\u6574\uf97e\uff1a ( ) ( ) ( ) ( ) ( ) ( ) ( ) \u2211 \u2211 \u2211 = \u2212 = \u2212 \u23a5 \u23a5 \u23a5 \u23a5 \u23a6 \u23a4 \u23a2 \u23a2 \u23a2 \u23a2 \u23a3 \u23a1 \u2212 = \u2202 \u2202 L i d M k k i d M j j i W i W i i k i , , , , \u03c9 k i R i d WGCLM W f W Score W Score W f F j 1 2 1 , 1 , , , exp , exp \u03c3 \u03bb \u03c9 \u03bb \u03bb \u03bb \u03bb (17) \u56e0\u6b64\uff0c\u6211\u5011\u53ef\u5c0d\u6bcf\u4e00\u7dad\u7279\u5fb5\u5206\u5225\uf901\u65b0\u5176\u6b0a\u91cd\uff0c\u4ee5\u6c42\u5f97\u8f03\u4f73\u6b0a\u91cd\uf96b\uf969\u5411\uf97e \u03bb \uff1a ( ) ( ) ( ) ( ) ( ) ( ) \u2211 \u2211 = = M k M j 1 \u2211 = \u23a2 \u23a2 \u23a2 \u23a2 \u23a3 \u23a1 \u2212 \u22c5 + = L i d k i d j i W i k i W i R i d d W f W Score W Score W f j i k i 1 2 , 1 , , , , , exp , exp \u02c6, , \u03c3 \u03bb \u03c9 \u03c9 \u03b7 \u03bb \u03bb \u03bb \u03bb \u2212 \u23a5 \u23a5 \u23a5 \u23a5 \u23a6 \u23a4 (18) 2\u3001\u6700\u5c0f\u5316\u932f\u8aa4\uf961\u8a13\uf996(Minimum Error Rate Training, MERT) \u6700\u5c0f\u5316\u932f\u8aa4\uf961\u8a13\uf996[20, 21]\u7684\u6392\u5e8f\u6e1b\u640d\u51fd\uf969\u88ab\u5b9a\u7fa9\u70ba\uff1a ( ) ( ) ( ) ( ) ( ) ( ) ( ) \u2211\u2211 \u2211 = = = \u2212 \u2212 = L i M k M j R i j i R i k i W i MERT W Score W Score W Score W Score F k i 1 exp exp , 1 1 , , , , , , , \u03b2 \u03b2 \u03c9 \u03bb \u03bb \u03bb \u03bb \u03bb (19) ( ( ) \u7d93\u904b\u7b97\u5f8c\uff0c\u6211\u5011\u53ef\u5f88\u6280\u5de7\u5730\u5c07\u5f0f(19)\u4e2d\u7684 ) \u03b2 \u03bb , R W Score ( ) exp \u9805\u6d88\u53bb\uff0c\u4f7f\u6392\u5e8f\u6e1b\u640d \u51fd\uf969\u5316\u7c21\u70ba\uff1a i ( ( )) ) ( ) ( \u2211\u2211 = L M k i W Score F , , exp \u03b2 \u03c9 \u03bb \u03bb \u2211 = = = i k M j j i W i MERT W Score k i 1 1 1 , , , exp , \u03b2 \u03bb k i W i , ,", |
| "eq_num": "(20)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 2 1 , 1 , , , 1 1 , exp , exp , \u239f \u239f \u23a0 \u239e \u239c \u239c \u239d \u239b \u2212 \u22c5 \u2202 \u2211 \u2211 = \u2032 \u2032 = = = M j j i M j k i d j i d j i i k d W Score W f W f W Score k i \u03b2 \u03b2 \u03bb \u03bb \u03bb (2 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) , , , exp \u22c5 \u22c5 = \u2202 \u2211\u2211 L M k i W i MERT W Score F \u03b2 \u03b2 \u03c9 \u03bb \u03bb 1) \u56e0\u6b64\u8207\u524d\u8ff0\u7684\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u76f8\u540c\uff0c\u6211\u5011\u53ef\u5c0d\u6bcf\u4e00\u7dad\u7279\u5fb5\u5206\u5225\uf901\u65b0\u5176\u6b0a\u91cd\uff0c\u4ee5\u6c42 \u5f97\u6700\u4f73\u6b0a\u91cd\uf96b\uf969\uff1a 2 1 , 1 , ,", |
| "eq_num": ", 1 1 ," |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u2211 = = M j j k k 1 , u u (24) \u800c\u6bcf\u4e00\u7fa4\u8a13\uf996\u8a9e\uf906\u7684\u55ae\uf99a\u8a5e\u5411\uf97e v p \u5247\u662f\u5176\u4e2d\u6240\u6709\u8a13\uf996\u8a9e\uf906\u7684\u6b63\u78ba\u8f49\u5beb\u8a9e\uf906\u7684\u55ae \uf99a\u8a5e\u5411\uf97e\u7684\u8cea\u5fc3\uff1a p L j j p p L p \u2211 = = 1 , v v (25) \u5176\u4e2d \u70ba p \u8a13\uf996\u8a9e\uf906\u7fa4\u4e2d\u5305\u542b\u7684\u8a13\uf996\u8a9e\uf906\u7684\uf906\uf969\u3002\u56e0\u6b64\uff0c\u6e2c\u8a66\u8a9e\uf906 \u8207 p \u8a13\uf996 \u8a9e\uf906\u7fa4\u7684\u7d44\u5408\u4fc2\uf969\u70ba\uff1a p L k y \u2211 = = P c c k p k p k 1 , , , cos cos \u03b3 (26) \u5176\u4e2d\uff0c \u5373\u70ba u k \u8207 v p \u7684\u9918\u5f26\u503c\uff1a p k, cos 2 2 , cos p k p k p k v u v u \u22c5 = (27) \u6700\u5f8c\uff0c\u5c0d\u6e2c\u8a66\u8a9e\uf906 \uf92d\uf96f\uff0c\u5176\u65b0\u7684\u8a9e\u8a00\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\u5411\uf97e \u53ef\u8868\u793a\u6210\u70ba P \u500b\u8a9e \u8a00\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\u7684\u7dda\u6027\u7d44\u5408\uff1a k y k \u03bb \u2211 \u22c5 = P \u03bb \u03bb \u03b3 = p p p k k 1 ,", |
| "eq_num": "(28)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Statistical Methods for Speech Recognition, the", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Jelinek", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. Jelinek, Statistical Methods for Speech Recognition, the MIT press, 1999.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Minimum Word Error Based Discriminative Training of Language Models", |
| "authors": [ |
| { |
| "first": "J.-W", |
| "middle": [], |
| "last": "Kuo", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proceedings of International Conference on Spoken Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "1277--1280", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J.-W. Kuo, B. Chen, \"Minimum Word Error Based Discriminative Training of Language Models,\" in Proceedings of International Conference on Spoken Language Processing, pp. 1277-1280, 2005.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "An empirical study on language model adaptation", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Gao", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Suzuki", |
| "suffix": "" |
| }, |
| { |
| "first": "W", |
| "middle": [], |
| "last": "Yuan", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "ACM Transactions on Asian Language Information Processing", |
| "volume": "5", |
| "issue": "3", |
| "pages": "209--227", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. Gao, H. Suzuki, and W. Yuan, \"An empirical study on language model adaptation,\" ACM Transactions on Asian Language Information Processing, Vol. 5, No. 3, pp. 209-227, 2006.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Discriminative n-gram language modeling", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Roark", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Saraclar", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Collins", |
| "suffix": "" |
| } |
| ], |
| "year": 2007, |
| "venue": "Computer Speech and Language", |
| "volume": "21", |
| "issue": "2", |
| "pages": "373--392", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B. Roark, M. Saraclar, and M. Collins, \"Discriminative n-gram language modeling,\" Computer Speech and Language, Vol. 21, No. 2, pp. 373-392, 2007.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Discriminative reranking for natural language parsing", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Collins", |
| "suffix": "" |
| }, |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Koo", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Computational Linguistics", |
| "volume": "31", |
| "issue": "1", |
| "pages": "25--70", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. Collins, T. Koo, \"Discriminative reranking for natural language parsing,\" Computational Linguistics, Vol. 31, No. 1, pp. 25-70, 2005.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Pattern Classification", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [ |
| "O" |
| ], |
| "last": "Duda", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [ |
| "E" |
| ], |
| "last": "Hart", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [ |
| "G" |
| ], |
| "last": "Stork", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification, Wiley, 2001.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Parameter estimation for statistical parsing models: Theory and practice of distribution-free methods", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Collins", |
| "suffix": "" |
| } |
| ], |
| "year": 2004, |
| "venue": "New Developments in Parsing Technology", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. Collins, \"Parameter estimation for statistical parsing models: Theory and practice of distribution-free methods,\" in H. Bunt, J. Carroll, G. Satta (eds.), New Developments in Parsing Technology, 2004.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "The perceptron: A probabilistic model for information storage and organization in the brain", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Rosenblatt", |
| "suffix": "" |
| } |
| ], |
| "year": 1958, |
| "venue": "Psychological Review", |
| "volume": "", |
| "issue": "6", |
| "pages": "386--408", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. Rosenblatt, \"The perceptron: A probabilistic model for information storage and organization in the brain,\" Psychological Review, No. 6, pp. 386-408, 1958.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Discriminative training methods for hidden Markov models: theory and experiments with perceptron algorithms", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Collins", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "1--8", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. Collins, \"Discriminative training methods for hidden Markov models: theory and experiments with perceptron algorithms,\" in Proceedings of the Conference on Empirical Methods in Natural Language Processing, pp. 1-8, 2002.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Conditional random fields: Probabilistic models for segmenting and labeling sequence data", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Lafferty", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Mccallum", |
| "suffix": "" |
| }, |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Pereira", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "Proceedings of International Conference on Machine Learning", |
| "volume": "", |
| "issue": "", |
| "pages": "282--289", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. Lafferty, A. McCallum, and F. Pereira, \"Conditional random fields: Probabilistic models for segmenting and labeling sequence data,\" in Proceedings of International Conference on Machine Learning, pp. 282-289, 2001.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Shallow parsing with conditional random fields", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Sha", |
| "suffix": "" |
| }, |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Pereira", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Proceedings of the North American Chapter of the Association for Computational Linguistics on Human Language Technology", |
| "volume": "", |
| "issue": "", |
| "pages": "134--141", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F. Sha, F. Pereira, \"Shallow parsing with conditional random fields,\" in Proceedings of the North American Chapter of the Association for Computational Linguistics on Human Language Technology, pp. 134-141, 2003.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Machine Learning. The McGraw-Hill Companies", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [ |
| "M" |
| ], |
| "last": "Mitchell", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T. M. Mitchell, Machine Learning. The McGraw-Hill Companies, 1997.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Large margin classification using the perceptron algorithm", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Freund", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [ |
| "E" |
| ], |
| "last": "Schapire", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Machine Learning", |
| "volume": "37", |
| "issue": "3", |
| "pages": "277--296", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Y. Freund and R.E. Schapire, \"Large margin classification using the perceptron algorithm,\" Machine Learning, Vol. 37, No. 3, pp. 277-296, 1999.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "A maximum entropy model for parsing", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Ratnaparkhi", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Roukos", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [ |
| "T" |
| ], |
| "last": "Ward", |
| "suffix": "" |
| } |
| ], |
| "year": 1994, |
| "venue": "Proceedings of International Conference on Spoken Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "803--806", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. Ratnaparkhi, S. Roukos, and R.T. Ward, \"A maximum entropy model for parsing,\" in Proceedings of International Conference on Spoken Language Processing, pp. 803-806, 1994.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Estimators for stochastic \"unification-based\" grammars", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Johnson", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Geman", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Canon", |
| "suffix": "" |
| }, |
| { |
| "first": "Z", |
| "middle": [], |
| "last": "Chi", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Riezler", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Proceedings of The annual meeting of the Association for Computational Linguistics on Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "535--541", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "M. Johnson, S. Geman, S. Canon, Z. Chi, and S. Riezler, \"Estimators for stochastic \"unification-based\" grammars,\" in Proceedings of The annual meeting of the Association for Computational Linguistics on Computational Linguistics, pp. 535-541, 1999.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Maximum mutual information estimation of hidden Markov model parameters for speech recognition", |
| "authors": [ |
| { |
| "first": "L", |
| "middle": [ |
| "R" |
| ], |
| "last": "Bahl", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [ |
| "F" |
| ], |
| "last": "Brown", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [ |
| "V" |
| ], |
| "last": "De Souza", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [ |
| "R" |
| ], |
| "last": "Mercer", |
| "suffix": "" |
| } |
| ], |
| "year": 1986, |
| "venue": "Proceedings of International Conference on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "49--52", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "L.R. Bahl, P.F. Brown, P.V. de Souza, and L.R. Mercer, \"Maximum mutual information estimation of hidden Markov model parameters for speech recognition,\" in Proceedings of International Conference on Acoustics, Speech and Signal Processing, pp. 49-52, 1986.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Discriminative learning for minimum error classification", |
| "authors": [ |
| { |
| "first": "B.-H", |
| "middle": [], |
| "last": "Juang", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Kataguru", |
| "suffix": "" |
| } |
| ], |
| "year": 1992, |
| "venue": "IEEE Transactions on Signal Processing", |
| "volume": "40", |
| "issue": "12", |
| "pages": "3043--3054", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B.-H. Juang and S. Kataguru, \"Discriminative learning for minimum error classification,\" IEEE Transactions on Signal Processing, Vol. 40, No. 12, pp. 3043-3054, 1992.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "A comparative study of discriminative methods for reranking LVCSR n-best hypotheses in domain adaptation and generalization", |
| "authors": [ |
| { |
| "first": "Z", |
| "middle": [], |
| "last": "Zhou", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Gao", |
| "suffix": "" |
| }, |
| { |
| "first": "F", |
| "middle": [ |
| "K" |
| ], |
| "last": "Soong", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Meng", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Proceedings of International Conference on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "141--144", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Z. Zhou, J. Gao, F.K. Soong, and H. Meng, \"A comparative study of discriminative methods for reranking LVCSR n-best hypotheses in domain adaptation and generalization,\" in Proceedings of International Conference on Acoustics, Speech and Signal Processing, pp. 141-144, 2006.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "A comparative study on methods of weighted language model training for reranking LVCSR n-best hypotheses", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Oba", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Hori", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Nakamura", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "Proceedings of International Conference on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "5126--5129", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T. Oba, T Hori, and A. Nakamura, \"A comparative study on methods of weighted language model training for reranking LVCSR n-best hypotheses,\" in Proceedings of International Conference on Acoustics, Speech and Signal Processing, pp. 5126-5129, 2010.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Minimum error rate training in statistical machine translation", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [ |
| "J" |
| ], |
| "last": "Och", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Proceedings of The annual meeting of the Association for Computational Linguistics on Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "160--167", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "F.J. Och, \"Minimum error rate training in statistical machine translation,\" in Proceedings of The annual meeting of the Association for Computational Linguistics on Computational Linguistics, pp. 160-167, 2003.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Discriminative rescoring based on minimization of word errors for transcribing broadcast news", |
| "authors": [ |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Akio", |
| "suffix": "" |
| }, |
| { |
| "first": "O", |
| "middle": [], |
| "last": "Takahiro", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Shinichi", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Shoei", |
| "suffix": "" |
| }, |
| { |
| "first": "I", |
| "middle": [], |
| "last": "Toru", |
| "suffix": "" |
| }, |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Tohru", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "Proceedings of the Annual Conference of the International Speech Communication Association", |
| "volume": "", |
| "issue": "", |
| "pages": "1574--1577", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "K. Akio, O. Takahiro, H. Shinichi, S. Shoei, I. Toru, and T. Tohru, \"Discriminative rescoring based on minimization of word errors for transcribing broadcast news,\" in Proceedings of the Annual Conference of the International Speech Communication Association, pp. 1574-1577, 2008.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "MATBN: A Mandarin Chinese broadcast news corpus", |
| "authors": [ |
| { |
| "first": "H.-M", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "J.-W", |
| "middle": [], |
| "last": "Kuo", |
| "suffix": "" |
| }, |
| { |
| "first": "S.-S", |
| "middle": [], |
| "last": "Cheng", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "International Journal of Computational Linguistics & Chinese Language Processing", |
| "volume": "10", |
| "issue": "2", |
| "pages": "219--236", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H.-M. Wang, B. Chen, J.-W. Kuo, and S.-S. Cheng, \"MATBN: A Mandarin Chinese broadcast news corpus,\" International Journal of Computational Linguistics & Chinese Language Processing, Vol. 10, No. 2, pp. 219-236, 2005.", |
| "links": null |
| }, |
| "BIBREF22": { |
| "ref_id": "b22", |
| "title": "SRI Language Modeling Toolkit", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Stolcke", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. Stolcke, SRI Language Modeling Toolkit, version 1.5.8, http://www.speech.sri.com/projects/srilm/.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "uris": null, |
| "type_str": "figure", |
| "text": "Vector)\u8868\u793a\u4e26\u9032\ufa08\u5206\u7fa4(Clustering)\u3002\u6211\u5011\u4f7f\u7528\u7684\u5206\u7fa4\u65b9\u6cd5\u70ba" |
| }, |
| "TABREF6": { |
| "text": "\u5be6\u9a57\u7d50\u679c\uff0c\u7576\u5206\u7fa4\u500b\uf969\u70ba 10 \u4e14\u03b1 \u8a2d\u70ba 0.6 \u6642\u5b57\u932f\u8aa4\uf961\u70ba 14.45\uff0c\u53ef\u9054\u5230\u76f8\u5c0d\u8fa8\uf9fc \uf961 11.82%\u7684\u986f\u8457\u63d0\u6607\u6548\u679c\uff0c\u6bd4\u8d77\u55ae\u7368\u4f7f\u7528\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd\u7684\u5b57\u932f\u8aa4\uf961 15.71 \u6216\u55ae\u7368\u4f7f\u7528\uf967\u540c\u6e2c\u8a66\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd\u7684\u5b57\u932f\u8aa4\uf961 15.11 \u90fd\u8981\u597d\u7684\u6548\u679c\u3002\u800c\u5f9e \u8868\u516b\u8207\u8868\u4e5d\u5247\u53ef\u770b\u51fa\u7576\u5206\u7fa4\u500b\uf969\u70ba 5 \u4e14\u03b1 \u8a2d\u70ba 0.4 \u6642\uff0c\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b \u7684\u5be6\u9a57\u7d50\u679c\u70ba 15.29\uff0c\u53ef\u9054\u5230\u76f8\u5c0d\u8fa8\uf9fc\uf961 6.71%\uff0c\u7531\u6b64\u53ef\u77e5\u7d50\u5408\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225", |
| "content": "<table><tr><td>\u8868\u4e8c\u3001\u8a13\uf996\u8a9e\uf9be\u8207\u6e2c\u8a66\u8a9e\uf9be\u4e4b\u7d71\u8a08\u8cc7\u8a0a \u8a13\uf996\u8a9e\uf9be 23 \u5c0f\u6642 30,600 \uf906 \u8a13\uf996\u8a9e\uf9be \u6e2c\u8a66\u8a9e\uf9be \u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u7684\u4e00\u822c\u5316(Generalization)\u80fd\uf98a\u8f03\u5176\u5b83\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\uf92d\u5f97\u5dee\u3002\u70ba\uf9ba \u6e2c\u8a66\u8a9e\uf9be \u8a9e\uf9be\u6642\u9593\u9577\ufa01 1.5 \u5c0f\u6642 \u8a9e\uf9be\uf906\uf969 1,997 \uf906 8 10 12 14 16 18 \u63a2\u7a76\u5176\u539f\u56e0\uff0c\u6211\u5011\u9032\u4e00\u6b65\u89c0\u5bdf\uf9ba\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u5728\u6bcf\u4e00\u6b21\u8a13\uf996\u8fed\u4ee3\u6642\uff0c\u8a13\uf996\u8a9e\uf9be\u8207 \u8868\u516b\u3001\u7d50\u5408\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u6b0a\u91cd\u8207\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd \uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6240\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd\u5411\uf97e \u4f54\u64da\u8f03\u91cd\u8981\u7684\u89d2\u8272\uff0c\u800c\u5728\u5168 j \u03bb \u8868\u56db\u3001\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u904b\u7528\u65bc\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u4e4b\u5be6\u9a57\u7d50\u679c(CER(%)) \u8868\uf9d1\u3001\u7d50\u5408\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u6b0a\u91cd (\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u3001\u5206\u7fa4\u500b\uf969=5)\u4e4b\u5be6\u9a57\u7d50\u679c(CER(%)) \u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\uff0c\u5247\u662f\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd all \u03bb \u8f03\u70ba\u91cd\u8981\u3002\u63a2\u7a76 CER(%) \u6e2c\u8a66\u8a9e\uf9be\u5b57\u932f\u8aa4\uf961\u8868\u73fe\u4e4b\u8b8a\u5316\u8da8\u52e2\uff0c\u5982\u5716\u56db\u6240\u793a\u3002\u7531\u5716\u56db\u53ef\u770b\u51fa\uff0c\u5c0d\u65bc\u6e2c\u8a66\u8a9e\uf9be \u800c\u8a00\uff0c\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u6703\u5728\u8f03\u5c11\u7684\u8a13\uf996\u8fed\u4ee3\u6b21\uf969\u6642\u5c31\u6703\u9054\u5230\u6700\u4f4e\u5b57\u932f\u8aa4\uf961\uff0c\u76f8\u8f03\u65bc \u5b83\u5728\u8a13\uf996\u8a9e\uf9be\u4e0a\u7684\u8868\u73fe\u3002\u56e0\u6b64\uff0c\u6211\u5011\uf974\u5c07\u7531\u8a13\uf996\u8a9e\uf9be\u6240\u7372\u5f97\u7684\u6700\u4f73\uf96b\uf969\u6b0a\u91cd\u4f7f\u7528 \u5728\u6e2c\u8a66\u8a9e\uf9be\u4e0a\u6642\uff0c\u53cd\u800c\u7121\u6cd5\u8b93\u6e2c\u8a66\u8a9e\uf9be\u9054\u5230\u6700\u4f73\u7684\u8a9e\u97f3\u8fa8\uf9fc\u6548\u679c\u3002\u9020\u6210\u6b64\u60c5\u6cc1\u7684 \u539f\u56e0\u61c9\u70ba\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u5728\u8a13\uf996\u904e\u7a0b\u4e2d\u5bb9\uf9e0\u767c\u751f\u904e\ufa01\u8a13\uf996(Overtraining)\u7684\u60c5\u6cc1\uff0c\u5c0e \u5206\u7fa4\u500b\uf969 \u6e2c\u8a66\u8a9e\uf9be\u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d\u63d0\u6607\uf961 \u76f8\u5c0d\u63d0\u6607\uf961 5 15.00 1.39 8.49 10 15.11 1.28 \u8207\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd(\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u3001\u5206\u7fa4\u500b\uf969=5)\u4e4b\u5be6\u9a57\u7d50\u679c(CER(%)) \u03b1 \u6e2c\u8a66\u8a9e\uf9be\u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d\u63d0\u6607\uf961 \u5176\u539f\u56e0\u61c9\u70ba\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\uf9e0\u6709\u904e\ufa01\u8a13\uf996\u7684\u554f\u984c\uff1b\u56e0\u70ba\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b \u76f8\u5c0d\u63d0\u6607\uf961 \u6240\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\u56e0\u70ba\u6709\u8003\u616e\u6e2c\u8a66\u8a9e\uf906\u672c\u8eab\u7684\u7279\u6027\uff0c\u6240\u4ee5\u52a0\u91cd\u5b83\u7684\u91cd\u8981 \u03b1 \u6e2c\u8a66\u8a9e\uf9be\u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d\u63d0\u6607\uf961 \u76f8\u5c0d\u63d0\u6607\uf961 0.1 15.42 0.97 5.94 \u6027\u4f3c\u4e4e\u53ef\u4ee5\u6e1b\u8f15\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\u904e\ufa01\u7b26\u5408\u8a13\uf996\u8a9e\uf9be\u7684\u554f\u984c\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u5c0d\u65bc\u5168 7.83 \u539f\u59cb 15.71 0.68 0.1 15.41 0.98 6.01 0.2 15.34 1.05 6.40 \u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\uff0c\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd\u5247\u986f\u7684\u8f03\u70ba\u91cd\u8981\uff0c\u53ef\u80fd\u662f 4.15 0.2 15.18 1.21 7.37 0.3 15.33 1.06 6.47 \u56e0\u70ba\u8981\u6e1b\u7de9\u8a13\uf996\u8a9e\uf9be\u5206\u7fa4\u5f8c\u5c0e\u81f4\u8a13\uf996\u8cc7\uf9be\uf97e\uf967\u8db3\u7684\u554f\u984c\u3002 \u8868\u4e09\u3001\u57fa\u790e\u5be6\u9a57\u7d50\u679c(CER(%)) \u8a13\uf996\u8a9e\uf9be \u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d \u63d0\u6607\uf961 \u76f8\u5c0d \u63d0\u6607\uf961 \u6e2c\u8a66\u8a9e\uf9be \u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d \u63d0\u6607\uf961 \u76f8\u5c0d \u63d0\u6607\uf961 \u57fa\u790e\u8fa8\uf9fc\uf961 11.26 --16.39 --\u611f\u77e5\u5668\u6f14\u7b97\u6cd5 6.02 5.25 46.58 15.71 0.68 4.15 GCLM 9.90 1.36 12.10 15.53 0.86 5.26 WGCLM (\u5b57\u932f\u8aa4\uf961) 9.86 1.40 12.46 15.44 0.95 5.77 WGCLM (\u6392\u5e8f) 9.87 1.39 12.33 15.49 0.90 5.48 MERT (\u5b57\u932f\u8aa4\uf961) 10.45 0.81 7.21 15.33 1.06 6.47 \u5716\u56db\u3001\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u8a13\uf996\u8a9e\uf9be\u8207\u6e2c\u8a66\u8a9e\uf9be\u5b57\u932f\u8aa4\uf961\u8da8\u52e2\u5716 0 2 4 6 1 3 5 7 9 111315171921232527293133353739414345 \u8a13\uf996\u8fed \u4ee3\u6b21\uf969 20 25 30 CER(%) \u81f4\u8a13\uf996\u5f8c\u7684\u6700\u4f73\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\u904e\u65bc\u7b26\u5408\u8a13\uf996\u8a9e\uf9be\uff1b\u4f7f\u5f97\u7576\u8a13\uf996\u597d\u7684\u8a9e\u8a00\u6a21\u578b\u7684\u6b0a \u91cd\uf96b\uf969\u5411\uf97e\u88ab\u4f7f\u7528\u5728\u6e2c\u8a66\u8a9e\uf9be\u6642\uff0c\u4e26\u7121\u6cd5\u767c\u63ee\u5c0d\u61c9\u7684\u6548\u679c\u3002 \u616e\uf9ba\u8a13\uf996\u8a9e\uf906\u4e2d\u6240\u6709\u7684\u5019\u9078\u8a5e\u5e8f\uf99c\u8207\u6700\u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f\uf99c\u9593\u7684\u95dc\u4fc2\uff1b\u800c\u611f\u77e5\u5668\u6f14\u7b97 \u6cd5\u5247\u53ea\u6709\u8003\u616e\u76ee\u524d\u8a13\uf996\u6b0a\u91cd\u4e0b\uff0c\u6392\u5e8f\u5206\uf969\u6700\u9ad8\u90a3\u689d\u5019\u9078\u8a5e\u5e8f\uf99c\u8207\u6700\u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f \u63a5\u4e0b\uf92d\uff0c\u6211\u5011\u6bd4\u8f03\uf978\u7a2e\u8003\u616e\u6a23\u672c\u6b0a\u91cd\u7684\u9451\u5225\u5f0f\u8a13\uf996\u6a21\u578b\uff0c\u4e5f\u5c31\u662f\u6b0a\u91cd\u5f0f\u5168\u57df \u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u8207\u6700\u5c0f\u5316\u932f\u8aa4\uf961\u8a13\uf996\uff1b\u5b83\u5011\u7686\u4f7f\u7528\uf9ba\uf978\u7a2e\uf967\u540c\u7684\u6a23\u672c\u6b0a\u91cd\uf92d (\u4e09)\u3001\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b \u672c\u7bc0\u8a0e\uf941\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u4e4b\u5be6\u9a57\u7d50\u679c\uff0c\u521d\u6b65\u5206\u5225\u4ee5\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u53ca\u5168 \u8207\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd(\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u3001\u5206\u7fa4\u500b\uf969=10)\u4e4b\u5be6\u9a57\u7d50\u679c(CER(%)) \u6e2c\u8a66\u8a9e\uf9be\u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d\u63d0\u6607\uf961 \u76f8\u5c0d\u63d0\u6607\uf961 0.1 15.43 0.96 5.88 \u4ee5\u5176\u9054\u5230\uf901\u52a0\u7684\u6548\u679c\u3002 \u03b1 \u6e2c\u8a66\u8a9e\uf9be\u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d\u63d0\u6607\uf961 \u76f8\u5c0d\u63d0\u6607\uf961 \u5411\uf97e\u3002\u672a\uf92d\uff0c\u6211\u5011\u5c07\u7814\u7a76\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\uf967\u540c\u7684\u7d44\u5408\u4fc2\uf969\u8a08\u7b97\u65b9\u6cd5\uff0c \u03b1 \u540c\u6b0a\u91cd\u5411\uf97e\uf901\u597d\u7684\u6548\u679c\uff0c\u56e0\u5176\u8003\u616e\uf9ba\uf967\u540c\u7684\u6e2c\u8a66\u8a9e\uf906\u7684\u7279\u6027\uff0c\u7d66\u4e88\u5176\uf967\u540c\u7684\u6b0a\u91cd \u5168\u3002 \u8868\u4e03\u3001\u7d50\u5408\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u6b0a\u91cd (\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u3001\u5206\u7fa4\u500b\uf969=10)\u4e4b\u5be6\u9a57\u7d50\u679c(CER(%)) \u6a21\u578b\u6b0a\u91cd\u5411\uf97e\u7684\u7f3a\u9ede\u3002\u5f9e\u5be6\u9a57\u7d50\u679c\u53ef\u767c\u73fe\uff0c\u6b64\u65b9\u6cd5\u53ef\u9054\u5230\u6bd4\u6240\u6709\u6e2c\u8a66\u8a9e\uf906\u7686\u70ba\u76f8 \uf99c\u9593\u7684\u95dc\u4fc2\u3002\u76f8\u8f03\u4e4b\u4e0b\uff0c\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u5728\u8a13\uf996\u6642\u8003\u616e\u7684\u5c64\u9762\u61c9\uf901\u70ba\u5468 10 15.67 0.72 4.41 \u539f\u59cb 15.53 0.86 5.26 0.8 14.74 1.65 0.9 14.88 1.51 \u8868\u4e5d\u3001\u7d50\u5408\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u6b0a\u91cd\u8207\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd \uf906\u7d50\u5408\uf967\u540c\u6b0a\u91cd\u5411\uf97e\uff0c\u4ee5\u6539\u5584\u50b3\u7d71\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6240\u6709\u6e2c\u8a66\u8a9e\uf906\u7686\u4f7f\u7528\u76f8\u540c\u8a9e\u8a00 9.23 \u7d50\u679c\u6240\u4ee3\u8868\u7684\u610f\u7fa9\u3002\u6211\u5011\u4e26\u63d0\u51fa\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\uff0c\u53ef\u91dd\u5c0d\u500b\u5225\u6e2c\u8a66\u8a9e 10.04 0.9 15.42 0.97 5.90 \u6a21\u578b\u7684\u57fa\u790e\u7cbe\u795e\uff0c\u4e26\u8a0e\uf941\uf9ba\u5e7e\u7a2e\u73fe\u6709\u7684\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\uff0c\u63a2\u8a0e\u5b83\u5011\u7684\u8868\u73fe\u8207\u5be6\u9a57 \uf961\u63d0\u6607\u3002\u63a2\u7a76\u5176\u539f\u56e0\uff0c\u6211\u5011\u63a8\u6e2c\u61c9\u70ba\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u5728\uf901\u65b0\u6b0a\u91cd\u6642\uff0c\u8003 5 15.48 0.91 5.57 0.7 14.81 1.58 9.63 0.8 15.43 0.96 5.86 \ufa09\u4f4e\u932f\u8aa4\uf961\u70ba\u8a13\uf996\u76ee\u6a19\u3002\u672c\uf941\u6587\u91dd\u5c0d\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u505a\u8a0e\uf941\uff0c\u4ecb\u7d39\uf9ba\u9451\u5225\u5f0f\u8a9e\u8a00 \u6211\u5011\u53ef\u4ee5\u767c\u73fe\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u8f03\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u80fd\u6709\uf901\u986f\u8457\u7684\u8a9e\u97f3\u8fa8\uf9fc \u5206\u7fa4\u500b\uf969 \u6e2c\u8a66\u8a9e\uf9be\u8fa8\uf9fc\u932f\u8aa4\uf961 \u7d55\u5c0d\u63d0\u6607\uf961 \u76f8\u5c0d\u63d0\u6607\uf961 0.6 14.79 1.60 9.78 0.7 15.47 0.92 5.62 \u5230\u8a9e\u97f3\u8a0a\u865f\u5c0d\u61c9\u6a5f\uf961\u6700\u9ad8\u7684\u8a5e\u5e8f\uf99c\u70ba\u76ee\u6a19\uff1b\u8fd1\uf98e\uf92d\u65b0\u8208\u7684\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u4ee5\u76f4\u63a5 \u53e6\u4e00\u65b9\u9762\uff0c\u7576\u6bd4\u8f03\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u8207\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u7684\u5be6\u9a57\u7d50\u679c\u6642\uff0c \u8868\u4e94\u3001\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u904b\u7528\u65bc\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u4e4b\u5be6\u9a57\u7d50 0.3 15.06 1.33 8.11 0.4 14.96 1.43 8.71 0.4 15.29 1.10 6.71 0.5 15.31 1.08 \uf9d1\u3001\u7d50\uf941\u8207\u672a\uf92d\u5c55\u671b 6.58 \u679c(CER(%)) 0.5 14.86 1.53 9.36 0.6 15.42 0.97 5.94 \u5728\u8a9e\u97f3\u8fa8\uf9fc\uf9b4\u57df\u4e2d\uff0c\u8a9e\u8a00\u6a21\u578b\u6709\u8457\u8209\u8db3\u8f15\u91cd\u7684\u5730\u4f4d\uff0c\u50b3\u7d71\u7684 \uf99a\u8a9e\u8a00\u6a21\u578b\u4ee5\u627e N</td></tr><tr><td>MERT (\u6392\u5e8f) \uf961\u8a13\uf996\uff0c\u7686\u4ee5\u5b57\u932f\u8aa4\uf961\u70ba\u6a23\u672c\u6b0a\u91cd\u7684\u65b9\u5f0f\u6703\u8f03\u6392\u5e8f\u70ba\u6a23\u672c\u6b0a\u91cd\u7684\u65b9\u5f0f\uf92d\u7684\u6709\u6548\u679c\u3002 10.54 0.73 6.45 15.40 0.99 6.03 \u97f3\u8fa8\uf9fc\u6642\u5b57\u932f\u8aa4\uf961\u4e4b\u9593\u7684\u95dc\u4fc2\uff0c\u5b83\u5011\u7684\u6700\u5f8c\u4e00\ufa08\u7686\u70ba\u50b3\u7d71\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u7684\u7d50\u679c\u3002 0.3 14.94 1.45 8.84 0.4 15.39 1.00 6.12 10 15 \u9032\ufa08\u8a13\uf996\uff0c\u5206\u5225\u662f\u5404\u5019\u9078\u8a5e\u5e8f\uf99c\u7684\u5b57\u932f\u8aa4\uf961\u8207\u5404\u5019\u9078\u8a5e\u5e8f\uf99c\u6839\u64da\u5b57\u932f\u8aa4\uf961\u6240\u505a\u7684 \u6392\u5e8f\u3002\u7531\u5be6\u9a57\u7d50\u679c\u53ef\u770b\u51fa\uff0c\uf967\u7ba1\u662f\u6b0a\u91cd\u5f0f\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u6216\u6700\u5c0f\u5316\u932f\u8aa4 \u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u505a\u70ba\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u70ba\u57fa\u790e\uf92d\u5efa\uf9f7\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e \u8a00\u6a21\u578b\u3002\u8868\u56db\u8207\u8868\u4e94\u6bd4\u8f03\uf978\u7a2e\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6240\u4f7f\u7528\u8a13\uf996\u8a9e\uf9be\u7684\u5206\u7fa4\u500b\uf969\u8207\u5728\u8a9e 0.1 15.32 1.07 6.56 0.2 15.07 1.32 8.05 0.2 15.37 1.02 6.23 0.3 15.38 1.01 6.14 \uf96b\u8003\u6587\u737b</td></tr><tr><td>5 \u70ba\u63a2\u7a76\u5176\u539f\u56e0\uff0c\u6211\u5011\u53d6\uf9ba\u8a13\uf996\u8a9e\uf9be\u5176\u4e2d\u4e00\uf906\u8a13\uf996\u8a9e\uf906\u7684\u524d 100 \u689d\u6700\u4f73\u5019\u9078\u8a5e\u5e8f\uf99c \u6bd4\u8f03\u9019\uf978\u7a2e\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\uff0c\u53ef\u770b\u51fa\u8a13\uf996\u8a9e\uf9be\u5206\u7fa4\u8207\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u7684 0.4 14.77 1.62 9.89 0.5 15.42 0.97 5.94 \u56db\u3001\u5be6\u9a57\u7d50\u679c\u8207\u8a0e\uf941 (\u4e00)\u3001\u5be6\u9a57\u8a9e\uf9be \u672c \uf941 \u6587 \u8a9e \u97f3 \u8fa8 \uf9fc \u5be6 \u9a57 \u6240 \u7528 \u7684 \u8a9e \u97f3 \u8a9e \uf9be \uf92d \u81ea \u516c \u8996 \u65b0 \u805e (Mandarian Across Taiwan\u2500Broadcast News, MATBN)[22]\u3002\u516c\u8996\u65b0\u805e\u8a9e\uf9be\u662f 2001 \uf98e\u81f3 2003 \uf98e\u9593\u7531 \u542b\u5167\u5834\u8207\u5916\u5834\uf978\u500b\u90e8\u5206\uff1a\u5167\u5834\u70ba\u4e3b\u64ad(Studio Anchors)\u8a9e\uf9be\uff0c\u5916\u5834\u5247\u6709\u63a1\u8a2a\u8a18\u8005 (Field Reporters)\u8a9e\uf9be\u8207\u53d7\u8a2a\u8005(Interviewees)\u8a9e\uf9be\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u5c0d\u65bc\u5be6\u9a57\u6240\u4f7f\u7528\u7684 \u80cc\u666f\u4e09\uf99a\u8a9e\u8a00\u6a21\u578b(Background Trigram Language Model)\uff0c\u5176\u8a13\uf996\u8a9e\uf9be\u5247\u662f\uf92d\u81ea\u4e2d \uf961(Character Error Rate, CER)\uff0c\u5728\u8a13\uf996\u8a9e\uf9be\u70ba 11.26\uff0c\u5728\u6e2c\u8a66\u8a9e\uf9be\u5247\u70ba 16.39\u3002\u7531 \u518d\u8005\uff0c\u7576\u6211\u5011\u6bd4\u8f03\u6700\u5c0f\u5316\u932f\u8aa4\uf961\u8a13\uf996\u8207\u5176\u4ed6\u65b9\u6cd5\u7684\u5be6\u9a57\u7d50\u679c\u6642\uff0c\u53ef\u770b\u51fa\u6700\u5c0f (\u56db)\u3001\u7d50\u5408\uf967\u540c\u6e2c\u8a66\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd\u8207\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u6b0a\u91cd ( ) all j combine j \u03bb \u03bb \u03bb \u22c5 \u2212 + \u22c5 = \u03b1 \u03b1 (29) 1 \u5982\u8868\u4e09\u6240\u793a\u3002\u672c\uf941\u6587\u5be6\u9a57\u6240\u7528\u7684\u57fa\u790e\u8fa8\uf9fc\u5668(\u50c5\u4f7f\u7528\u80cc\u666f\u4e09\uf99a\u8a9e\u8a00\u6a21\u578b)\u7684\u5b57\u932f\u8aa4 \u8003\u616e\uf9ba\u6bcf\u689d\u5f8c\u9078\u8a5e\u5e8f\uf99c\u7684\u6392\u5e8f\u6216\u932f\u8aa4\uf961\uff0c\u56e0\u6b64\u53ef\u9054\u5230\uf901\u597d\u7684\u6548\u679c\u3002 \u9996\u5148\uff0c\u6211\u5011\u6bd4\u8f03\u672c\uf941\u6587\u6240\u4ecb\u7d39\u7684\u5e7e\u7a2e\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u6f14\u7b97\u6cd5\u7684\u57fa\u790e\u5be6\u9a57\u7d50\u679c\uff0c \u7684\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\u3002 \uf901\u597d\u7684\u6548\u679c\u3002\u7531\u6b64\u53ef\u77e5\uff0c\u6a23\u672c\u6b0a\u91cd\u5c0d\u9451\u5225\u5f0f\u8a9e\u8a00\u8a13\uf996\u6709\u4e00\u5b9a\u7684\u5e6b\u52a9\uff0c\u56e0\u70ba\u8a13\uf996\u6642 \u4e2d\u592e\u7814\u7a76\u9662\u8cc7\u8a0a\u6240\u53e3\u8a9e\u5c0f\u7d44\u8207\u53f0\u7063\u516c\u5171\u96fb\u8996\u53f0\u5408\u4f5c\u8490\u96c6\u800c\u6210\uff1b\u6bcf\u4e00\u7bc7\u65b0\u805e\u5831\u5c0e\u5305 \u5716\u4e94\u3001\u8a13\uf996\u8a9e\uf906 100 \u689d\u6700\u4f73\u8fa8\uf9fc\u7d50\u679c\u4e4b\u5b57\u932f\u8aa4\uf961 0 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 \u524d100\u689d\u6700 \u4f73\u8fa8\uf9fc\u7d50\u679c (\u4e8c)\u3001\u57fa\u790e\u5be6\u9a57\u7d50\u679c \uf92d\u89c0\u5bdf\u6bcf\u4e00\u689d\u5019\u9078\u8a5e\u5e8f\uf99c\u7684\u5b57\u932f\u8aa4\uf961\uff0c\u5982\u5716\u4e94\u6240\u793a\u3002\u7531\u5716\u4e94\u53ef\u770b\u51fa\uff0c\u524d 100 \u689d\u6700 \u7d50\u5408\u7684\u63d0\u6607\u6548\u679c\u8f03\u5176\u8207\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u7684\u7d50\u5408\u70ba\u4f73\u3002\u63a2\u7a76\u5176\u539f\u56e0\u53ef\u80fd\u70ba 0.5 14.59 1.80 0.6 15.37 1.02 6.21 11.01 \u4f73\u8fa8\uf9fc\u7d50\u679c\u4e2d\u6709\u8a31\u591a\u5019\u9078\u8a5e\u5e8f\uf99c\u64c1\u6709\u76f8\u540c\u7684\u5b57\u932f\u8aa4\uf961\uff0c\uf974\u4ee5\u6392\u5e8f\u70ba\u6a23\u672c\u6b0a\u91cd\uff0c\u64c1 \u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u56e0\u5bb9\uf9e0\u5c0e\u81f4\u904e\ufa01\u8a13\uf996\uff0c\u800c\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u56e0\u6709\u8003\u616e\u6e2c\u8a66 0.6 14.45 1.94 0.7 15.41 0.98 5.97 11.82 \u6709\u76f8\u540c\u7684\u5b57\u932f\u8aa4\uf961\u7684\u5019\u9078\u8a5e\u5e8f\uf99c\u4e5f\u6703\u6709\uf967\u540c\u7684\u6392\u5e8f\uff0c\u56e0\u800c\u5f71\u97ff\uf9ba\u4ee5\u6392\u5e8f\u70ba\u6a23\u672c\u6b0a \u8a9e\uf906\u7684\u7279\u6027\uf92d\u9078\u64c7\u5404\u500b\u8a13\uf996\u8a9e\uf9be\u6240\u7522\u751f\u7684\u8a9e\u8a00\u7279\u5fb5\u6b0a\u91cd\uf96b\uf969\u5411\uf97e\u4f5c\u7d44\u5408\uff0c\u6240\u4ee5\u53ef 0.7 14.46 1.93 11.77 0.8 15.45 0.94 5.72 \u91cd\u7684\u65b9\u5f0f\u7684\u9451\u5225\u80fd\uf98a\uff0c\u5c0e\u81f4\u91cd\u65b0\u6392\u5e8f\u7d50\u679c\u8f03\u5dee\u3002\uf974\u5c07\u4e0a\u8ff0\uf978\u7a2e\u8003\u616e\u6a23\u672c\u6b0a\u91cd\u7684\u9451 \u5225\u5f0f\u8a13\uf996\u6a21\u578b\u8207\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u548c\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u5be6\u9a57\u7d50\u679c\u505a\u6bd4\u8f03\uff0c\u5247\u53ef \u770b\u51fa\u8003\u616e\u6a23\u672c\u6b0a\u91cd\u7684\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6703\u8f03\u672a\u8003\u616e\u6a23\u672c\u6b0a\u91cd\u7684\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6709 \u4ee5\u6e1b\u8f15\u50b3\u7d71\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u5bb9\uf9e0\u904e\ufa01\u8a13\uf996\u7684\u554f\u984c\u3002\u53e6\u5916\uff0c\u6211\u5011\u4e5f\u53ef\u770b\u51fa\u5206\u7fa4\uf969\u76ee\u904e \u7684\u5206\u7fa4\uf969\u76ee\u5c0e\u81f4\u6bcf\u4e00\u7fa4\u7684\u8a13\uf996\u8a9e\uf906\u7684\uf906\uf969\uf967\u8db3\uff0c\u800c\u7121\u6cd5\u8a13\uf996\u51fa\u5177\u6709\u8db3\u5920\u9451\u5225\u80fd\uf98a 0.9 14.87 1.52 9.25 \u591a\u6642\uff0c\u53ef\u80fd\u6703\u5c0e\u81f4\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u7684\u6548\u80fd\u4e0b\ufa09\uff1b\u63a2\u7a76\u5176\u539f\u56e0\u61c9\u70ba\u904e\u591a 0.8 14.56 1.83 11.14 0.9 15.56 0.83 5.09</td></tr><tr><td>\u592e\u901a\u8a0a\u793e 2001 \uf98e\u81f3 2002 \uf98e\u7684\u6587\u5b57\u65b0\u805e\u8a9e\uf9be\uff0c\u5305\u542b\uf9ba\u7d04\u4e00\u5104\u4e94\u5343\u842c\u500b\u4e2d\u6587\u5b57\uff0c\u7d93 \u65b7\u8a5e\u5f8c\u7d04\u6709\u516b\u5343\u842c\u8a5e\u3002\u6211\u5011\u63a1\u7528 SRI Language Modeling Toolkit[23]\u8a13\uf996\u5be6\u9a57\u6240\u9700 \u8981\u7684\u4e09\uf99a\u8a9e\u8a00\u6a21\u578b\u3002\u800c\u8a13\uf996\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6240\u9700\u7684\u8a9e\u97f3\u8a9e\uf9be\u662f\u53d6\u81ea\u516c\u8996\u65b0\u805e\uff0c\u5171 30,600 \uf906\uff0c\u7d04 23 \u5c0f\u6642\u3002\u6700\u5f8c\uff0c\u8a9e\u97f3\u8fa8\uf9fc\u4e4b\u6e2c\u8a66\u8a9e\uf9be\u8207\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u8a9e\uf9be \u5c6c\u65bc\u540c\u6642\u671f(\u4ea6\u662f\u53d6\u81ea\u516c\u8996\u65b0\u805e)\uff0c\u5171 1,997 \uf906(\u7d04 1.5 \u5c0f\u6642)\uff0c\u5982\u8868\u4e8c\u6240\u793a\u3002 \u8868\u4e09\u53ef\u770b\u51fa\uff0c\u672c\uf941\u6587\u6240\u4ecb\u7d39\u7684\u5e7e\u7a2e\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u6f14\u7b97\u6cd5\u7576\u4e2d\uff0c\u4ee5\u6700\u5c0f\u5316\u932f\u8aa4\uf961\u8a13 \uf996(MERT)\u5728\u6e2c\u8a66\u8a9e\uf9be\u6709\u6700\u4f73\u7684\u6548\u80fd\uff0c\u63a5\u8457\u4f9d\u5e8f\u70ba\u6b0a\u91cd\u5f0f\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21 \u578b(WGCLM)\u3001\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b(GCLM)\u53ca\u611f\u77e5\u5668\u6f14\u7b97\u6cd5(Percepton)\u3002 \u672c\u7bc0\u5c07\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6240\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd\u5411\uf97e \u8207\u50b3\u7d71\u9451\u5225\u5f0f\u8a9e\u8a00 j \u03bb \u5316\u932f\u8aa4\uf961\u8a13\uf996\u7372\u5f97\u8f03\u4f73\u7684\u8a9e\u97f3\u8fa8\uf9fc\uf961\u3002\u5176\u539f\u56e0\u9664\uf9ba\u662f\u6a23\u672c\u6b0a\u91cd\u7684\u5e6b\u52a9\u5916\uff0c\u9084\u53ef\u80fd \u662f\u7531\u65bc\u6700\u5c0f\u5316\u932f\u8aa4\uf961\u8a13\uf996\u7684\u76ee\u6a19\u51fd\uf969(\uf96b\u8003\u5f0f(20))\u4e26\u6c92\u6709\u7528\u5230\u6700\u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f\uf99c \u4f5c\u70ba\u8a13\uf996\u6a19\u6e96\uff0c\u800c\u662f\u8003\u616e\u6240\u6709\u5019\u9078\u8a5e\u5e8f\uf99c\u7684\u5f71\u97ff\uff0c\uf967\u50cf\u5176\u5b83\u4e09\u7a2e\u9451\u5225\u5f0f\u8a13\uf996\u6a21\u578b \u6a21\u578b\u4f7f\u7528\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd all \u03bb \u4f5c\u7dda\u6027\u7d44\u5408\uff0c\u5f97\u5230\u4e00\u7d44\u65b0\u7684\u6b0a\u91cd\u5411\uf97e \uff0c\u4ee5\u671f\uf978\u7a2e\uf967\u540c\u8a13\uf996\u76ee\u6a19\u6240\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd\u5411\uf97e\u53ef\u4ee5\u5e6b\u52a9\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u7372 \u5230\uf901\u597d\u7684\u91cd\u65b0\u6392\u5e8f\u7d50\u679c\uff0c\u5176\u7dda\u6027\u7d44\u5408\u7684\u65b9\u5f0f\u5982\u5f0f(29)\u6240\u793a\uff1a combine j \u03bb \u5176\u4e2d\u03b1 \u70ba\u8abf\u6574\u4fc2\uf969\uff0c\u7528\uf92d\u63a7\u5236\u8a9e\uf906\u76f8\u95dc\u4e4b\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\u6240\u8a13\uf996\u7684\u8a9e\u8a00\u7279\u5fb5\u6b0a\u91cd \u5411\uf97e \u03bb \u8207\u4f7f\u7528\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u7684\u8a9e\u8a00\u7279\u5fb5\u6b0a\u91cd \u7684\u8ca2\u737b\uff1b\u5728\u672c\u7814\u7a76 j all \u03bb \u5f0f\u8a9e\u8a00\u6a21\u578b\u8a13\uf996\u6b0a\u91cd\u8207\u6240\u6709\u8a13\uf996\u8a9e\uf906\u8a13\uf996\u7684\u7279\u5fb5\u6b0a\u91cd\u53ef\u9054\u5230\u4e92\u88dc\u7684\u6548\u679c\uff0c\u56e0\u800c\u9054 \u03b1 \u7684\u503c\u662f \u5230\u986f\u8457\u7684\u63d0\u6607\u3002 \u4ecb\u65bc 0.1 \u5230 0.9 \u4e4b\u9593\u3002 \u5bb9\uf9e0\u53d7\u6700\u4f4e\u932f\u8aa4\uf961\u8a5e\u5e8f\uf99c\u5f71\u97ff\u8a13\uf996\u7d50\u679c\u3002\u56e0\u6b64\uff0c\u6700\u5c0f\u5316\u932f\u8aa4\uf961\u8a13\uf996\u53ef\u5f97\u5230\uf901\u5177\u4e00 \u5f9e\u8868\u4e09\u4ea6\u53ef\u770b\u51fa\uff0c\u96d6\u7136\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u5728\u8a13\uf996\u8a9e\uf9be\u53ef\u9054\u5230\u6700\u4f73\u7684\u8a9e\u97f3\u8fa8\uf9fc\uf961\uff0c \u4f46\u7576\u5b83\u4f7f\u7528\u5728\u6e2c\u8a66\u8a9e\uf9be\u6642\uff0c\u5176\u6548\u679c\u537b\u662f\u8f03\u5176\u5b83\u9451\u5225\u5f0f\u8a9e\u8a00\u6a21\u578b\uf92d\u5f97\u4f4e\uff1b\u9019\u986f\u793a\u51fa \u822c\u5316\u7684\u6b0a\u91cd\u5411\uf97e\uff0c\u8b93\u8a13\uf996\u5f8c\u7684\u6700\u4f73\u6b0a\u91cd\u5411\uf97e\uf967\u53ea\u7b26\u5408\u8a13\uf996\u8a9e\uf9be\uff0c\u5728\u6e2c\u8a66\u8a9e\uf9be\u4e5f\u80fd \u9054\u5230\uf967\u932f\u7684\u6548\u679c\u3002 \u53e6\u5916\uff0c\u6211\u5011\uf974\u89c0\u5bdf\u5728\u6700\u4f4e\u5b57\u932f\u8aa4\uf961\u6240\u8a2d\u7684\u03b1 \u503c\u6642\uff0c\u53ef\u767c\u73fe\u5c0d\u65bc\u611f\u77e5\u5668\u6f14\u7b97\u6cd5 \u8868\uf9d1\u8207\u8868\u4e03\u70ba\u4f7f\u7528\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u7684\u5be6\u9a57\u7d50\u679c\uff0c\u800c\u8868\u516b\u8207\u8868\u4e5d\u5247\u70ba\u4f7f\u7528\u5168\u57df\u689d \u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u7684\u5be6\u9a57\u7d50\u679c\u3002\u7531\u8868\uf9d1\u8207\u8868\u4e03\u53ef\u770b\u51fa\uff0c\u5c0d\u65bc\u4f7f\u7528\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u7684 \u03b1 \u503c\u70ba 0.6\uff0c\u5168\u57df\u689d\u4ef6\u5f0f\u5c0d\uf969\u7dda\u6027\u6a21\u578b\u70ba 0.4\uff1b\u4e5f\u5c31\u662f\uf96f\uff0c\u5728\u611f\u77e5\u5668\u6f14\u7b97\u6cd5\u4e2d\uff0c\u8a9e</td></tr></table>", |
| "html": null, |
| "num": null, |
| "type_str": "table" |
| } |
| } |
| } |
| } |
