| { |
| "paper_id": "2020", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T12:54:36.738374Z" |
| }, |
| "title": "\u5f15 \u5f15 \u5f15\u5165 \u5165 \u5165\u6e90 \u6e90 \u6e90\u7aef \u7aef \u7aef\u4fe1 \u4fe1 \u4fe1\u606f \u606f \u606f\u7684 \u7684 \u7684\u673a \u673a \u673a\u5668 \u5668 \u5668\u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u81ea \u81ea \u81ea\u52a8 \u52a8 \u52a8\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5\u7814 \u7814 \u7814\u7a76 \u7a76 \u7a76", |
| "authors": [ |
| { |
| "first": "\u7f57", |
| "middle": [ |
| "\u7f57" |
| ], |
| "last": "\u7f57\u742a \u742a \u742a \u674e \u674e \u674e\u8302 \u8302 \u8302\u897f \u897f \u897f", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Jiangxi Normal University Nanchang", |
| "location": { |
| "postCode": "330022", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "\u6c5f\u897f\u5e08\u8303\u5927\u5b66", |
| "middle": [], |
| "last": "\u8ba1\u7b97\u673a\u4fe1\u606f\u5de5\u7a0b\u5b66\u9662 / \u6c5f\u897f \u5357\u660c", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Jiangxi Normal University Nanchang", |
| "location": { |
| "postCode": "330022", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Qi", |
| "middle": [], |
| "last": "Luo", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Jiangxi Normal University Nanchang", |
| "location": { |
| "postCode": "330022", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Maoxi", |
| "middle": [], |
| "last": "Li", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Jiangxi Normal University Nanchang", |
| "location": { |
| "postCode": "330022", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "Automatic evaluation of machine translation is one of the most critical tasks in machine translation. However, the source sentence information is completely ignored and only the reference is used to measure the translation quality in previous work. For this shortcoming, the paper presents a novel automatic evaluation metric incorporating the source information: extracting the quality embeddings that describes the translation quality from a tuple consist of the machine translations and their corresponding source sentences, and incorporating it into the automatic evaluation method based on contextual embeddings by using a deep neural network. The experimental results on the dataset of WMT'19 Metrics task show that the proposed method can effectively enhance the correlation between the results of the automatic evaluation metrics and that of the human judgments. Deep analysis further reveals that the information of the source sentences plays an important role in automatic evaluation of machine translation.", |
| "pdf_parse": { |
| "paper_id": "2020", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "Automatic evaluation of machine translation is one of the most critical tasks in machine translation. However, the source sentence information is completely ignored and only the reference is used to measure the translation quality in previous work. For this shortcoming, the paper presents a novel automatic evaluation metric incorporating the source information: extracting the quality embeddings that describes the translation quality from a tuple consist of the machine translations and their corresponding source sentences, and incorporating it into the automatic evaluation method based on contextual embeddings by using a deep neural network. The experimental results on the dataset of WMT'19 Metrics task show that the proposed method can effectively enhance the correlation between the results of the automatic evaluation metrics and that of the human judgments. Deep analysis further reveals that the information of the source sentences plays an important role in automatic evaluation of machine translation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "0 \u5f15 \u5f15 \u5f15\u8a00 \u8a00 \u8a00 \u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u662f\u673a\u5668\u7ffb\u8bd1\u7684\u91cd\u8981\u7ec4\u6210\u90e8\u5206\u3002\u5b83\u4e0d\u4ec5\u80fd\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u5ea6\u91cf\u7ffb\u8bd1\u7cfb\u7edf\u7684\u6574\u4f53 \u6027\u80fd\uff0c\u8fd8\u80fd\u5728\u7ffb\u8bd1\u7cfb\u7edf\u5f00\u53d1\u65f6\u6307\u5bfc\u5176\u7279\u5f81\u6743\u503c\u7684\u4f18\u5316\u3002\u56e0\u6b64\uff0c\u7814\u7a76\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u5bf9\u673a\u5668\u7ffb \u8bd1\u7684\u53d1\u5c55\u548c\u5e94\u7528\u6709\u7740\u91cd\u8981\u7684\u610f\u4e49\u3002 \u8fd1\u5e74\u6765\uff0c\u8bb8\u591a\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u88ab\u76f8\u7ee7\u63d0\u51fa\uff0c\u5b83\u4eec\u5c06\u673a\u5668\u7ffb\u8bd1\u7cfb\u7edf\u7684\u8f93\u51fa\u8bd1\u6587\u4e0e \u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8fdb\u884c\u5bf9\u6bd4\u6765\u5b9a\u91cf\u523b\u753b\u8bd1\u6587\u7684\u8d28\u91cf\u3002\u6839\u636e\u5bf9\u6bd4\u65f6\u6d89\u53ca\u7684\u8bed\u8a00\u77e5\u8bc6\u5c42\u6b21\uff0c\u5b83\u4eec\u5206 \u4e3a\u57fa\u4e8e\u8bcd\u8bed\u5339\u914d\u7684\u65b9\u6cd5\uff0c\u5982BLEU (Papineni et al., 2002) \u548cNIST (Doddington, 2002) \u7b49\uff1b\u57fa\u4e8e \u6d45\u5c42\u53e5\u6cd5\u7ed3\u6784\u5339\u914d\u7684\u65b9\u6cd5\uff0c\u5982POSBLEU (Popovi\u0107 and Ney, 2009) \u548cPOSF (Popovi\u0107 and Ney, 2009) \u7b49\uff1b\u57fa\u4e8e\u6df1\u5c42\u8bed\u4e49\u4fe1\u606f\u5339\u914d\u7684\u65b9\u6cd5\uff0c\u5982\u5f15\u5165\u590d\u8ff0\u7684\u6307\u6807Meteor Universal (Banerjee and Lavie, 2005) \u548cTERp (Snover et al., 2008) \u7b49\u3001\u5f15\u5165\u8bed\u4e49\u89d2\u8272\u6807\u6ce8\u7684\u6307\u6807MEANT (Lo, 2017) \u7b49 \u7b49\u3002\u968f\u7740\u6df1\u5ea6\u5b66\u4e60\u7684\u53d1\u5c55\u548c\u5176\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4e2d\u7684\u5e7f\u6cdb\u5e94\u7528\uff0c\u4e00\u4e9b\u7814\u7a76\u8005\u5229\u7528\u8bcd\u8bed\u6df1\u5ea6\u8868\u793a \u548c\u795e\u7ecf\u7f51\u7edc\u7ed3\u6784\u5bf9\u6bd4\u7ffb\u8bd1\u7cfb\u7edf\u8f93\u51fa\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8fdb\u884c\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\uff0c\u5982\u57fa\u4e8e\u9759\u6001\u8bcd\u5411 \u91cfword2vec (Mikolov et al., 2013) \u7684\u65b9\u6cd5 (Chen and Guo, 2015) \u3001\u57fa\u4e8e\u52a8\u6001\u8bcd\u5411\u91cfBERT (Devlin et al., 2018) \u7684\u65b9\u6cd5 (Mathur et al., 2019) \u3001\u548c\u57fa\u4e8e\u795e\u7ecf\u7f51\u7edc\u7ed3\u6784\u7684\u65b9\u6cd5ReVal (Gupta et al., 2015) \u548cRUSE (Shimanaka et al., 2018) \u7b49\u7b49\u3002 \u7136\u800c\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u8bc4\u4ef7\u673a\u5668\u8bd1\u6587\u7684\u4e3b\u8981\u601d\u8def\u8fd8\u662f\u9075\u5faaBLEU (Papineni et al., 2002) \u7684\u57fa\u672c\u89c2 \u70b9\uff1a\"\u673a\u5668\u8bd1\u6587\u8d8a\u63a5\u8fd1\u4e8e\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\uff0c\u5176\u8bd1\u6587\u8d28\u91cf\u8d8a\u9ad8\"\u3002\u4ece\u8fd9\u4e2a\u89c2\u70b9\u51fa\u53d1\uff0c\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u5373 \u662f\u8ba1\u7b97\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684\u76f8\u4f3c\u5ea6\u3002\u56e0\u6b64\uff0c\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u8fc7\u7a0b\u4e2d\u5b8c\u5168\u5ffd\u7565\u4e86\u6e90\u8bed\u8a00\u53e5 \u5b50\uff0c\u5373\u5728\u6ca1\u6709\u5bf9\u6e90\u8bed\u8a00\u53e5\u5b50\u5145\u5206\u5229\u7528\u7684\u57fa\u7840\u4e0a\u8fdb\u884c\u8bd1\u6587\u7684\u81ea\u52a8\u8bc4\u4ef7\u3002\u6240\u4ee5\uff0c\u627e\u5230\u7ed3\u5408\u6e90\u8bed\u8a00\u53e5 \u5b50\u8fdb\u884c\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u7684\u5207\u5165\u70b9\uff0c\u52bf\u5fc5\u80fd\u63d0\u9ad8\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4e0e\u4eba\u5de5\u8bc4\u4ef7\u7684\u76f8\u5173\u6027\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u5c1d \u8bd5\u5f15\u5165\u4ece\u6e90\u8bed\u8a00\u53e5\u5b50\u548c\u5176\u673a\u5668\u8bd1\u6587\u4e2d\u63d0\u53d6\u7684\u8d28\u91cf\u5411\u91cf(Quality Embedding, QE)\uff0c\u5e76\u5c06\u5176\u4e0e\u57fa \u4e8e\u8bed\u5883\u8bcd\u5411\u91cf\u7684\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5 (Mathur et al., 2019) \u8fdb\u884c\u6df1\u5ea6\u878d\u5408\u6765\u589e\u5f3a\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\uff0c\u63d0 \u9ad8\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4e0e\u4eba\u5de5\u8bc4\u4ef7\u7684\u76f8\u5173\u6027\u3002 1 \u76f8 \u76f8 \u76f8\u5173 \u5173 \u5173\u5de5 \u5de5 \u5de5\u4f5c \u4f5c \u4f5c \u5728\u57fa\u4e8e\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4e2d\uff0cLo(2017)\uff0c\u548cChen(2015)\u7b49\u4eba\u63d0\u51fa\u5229\u7528\u8bcd\u8bed \u7684\u5206\u5e03\u5f0f\u8868\u793a\uff0c\u9759\u6001\u9884\u8bad\u7ec3\u7684\u8bcd\u5411\u91cfword2vec (Mikolov et al., 2013) \uff0c\u6765\u63d0\u9ad8\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5 \u53c2\u8003\u8bd1\u6587\u5bf9\u6bd4\u65f6\u540c\u4e49\u8bcd\u3001\u8fd1\u4e49\u8bcd\u548c\u590d\u8ff0\u7b49\u5339\u914d\u7684\u51c6\u786e\u7387\u3002Guzm\u00e1n(2019)\u7b49\u4eba\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8e \u8bcd\u5411\u91cf\u548c\u795e\u7ecf\u7f51\u7edc\u7684\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\uff0c\u5176\u76ee\u6807\u662f\u5728\u7ed9\u5b9a\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684\u60c5\u51b5\u4e0b\uff0c\u4ece\u4e00 \u5bf9\u673a\u5668\u8bd1\u6587\u4e2d\u9009\u62e9\u6700\u4f73\u8bd1\u6587\uff0c\u4f7f\u7528\u795e\u7ecf\u7f51\u7edc\u53ef\u4ee5\u65b9\u4fbf\u5730\u878d\u5408\u7531\u8bcd\u5411\u91cf\u6355\u83b7\u7684\u4e30\u5bcc\u8bed\u6cd5\u548c\u8bed\u4e49 \u8868\u793a\u3002Gupta(2015)\u7b49\u4eba\u7528\u57fa\u4e8e\u6811\u7ed3\u6784\u7684\u957f\u77ed\u65f6\u8bb0\u5fc6\u7f51\u7edc (Tai et al., 2015) (Devlin et al., 2018) \u8bed\u5883\u8bcd\u5411\u91cf\u4f7f\u7528Bi-LSTM\u7f51\u7edc \u7ed3\u6784\u5b66\u4e60\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684\u53e5\u5b50\u8868\u793a\uff0c\u5e76\u5c06\u81ea\u7136\u8bed\u8a00\u63a8\u7406\u4e2d\u542f\u53d1\u5f0f\u65b9\u6cd5 (Mou et al., 2015) \u548c\u589e\u5f3a\u5e8f\u5217\u63a8\u7406\u6a21\u578b (Chen et al., 2016) (Mou et al., 2015) \u4ee5\u53caESIM\u65b9\u6cd5 (Chen et al., 2016) ", |
| "cite_spans": [ |
| { |
| "start": 206, |
| "end": 229, |
| "text": "(Papineni et al., 2002)", |
| "ref_id": "BIBREF19" |
| }, |
| { |
| "start": 236, |
| "end": 254, |
| "text": "(Doddington, 2002)", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 281, |
| "end": 304, |
| "text": "(Popovi\u0107 and Ney, 2009)", |
| "ref_id": "BIBREF20" |
| }, |
| { |
| "start": 311, |
| "end": 334, |
| "text": "(Popovi\u0107 and Ney, 2009)", |
| "ref_id": "BIBREF20" |
| }, |
| { |
| "start": 376, |
| "end": 402, |
| "text": "(Banerjee and Lavie, 2005)", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 409, |
| "end": 430, |
| "text": "(Snover et al., 2008)", |
| "ref_id": "BIBREF23" |
| }, |
| { |
| "start": 450, |
| "end": 460, |
| "text": "(Lo, 2017)", |
| "ref_id": "BIBREF13" |
| }, |
| { |
| "start": 555, |
| "end": 577, |
| "text": "(Mikolov et al., 2013)", |
| "ref_id": "BIBREF16" |
| }, |
| { |
| "start": 582, |
| "end": 602, |
| "text": "(Chen and Guo, 2015)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 616, |
| "end": 637, |
| "text": "(Devlin et al., 2018)", |
| "ref_id": "BIBREF6" |
| }, |
| { |
| "start": 642, |
| "end": 663, |
| "text": "(Mathur et al., 2019)", |
| "ref_id": "BIBREF15" |
| }, |
| { |
| "start": 683, |
| "end": 703, |
| "text": "(Gupta et al., 2015)", |
| "ref_id": "BIBREF9" |
| }, |
| { |
| "start": 710, |
| "end": 734, |
| "text": "(Shimanaka et al., 2018)", |
| "ref_id": "BIBREF22" |
| }, |
| { |
| "start": 766, |
| "end": 789, |
| "text": "(Papineni et al., 2002)", |
| "ref_id": "BIBREF19" |
| }, |
| { |
| "start": 1032, |
| "end": 1053, |
| "text": "(Mathur et al., 2019)", |
| "ref_id": "BIBREF15" |
| }, |
| { |
| "start": 1187, |
| "end": 1209, |
| "text": "(Mikolov et al., 2013)", |
| "ref_id": "BIBREF16" |
| }, |
| { |
| "start": 1382, |
| "end": 1400, |
| "text": "(Tai et al., 2015)", |
| "ref_id": "BIBREF26" |
| }, |
| { |
| "start": 1401, |
| "end": 1422, |
| "text": "(Devlin et al., 2018)", |
| "ref_id": "BIBREF6" |
| }, |
| { |
| "start": 1476, |
| "end": 1494, |
| "text": "(Mou et al., 2015)", |
| "ref_id": "BIBREF17" |
| }, |
| { |
| "start": 1505, |
| "end": 1524, |
| "text": "(Chen et al., 2016)", |
| "ref_id": "BIBREF5" |
| }, |
| { |
| "start": 1525, |
| "end": 1543, |
| "text": "(Mou et al., 2015)", |
| "ref_id": "BIBREF17" |
| }, |
| { |
| "start": 1553, |
| "end": 1572, |
| "text": "(Chen et al., 2016)", |
| "ref_id": "BIBREF5" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "(Long Short-Term Memory network, LSTM)\u5bf9\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8fdb\u884c\u7f16\u7801\uff0c\u6839\u636e\u4e24\u8005\u4e4b\u95f4\u5143\u7d20\u5dee\u5f02\u548c \u5939\u89d2\u8ba1\u7b97\u673a\u5668\u8bd1\u6587\u7684\u8d28\u91cf\u5f97\u5206\u3002Shimanaka(2018) \u7b49\u4eba\u4f7f\u7528\u53cc\u5411LSTM(Bidirectional LSTM, Bi-LSTM)\u5bf9\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8fdb\u884c\u7f16\u7801\uff0c\u5e76\u5229\u7528\u591a\u5c42\u611f\u77e5\u673a\u56de\u5f52\u6a21\u578b\u8ba1\u7b97\u673a\u5668\u8bd1\u6587 \u7684\u8d28\u91cf\u5f97\u5206\u3002Mathur(2019)\u7b49\u4eba\u57fa\u4e8eBERT", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "(Enhanced Sequential Inference Model, ESIM) \u5f15\u5165\u5230\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4e2d\uff0c\u8be5\u65b9\u6cd5\u5728WMT'19\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4efb\u52a1(Metrics Task)\u4e0a\u53d6\u5f97\u4e86\u4f18 \u5f02\u7684\u6210\u7ee9\uff0c\u56e0\u6b64\uff0c\u672c\u6587\u5c06\u5728Mathur(2019)\u7b49\u4eba\u7684\u5de5\u4f5c\u57fa\u7840\u4e0a\uff0c\u5c06\u5229\u7528\u6e90\u8bed\u8a00\u53e5\u5b50\u63d0\u53d6\u7684\u8d28\u91cf\u5411 \u91cf\u878d\u5165\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4e2d\uff0c\u8fdb\u4e00\u6b65\u589e\u5f3a\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u7684\u6027\u80fd\u3002 2 \u80cc \u80cc \u80cc\u666f \u666f \u666f\u77e5 \u77e5 \u77e5\u8bc6 \u8bc6 \u8bc6 2.1 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u8bed \u8bed \u8bed\u5883 \u5883 \u5883\u8bcd \u8bcd \u8bcd\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u7684 \u7684 \u7684\u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u81ea \u81ea \u81ea\u52a8 \u52a8 \u52a8\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7 \u81ea\u7136\u8bed\u8a00\u63a8\u65ad\u5173\u6ce8\u5047\u8bbe\u7ed3\u8bba(hypothesis)\u662f\u5426\u53ef\u4ee5\u4ece\u524d\u63d0\u8bed\u53e5(premise)\u4e2d\u63a8\u65ad\u83b7\u53d6\uff0c \u5b83\u4e0e\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4efb\u52a1\u975e\u5e38\u7c7b\u4f3c\u3002\u8bd1\u6587\u7684\u8d28\u91cf\u8d8a\u597d\uff0c\u673a\u5668\u8bd1\u6587\u88ab\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8868\u793a(\u63a8\u65ad)\u7684 \u7a0b\u5ea6\u8d8a\u9ad8\uff0c\u540c\u65f6\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u88ab\u673a\u5668\u8bd1\u6587\u8868\u793a(\u63a8\u65ad)\u7684\u7a0b\u5ea6\u4e5f\u8d8a\u9ad8\uff1b\u53cd\u4e4b\u4ea6\u7136\u3002\u5728\u81ea\u7136\u8bed \u8a00\u63a8\u65ad\u7684\u6846\u67b6\u4e0b\uff0cMathur(2019)\u7b49\u4eba\u4f7f\u7528\u8bed\u5883\u8bcd\u5411\u91cf\u5206\u522b\u8868\u793a\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\uff0c\u5e76\u6839 \u636e\u4e24\u4e2a\u8868\u793a\u7684\u4ea4\u4e92\u7a0b\u5ea6\u6765\u5ea6\u91cf\u673a\u5668\u8bd1\u6587\u7684\u8d28\u91cf\u3002\u4f7f\u7528\u81ea\u7136\u8bed\u8a00\u63a8\u65ad\u4e2d\u542f\u53d1\u5f0f\u65b9\u6cd5", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\uff0cMathur\u7b49\u4eba\u5206\u522b\u63d0\u51fa\u4e86(Bi-LSTM+attention) BERT \u8bd1 \u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u548c(ESIM) BERT \u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u3002 2.1.1 (Bi-LSTM+attention) BERT \u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u81ea \u81ea \u81ea\u52a8 \u52a8 \u52a8\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 \u5c06\u957f\u5ea6\u4e3al r \u7684\u4eba\u5de5\u53c2\u8003\u8bd1\u6587r\u548c\u957f\u5ea6\u4e3al t \u7684\u673a\u5668\u8bd1\u6587t\u5206\u522b\u5229\u7528BERT\u8bed\u5883\u8bcd\u5411\u91cf\u8fdb\u884c\u8868\u793a\uff0c \u4f7f\u7528Bi-LSTM\u7f51\u7edc\u5bf9\u5176\u8fdb\u884c\u7f16\u7801\u5f97\u5230\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u548c\u673a\u5668\u8bd1\u6587\u5305\u542b\u4e0a\u4e0b\u6587\u542b\u4e49\u7684\u65b0\u7684\u5411\u91cf\u8868 \u793ah r 1:x (x = 1. . . l r )\u3001h t 1:y (y = 1. . . l t )\uff0c\u901a\u8fc7\u5411\u91cf\u70b9\u79ef\u6c42\u5f97\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u548c\u673a\u5668\u8bd1\u6587\u7684\u76f8\u4f3c\u5ea6\u77e9 \u9635A\uff0cA\u4e2d\u5143\u7d20a i,j = h T r i h t j \uff0c\u5229\u7528\u76f8\u4f3c\u5ea6\u77e9\u9635A\uff0c\u7ed3\u5408h r \u548ch t \uff0c\u8ba1\u7b97\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u548c\u673a\u5668\u8bd1\u6587\u7684 \u76f8\u4e92\u8868\u793a\uff1ah t = lr i=1 exp(a i,j ) j exp(a i,j) \u2022 h r (1) h r = lt j=1 exp(a i,j ) i exp(a i,j) \u2022 h t (2)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5176\u4e2d\u7b26\u53f7h t \u8868\u793ah r \u4e2d\u6bcf\u4e2a\u8bcd\u4e0eh t \u7684\u76f8\u5173\u7a0b\u5ea6\uff0ch r \u8868\u793ah t \u4e2d\u6bcf\u4e2a\u8bcd\u4e0eh r \u7684\u76f8\u5173\u7a0b\u5ea6\u3002 \u4e3a\u4e86\u907f\u514d\u5411\u91cfh t \u548ch r \u7b80\u5355\u6c42\u548c\u5bb9\u6613\u5bfc\u81f4\u7ed3\u679c\u5bf9\u5e8f\u5217\u957f\u5ea6\u654f\u611f\u7684\u95ee\u9898 (Chen et al., 2016 )\uff0c\u5bf9 \u5411\u91cfh t \u548ch r \u5206\u522b\u8fdb\u884c\u6700\u5927\u6c60\u5316\u548c\u5e73\u5747\u6c60\u5316\uff0c\u5c06\u6c60\u5316\u7ed3\u679c\u5206\u522b\u62fc\u63a5\u5f97\u5230\u5411\u91cfv t \u548cv r \uff0c\u5e76\u4e14\u542f\u53d1\u5f0f \u65b9\u6cd5 (Mou et al., 2015) \u88ab\u7528\u4f5c\u5bf9\u5c40\u90e8\u63a8\u7406\u8fdb\u884c\u589e\u5f3a\u5f97\u5230\u589e\u5f3a\u540e\u7684\u8868\u793a\u5411\u91cfm:", |
| "cite_spans": [ |
| { |
| "start": 91, |
| "end": 109, |
| "text": "(Chen et al., 2016", |
| "ref_id": "BIBREF5" |
| }, |
| { |
| "start": 171, |
| "end": 189, |
| "text": "(Mou et al., 2015)", |
| "ref_id": "BIBREF17" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "m = [v t \u2295 v r \u2295 (v t v r ) \u2295 (v t \u2212 v r )]", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "(3)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5176\u4e2d\u7b26\u53f7\"\u2295\"\u8868\u793a\u5411\u91cf\u62fc\u63a5\u64cd\u4f5c\uff1b\u7b26\u53f7\" \"\u8868\u793a\u4e24\u4e2a\u5411\u91cf\u9010\u5143\u7d20\u76f8\u4e58\u64cd\u4f5c\u3002\u6700\u540e\u5411\u91cfm\u88ab\u4f5c \u4e3a\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u7684\u8f93\u5165\u7528\u4e8e\u9884\u6d4b\u673a\u5668\u8bd1\u6587\u88ab\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8868\u793a\u7684\u7a0b\u5ea6\uff0c\u5373\u8bd1\u6587\u8d28\u91cf\u7684\u5f97\u5206\u3002 2.1.2 (ESIM) BERT \u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u81ea \u81ea \u81ea\u52a8 \u52a8 \u52a8\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 ESIM\u65b9\u6cd5\u5229\u7528\u5f0f(4)\u548c(5)\u8ba1\u7b97\u673a\u5668\u8bd1\u6587\u88ab\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8868\u793a\u7684\u589e\u5f3a\u5411\u91cfm t \u548c\u4eba\u5de5\u53c2\u8003 \u8bd1\u6587\u88ab\u673a\u5668\u8bd1\u6587\u8868\u793a\u7684\u589e\u5f3a\u5411\u91cfm r \u3002\u4e3a\u964d\u4f4e\u6a21\u578b\u53c2\u6570\u590d\u6742\u6027\uff0c\u5229\u7528\u4e00\u4e2a\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u5c42 \u5c06m t \u548cm r \u8f6c\u6362\u81f3\u6a21\u578b\u7684\u7ef4\u5ea6\u3002Bi-LSTM\u7f51\u7edc\u88ab\u7528\u4f5c\u5bf9\u964d\u7ef4\u540e\u7684\u4fe1\u606f\u8fdb\u884c\u7f16\u7801\uff0c\u4ee5\u4fbf\u5f97\u5230\u5176 \u5c40\u90e8\u4fe1\u606f\u7684\u4e0a\u4e0b\u6587\u8868\u793a\u5411\u91cf\u3002\u5c06\u7f16\u7801\u540e\u7684\u5411\u91cf\u8fdb\u884c\u5e73\u5747\u6c60\u5316\u548c\u6700\u5927\u6c60\u5316\uff0c\u5e76\u5c06\u6c60\u5316\u540e\u7684\u7ed3 \u679cv r,avg \u3001v r,max \u548cv t,avg \u3001v t,max \u8fdb\u884c\u62fc\u63a5\uff0c\u5f62\u6210\u56fa\u5b9a\u957f\u5ea6\u5411\u91cfp\uff0c\u5373\uff1a m t = [h t \u2295h t \u2295 (h t h t ) \u2295 (h t \u2212h t )]", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "(4) (Bahdanau et al., 2014 )\u7684\u65b9 \u6cd5 (Kim et al., 2017; Li et al., 2018) \u548c\u57fa\u4e8eTransformer\u6a21\u578b (Vaswani et al., 2017) \u7684\u65b9\u6cd5 (Fan et al., 2019; ", |
| "cite_spans": [ |
| { |
| "start": 4, |
| "end": 26, |
| "text": "(Bahdanau et al., 2014", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 33, |
| "end": 51, |
| "text": "(Kim et al., 2017;", |
| "ref_id": "BIBREF11" |
| }, |
| { |
| "start": 52, |
| "end": 68, |
| "text": "Li et al., 2018)", |
| "ref_id": "BIBREF12" |
| }, |
| { |
| "start": 86, |
| "end": 108, |
| "text": "(Vaswani et al., 2017)", |
| "ref_id": "BIBREF27" |
| }, |
| { |
| "start": 113, |
| "end": 131, |
| "text": "(Fan et al., 2019;", |
| "ref_id": "BIBREF8" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "m r = [h r \u2295h r \u2295 (h r h r ) \u2295 (h r \u2212h r )] (5) p = [v r,avg \u2295 v r,max \u2295 v t,avg \u2295 v t,max ] (6) \u6700\u540e\u5411\u91cfp\u88ab\u4f5c\u4e3a\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u7684\u8f93\u5165\u7528\u4e8e\u9884\u6d4b\u673a\u5668\u8bd1\u6587\u8d28\u91cf\u7684\u5f97\u5206\u3002 2.2 \u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u8d28 \u8d28 \u8d28\u91cf \u91cf \u91cf\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u63d0 \u63d0 \u63d0\u53d6 \u53d6 \u53d6\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 \u8bd1\u6587\u8d28\u91cf\u5411\u91cf\u662f\u8bd1\u6587\u8d28\u91cf\u4f30\u8ba1\u4e2d\u63cf\u8ff0\u7ffb\u8bd1\u8d28\u91cf\u7684\u5411\u91cf\uff0c\u5b83\u4ece\u6e90\u8bed\u8a00\u53e5\u5b50\u548c\u5176\u76f8\u5e94\u7684\u8bd1\u6587\u4e2d \u62bd\u53d6\uff0c\u5b8c\u5168\u4e0d\u9700\u8981\u501f\u52a9\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u8fdb\u884c\u8ba1\u7b97\u3002\u76ee\u524d\u4e3b\u6d41\u7684\u8d28\u91cf\u5411\u91cf\u63d0\u53d6\u65b9\u6cd5\u5305\u62ec\u57fa\u4e8e\u5faa\u73af\u795e \u7ecf\u7f51\u7edc(Recurrent Neural Network, RNN)\u7684\u7f16\u7801\u5668-\u89e3\u7801\u5668\u6a21\u578b", |
| "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": "3 \u7ed3 \u7ed3 \u7ed3\u5408 \u5408 \u5408\u8d28 \u8d28 \u8d28\u91cf \u91cf \u91cf\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u7684 \u7684 \u7684\u673a \u673a \u673a\u5668 \u5668 \u5668\u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u81ea \u81ea \u81ea\u52a8 \u52a8 \u52a8\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7 \u4e3a\u4e86\u628a\u6e90\u8bed\u8a00\u53e5\u5b50\u4fe1\u606f\u5f15\u5165\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4e2d\uff0c\u6211\u4eec\u4ee5\u8d28\u91cf\u5411\u91cf\u4f5c\u4e3a\u5207\u5165\u70b9\uff0c\u5c06\u7ed9\u5b9a\u6e90\u8bed\u8a00 \u53e5\u5b50\u60c5\u51b5\u4e0b\u673a\u5668\u8bd1\u6587\u8d28\u91cf\u7684\u8868\u793a\u548c\u7ed9\u5b9a\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u60c5\u51b5\u4e0b\u673a\u5668\u8bd1\u6587\u7684\u589e\u5f3a\u8868\u793a\u8fdb\u884c\u878d\u5408\u3002\u6a21 \u578b\u7ed3\u6784\u5982\u56fe1\u6240\u793a\uff0c\u5176\u4e2d\u7b26\u53f7src\u3001mt\u548cref \u5206\u522b\u8868\u793a\u6e90\u8bed\u8a00\u53e5\u5b50\u3001\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u3002 \u56fe\u5de6\u8fb9\u63cf\u8ff0\u901a\u8fc7UNQE\u65b9\u6cd5(Li et al., 2018)\u4ece\u6e90\u8bed\u8a00\u53e5\u5b50\u548c\u5176\u673a\u5668\u8bd1\u6587\u4e2d\u63d0\u53d6\u51fa\u63cf\u8ff0\u7ffb\u8bd1\u8d28 \u91cf\u7684\u8bcd\u8bed\u7ea7\u8d28\u91cf\u5411\u91cf\uff0c\u5e76\u5c06\u5176\u5229\u7528Bi-LSTM\u7f51\u7edc\u5904\u7406\u6210\u53e5\u5b50\u7ea7\u522b\u7684\u8d28\u91cf\u5411\u91cf\uff1b\u56fe\u53f3\u8fb9\u63cf\u8ff0\u901a \u8fc7(Bi-LSTM+attention) BERT \u6216(ESIM) BERT \u65b9\u6cd5(Mathur et al., 2019)\u5c06\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1 \u6587\u62bd\u8c61\u4e3a\u4ea4\u4e92\u8868\u793a\u7684\u589e\u5f3a\u5411\u91cf\uff0c\u56fe\u4e0a\u8868\u793a\u5c06\u8d28\u91cf\u5411\u91cf\u4e0e\u4ea4\u4e92\u8868\u793a\u7684\u589e\u5f3a\u5411\u91cf\u8fdb\u884c\u62fc\u63a5\uff0c\u5c06\u62fc\u63a5 \u540e\u7684\u5411\u91cf\u8f93\u5165\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u4ee5\u9884\u6d4b\u673a\u5668\u8bd1\u6587\u8d28\u91cf\u5f97\u5206\u3002 \u56fe 1. \u5f15\u5165\u8bd1\u6587\u8d28\u91cf\u5411\u91cf\u589e\u5f3a\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u7684\u6a21\u578b\u67b6\u6784 3.1 (Bi-LSTM+attention) BERT+QE \u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u81ea \u81ea \u81ea\u52a8 \u52a8 \u52a8\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 \u7531 \u4e8e \u4ece \u6e90 \u8bed \u8a00 \u53e5 \u5b50 \u548c \u673a \u5668 \u8bd1 \u6587 \u4e2d \u62bd \u53d6 \u7684 \u8d28 \u91cf \u5411 \u91cf \u662f \u8bcd \u8bed \u7ea7 \u7684 \uff0c \u5373 \u673a \u5668 \u8bd1 \u6587 \u4e2d \u6bcf \u4e2a \u8bcd (token)\u4f7f\u7528\u4e00\u4e2a\u5b9e\u6570\u5411\u91cf\u63cf\u8ff0\u5176\u7ffb\u8bd1\u8d28\u91cf\uff0c\u800c\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684\u4ea4\u4e92\u8868\u793a\u589e\u5f3a \u5411\u91cf\u662f\u53e5\u5b50\u7ea7\uff0c\u4e3a\u4e86\u5728\u540c\u4e00\u5c42\u6b21\u5c06\u4e8c\u8005\u8fdb\u884c\u878d\u5408\uff0c\u9700\u8981\u5c06\u8d28\u91cf\u5411\u91cf\u8fdb\u4e00\u6b65\u62bd\u8c61\u6210\u53e5\u5b50\u7ea7\u522b\u8868 \u793a\u3002Bi-LSTM\u7f51\u7edc\u88ab\u7528\u6765\u5bf9\u8bcd\u8bed\u7ea7\u8d28\u91cf\u5411\u91cfe qe 1:k (k = 1. . . l t )\u8fdb\u884c\u7f16\u7801\uff0c\u5f97\u5230e qe 1:k \u7684\u5305\u542b\u4e0a\u4e0b \u6587\u4fe1\u606f\u7684\u5411\u91cfh qe,k (k = 1. . . l t )\uff0c\u901a\u8fc7\u5bf9h qe \u8fdb\u884c\u6700\u5927\u6c60\u5316\u548c\u5e73\u5747\u6c60\u5316\u5904\u7406\uff0c\u5c06\u6c60\u5316\u540e\u7684\u7ed3\u679c\u62fc\u63a5 \u5373\u5f97\u5230\u4e86\u53e5\u5b50\u7684\u8d28\u91cf\u5411\u91cf\u8868\u793av qe \uff1a h qe,k = Bi-LSTM (e qe , k) , \u2200k \u2208 [1, . . . , l t ] (7) v qe,max = lt max k=1 h qe,k , v qe,avg = 1 l t lt k=1 h qe,k", |
| "eq_num": "(8)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "v qe = [v qe,avg \u2295 v qe,max ]", |
| "eq_num": "(9)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5176\u4e2d\u7b26\u53f7v qe,avg \u8868\u793a\u5bf9h qe \u8fdb\u884c\u5e73\u5747\u6c60\u5316\u540e\u7684\u7ed3\u679c\uff0cv qe,max \u8868\u793a\u5bf9h qe \u8fdb\u884c\u6700\u5927\u6c60\u5316\u540e\u7684\u7ed3 \u679c\uff0ck\u8868\u793a\u53e5\u5b50\u4e2d\u7684\u8bcd\u5e8f\u53f7\u3002", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5728\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684\u4ea4\u4e92\u8868\u793a\u589e\u5f3a\u5411\u91cf\u65b9\u9762\uff0cBi-LSTM\u7f51\u7edc\u88ab\u7528\u6765\u5bf9\u4eba\u5de5\u53c2\u8003\u8bd1 \u6587\u548c\u673a\u5668\u8bd1\u6587\u7684\u8bed\u5883\u8bcd\u5411\u91cf\u7f16\u7801\uff0c\u5229\u7528\u5f0f(1)-(2)\u6c42\u5f97\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u548c\u673a\u5668\u8bd1\u6587\u7684\u76f8\u4e92\u8868\u793a\uff0c\u968f \u540e\u5229\u7528\u5f0f(8)\u7684\u6c60\u5316\u64cd\u4f5c\u548c\u5f0f(9)\u7684\u62fc\u63a5\u64cd\u4f5c\u6c42\u5f97\u4e86\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u53e5\u5b50\u8868\u793av r \u548c\u673a\u5668\u8bd1\u6587\u53e5\u5b50\u8868 \u793av t \u3002 \u4e3a\u4e86\u5c06\u6e90\u7aef\u4fe1\u606f\u6709\u6548\u5730\u5f15\u5165\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u6a21\u578b\u4e2d\uff0c\u6211\u4eec\u5c06v r \u548cv t \u8fdb\u884c\u5c40\u90e8\u4fe1\u606f\u589e\u5f3a\u7ec4 \u5408\uff0c\u540c\u65f6\u5c06\u589e\u5f3a\u540e\u7684\u4fe1\u606f\u4e0e\u5f0f(9)\u5904\u7406\u540e\u7684\u53e5\u5b50\u7ea7\u522b\u8d28\u91cf\u5411\u91cfv qe \u62fc\u63a5\u8d77\u6765\u5f62\u6210\u65b0\u7684\u56fa\u5b9a\u957f\u5ea6\u5411 \u91cfm\uff1am = [v t \u2295 v r \u2295 (v t v r ) \u2295 (v t \u2212 v r ) \u2295 v qe ] (10) \u5176 \u4e2d \u7b26 \u53f7v t \u662fh t \u7684 \u5e73 \u5747 \u6c60 \u5316 \u540e \u5411 \u91cfv t,avg \u548c \u6700 \u5927 \u6c60 \u5316 \u540e \u5411 \u91cfv t,max \u62fc \u63a5 \u540e \u5f62 \u6210 \u7684 \u5411 \u91cf \uff1bv r \u662fh r \u7684 \u5e73 \u5747 \u6c60 \u5316 \u540e \u5411 \u91cfv r,avg \u548c \u6700 \u5927 \u6c60 \u5316 \u540e \u5411 \u91cfv r,max \u62fc \u63a5 \u540e \u5f62 \u6210 \u7684 \u5411 \u91cf \u3002 \u6700 \u540e \u5c06 \u5411 \u91cfm\u4f5c\u4e3a\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u7684\u8f93\u5165\uff0c\u4f7f\u7528\u5176\u9884\u6d4b\u8bd1\u6587\u7684\u8d28\u91cf\u5f97\u5206\uff1a y score = w T ReLU W Tm + b + b (11) \u5176\u4e2d\u53c2\u6570w\uff0cW \uff0cb\uff0cb \u5747\u4e3a\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u7684\u6743\u503c\u3002 \u4e3a\u4e86\u8bad\u7ec3\u6a21\u578b\u7684\u6240\u6709\u53c2\u6570\uff0c\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u5f97\u5206y score \u4e0e\u4eba\u5de5\u8bc4\u4ef7\u5f97\u5206h\u7684\u5747\u65b9\u8bef\u5dee\u88ab\u7528\u6765\u5bf9 \u6a21\u578b\u8fdb\u884c\u4f18\u5316\uff0c\u4f18\u5316\u76ee\u6807\u6b63\u5f0f\u63cf\u8ff0\u4e3a\uff1a loss = 1 M M i=1 (y (i) score \u2212 h (i) ) 2 (12) \u5176\u4e2dy (i)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "score \u4e3a\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u5bf9\u5f85\u8bc4\u4ef7\u673a\u5668\u8bd1\u6587\u7684\u6253\u5206\uff0ch (i) \u4e3a\u4eba\u5de5\u8bc4\u4ef7\u7ed3\u679c\uff0cM \u4e3a\u8bad\u7ec3\u96c6\u5305 \u542b\u7684\u6837\u672c\u6570\u91cf\u3002 3.2 (ESIM) BERT+QE \u8bd1 \u8bd1 \u8bd1\u6587 \u6587 \u6587\u81ea \u81ea \u81ea\u52a8 \u52a8 \u52a8\u8bc4 \u8bc4 \u8bc4\u4ef7 \u4ef7 \u4ef7\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 \u4e3a\u4e86\u63a7\u5236\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u6a21\u578b\u7684\u590d\u6742\u6027\uff0c\u5c06\u5bf9\u5f0f(4)\u548c(5)\u5f97\u5230\u7684\u673a\u5668\u8bd1\u6587\u548c\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684\u5c40 \u90e8\u4fe1\u606f\u8868\u793am t \u3001m r \u4f7f\u7528\u4e00\u4e2a\u6620\u5c04F \u8f6c\u6362\u81f3\u6a21\u578b\u7684\u7ef4\u5ea6\u540e\uff0c\u7ecf\u8fc7Bi-LSTM\u8fdb\u884c\u7f16\u7801\uff0c\u4ee5\u5f97\u5230\u5176\u5c40 \u90e8\u4fe1\u606f\u7684\u4e0a\u4e0b\u6587\u8868\u793a\u5411\u91cfm t \u548cm r \uff0c\u5982\u5f0f(13)-(14)\u6240\u793a\u3002\u4e3a\u4e86\u5f15\u5165\u6e90\u7aef\u4fe1\u606f\u589e\u5f3a\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4 \u4ef7\uff0c\u6211\u4eec\u5c06m t \u548cm r \u5e73\u5747\u6c60\u5316\u548c\u6700\u5927\u6c60\u5316\u540e\u7684\u5411\u91cf\u4e0e\u673a\u5668\u8bd1\u6587\u8d28\u91cf\u4f30\u8ba1\u7684v qe,avg \u548cv qe,max \u5411\u91cf\u62fc \u63a5\u5f97\u5230\u65b0\u7684\u4fe1\u606f\u7ec4\u5408\u5411\u91cfp\u3002\u5c06\u62fc\u63a5\u540e\u7684\u4fe1\u606f\u8868\u793a\u5411\u91cf\u4f5c\u4e3a\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u7684\u8f93\u5165\u4ee5\u9884\u6d4b\u673a\u5668\u8bd1\u6587 \u7684\u8d28\u91cf\u5206\u6570\uff1am (Ma et al., 2019) \u7684\u5fb7\u82f1\u4efb\u52a1\u3001\u4e2d\u82f1\u4efb\u52a1\u548c\u82f1\u4e2d\u4efb\u52a1\u4e0a\u8fdb\u884c\u5b9e\u9a8c\u3002\u4e3a\u4e86\u6bd4\u8f83\u4e0d\u540c\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u7684 \u6027\u80fd\uff0c\u6211\u4eec\u9075\u5faaWMT\u8bc4\u6d4b\u5b98\u65b9\u7684\u505a\u6cd5\u5229\u7528\u76ae\u5c14\u68ee\u76f8\u5173\u7cfb\u6570\u4e0e\u80af\u5fb7\u5c14\u76f8\u5173\u7cfb\u6570\u5206\u522b\u8ba1\u7b97\u81ea\u52a8\u8bc4 \u4ef7\u7ed3\u679c\u548c\u4eba\u5de5\u8bc4\u4ef7\u7ed3\u679c\u7684\u7cfb\u7edf\u7ea7\u522b\u76f8\u5173\u6027\u548c\u53e5\u5b50\u7ea7\u522b\u76f8\u5173\u6027\uff0c\u76ae\u5c14\u68ee\u76f8\u5173\u7cfb\u6570\u6216\u80af\u5fb7\u5c14\u76f8\u5173\u7cfb \u6570\u8d8a\u5927\uff0c\u76f8\u5173\u6027\u8d8a\u597d\u3002 UNQE\u63d0\u53d6\u7684\u4e2d\u82f1\u3001\u82f1\u4e2d\u4efb\u52a1\u4e0a\u7684\u8d28\u91cf\u5411\u91cf\u7ef4\u5ea6\u4e3a700\uff0c\u5fb7\u82f1\u4efb\u52a1\u4e0a\u8d28\u91cf\u5411\u91cf\u7ef4\u5ea6\u4e3a500\u3002 \u6a21\u578b\u4e2dBi-LSTM\u9690\u85cf\u5c42\u72b6\u6001\u7ef4\u5ea6\u5747\u56fa\u5b9a\u4e3a300\uff0cDropout\u8bbe\u7f6e\u4e3a0.2\uff0c\u4f7f\u7528Adam\u4f18\u5316\u5668\u4f18\u5316\u8bad \u7ec3\uff0c\u521d\u59cb\u5b66\u4e60\u7387\u4e3a0.0004\uff0c\u8bad\u7ec3\u6279\u6b21\u5927\u5c0f\u4e3a32\uff0c\u4f7f\u7528\"bert-base-uncased\"\u63d0\u53d6\u82f1\u6587\u53e5\u5b50\u8bed\u5883\u8bcd \u5411\u91cf\uff0c\u4f7f\u7528\"bert-base-chinese\"\u63d0\u53d6\u4e2d\u6587\u53e5\u5b50\u8bed\u5883\u8bcd\u5411\u91cf\u3002 \u5728\u5b9e\u9a8c\u4e2d\uff0c\u6211\u4eec\u4e0d\u4ec5\u5c06\u672c\u6587\u63d0\u51fa\u7684\u65b9\u6cd5\u4e0eBLEU (Papineni et al., 2002) \u3001chrF (Popovi\u0107, 2015) \u4ee5 \u53caBEER (Stanojevi\u0107 and Sima'an, 2014) (Li et al., 2018) \u8fdb\u884c\u4e86\u5bf9\u6bd4\u3002\u9700\u8981\u8bf4\u660e\u7684\u662fMathur\u7b49\u4eba\u662f\u6df7\u5408\u6240\u6709\u76f8\u540c\u76ee\u6807\u8bed\u8a00 (\u6bd4\u5982\u5fb7\u82f1\u548c\u4e2d\u82f1)\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u8bad\u7ec3\u96c6\u8bed\u6599\u8fdb\u884c\u6a21\u578b\u8bad\u7ec3\uff0c\u800c\u6211\u4eec\u5f15\u5165\u4e86\u6e90\u7aef\u4fe1\u606f\uff0c\u8003\u8651\u5b9e \u9645\u8bd1\u6587\u6253\u5206\u9700\u6c42\u4e14\u907f\u514d\u53d7\u4e0d\u540c\u6e90\u8bed\u8a00\u5dee\u5f02\u6027\u7684\u8d1f\u9762\u5f71\u54cd\uff0c\u6211\u4eec\u9488\u5bf9\u6bcf\u4e2a\u8bed\u8a00\u5bf9\u5229\u7528\u5176\u8bad\u7ec3\u96c6\u6570 \u636e\u5355\u72ec\u8bad\u7ec3\u6a21\u578b\u3002\u5fb7\u82f1\u8bed\u8a00\u5bf9\u4f7f\u7528\u7684\u662fWMT'15-17 Metrics task (Bojar et al., 2015; Bojar et al., 2016; Ondrej et al., 2017 \u88683\u7684\u6570\u636e\u8868\u660e\u5f15\u5165\u6e90\u8bed\u8a00\u53e5\u5b50\u4fe1\u606f\u7684\u65b9\u6cd5\"(Bi-LSTM+attention) BERT+QE \"\u548c\"(ESIM) \u88684\u7684 \u6570 \u636e \u8868 \u660e \u672c \u6587 \u6240 \u63d0 \u65b9 \u6cd5\"(Bi-LSTM+attention) BERT+QE \"\u548c\"(ESIM) BERT+QE \"\u5728 \u5fb7 \u82f1\u3001\u4e2d\u82f1\u548c\u82f1\u4e2d\u4e09\u4e2a\u8bed\u8a00\u5bf9\u8bc4\u6d4b\u4efb\u52a1\u4e0a\uff0c\u4e0e\u4eba\u5de5\u8bc4\u4ef7\u7684\u7cfb\u7edf\u7ea7\u522b\u76f8\u5173\u7cfb\u6570\u7684\u5747\u503c\u5206\u522b\u9ad8 \u4e8e\"(Bi-LSTM+attention) BERT \"\u548c\"(ESIM) BERT \"\u3002\"(Bi-LSTM+attention) BERT+QE \"\u76f8\u5bf9\u4e8e\"(Bi-LSTM+attention) BERT \"\u65b9\u6cd5\u5728\u5fb7\u82f1\u3001\u4e2d\u82f1\u4efb\u52a1\u4e0a\u63d0\u5347\u4e860.8%\u548c1.7%\uff0c\u5728\u82f1\u4e2d\u4efb\u52a1\u4e0a\u4fdd\u6301\u4e00 \u81f4\uff0c\"(ESIM) BERT+QE \"\u76f8\u5bf9\u4e8e\"(ESIM) BERT \"\u65b9\u6cd5\u5728\u4e2d\u82f1\u3001\u82f1\u4e2d\u4efb\u52a1\u4e0a\u5206\u522b\u63d0\u5347\u4e860.7%\u548c0.3%\uff0c \u5728\u5fb7\u82f1\u4e0a\u4fdd\u6301\u4e00\u81f4\u3002\u8fd9\u8bf4\u660e\u5f15\u5165\u6e90\u7aef\u4fe1\u606f\u80fd\u589e\u5f3a\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u4e0e\u4eba\u5de5\u8bc4\u4ef7\u7684\u7cfb\u7edf\u7ea7\u522b\u76f8\u5173 \u6027\u3002 \u4ee4\u4eba\u60ca\u5947\u7684\u662f\u4ec5\u4f7f\u7528\u6e90\u7aef\u4fe1\u606f\uff0c\u5b8c\u5168\u4e0d\u4f7f\u7528\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684UNQE\u65b9\u6cd5\u4e5f\u4e0e\u4eba\u5de5\u8bc4\u4ef7 \u7ed3\u679c\u6709\u8f83\u597d\u7684\u76f8\u5173\u6027\u3002\u5c3d\u7ba1\u5176\u5728\u5e73\u5747\u76f8\u5173\u6027\u4e0a\u52a3\u4e8e\u6240\u6709\u4f7f\u7528\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u7684\u65b9\u6cd5\uff0c\u4f46\u662f\u5b83 \u4e0esentBLEU\u65b9\u6cd5\u5728\u5e73\u5747\u53e5\u5b50\u7ea7\u522b\u76f8\u5173\u6027\u548c\u5e73\u5747\u7cfb\u7edf\u7ea7\u522b\u76f8\u5173\u6027\u4e0a\u5dee\u8ddd\u5e76\u4e0d\u5927\uff0c\u5728\u82f1\u4e2d\u7684\u53e5 \u5b50\u7ea7\u522b\u76f8\u5173\u6027(0.258)\u4e0a\u751a\u81f3\u7a0d\u9ad8\u4e8eBEER\u65b9\u6cd5(0.232)\uff0c\u5728\u82f1\u4e2d\u7684\u7cfb\u7edf\u7ea7\u522b\u76f8\u5173\u6027(0.916)\u4e0a\u9ad8 \u4e8eBLEU(0.901)\u3001BEER(0.803)\u3001chrF(0.880)\u7b49\u65b9\u6cd5\u3002\u8fd9\u8bf4\u660e\u4e86\u6e90\u7aef\u4fe1\u606f\u5bf9\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u975e\u5e38 \u6709\u5e2e\u52a9\uff0c\u4ece\u4e00\u4e2a\u4fa7\u9762\u4f50\u8bc1\u4e86\u6b63\u786e\u5730\u5c06\u8d28\u91cf\u5411\u91cf\u5f15\u5165\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u5fc5\u5c06\u63d0\u9ad8\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u7684\u6027 \u80fd\u3002 ", |
| "cite_spans": [ |
| { |
| "start": 375, |
| "end": 392, |
| "text": "(Ma et al., 2019)", |
| "ref_id": "BIBREF14" |
| }, |
| { |
| "start": 737, |
| "end": 760, |
| "text": "(Papineni et al., 2002)", |
| "ref_id": "BIBREF19" |
| }, |
| { |
| "start": 767, |
| "end": 782, |
| "text": "(Popovi\u0107, 2015)", |
| "ref_id": "BIBREF21" |
| }, |
| { |
| "start": 791, |
| "end": 821, |
| "text": "(Stanojevi\u0107 and Sima'an, 2014)", |
| "ref_id": "BIBREF25" |
| }, |
| { |
| "start": 822, |
| "end": 839, |
| "text": "(Li et al., 2018)", |
| "ref_id": "BIBREF12" |
| }, |
| { |
| "start": 996, |
| "end": 1016, |
| "text": "(Bojar et al., 2015;", |
| "ref_id": "BIBREF2" |
| }, |
| { |
| "start": 1017, |
| "end": 1036, |
| "text": "Bojar et al., 2016;", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 1037, |
| "end": 1056, |
| "text": "Ondrej et al., 2017", |
| "ref_id": "BIBREF18" |
| } |
| ], |
| "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": "t,i = Bi-LSTM (F (m t,i ) , i) , \u2200i \u2208 [1, . . . , l t ]", |
| "eq_num": "(13)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "m r,j = Bi-LSTM (F (m r,j ) , j) , \u2200j \u2208 [1, . . . , l r ]", |
| "eq_num": "(14)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "p = [\u1e7d r,avg \u2295\u1e7d r,max \u2295\u1e7d t,avg \u2295\u1e7d t,max \u2295 v qe,avg \u2295 v qe,max ]", |
| "eq_num": "(15)" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "y score = w T ReLU W Tp + b + b", |
| "eq_num": "(16" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u7b49 \u7ecf \u5178 \u7684 \u65b9 \u6cd5 \u8fdb \u884c \u4e86 \u6bd4 \u8f83 \uff0c \u800c \u4e14 \u4e0eMathur(2019)\u7b49 \u4eba \u63d0 \u51fa \u7684 \u81ea \u52a8 \u8bc4 \u4ef7 \u65b9 \u6cd5 \u3001 \u4e0e \u4e0d \u4f7f \u7528 \u4eba \u5de5 \u53c2 \u8003 \u8bd1 \u6587 \u7684 \u8bd1 \u6587 \u8d28 \u91cf \u4f30 \u8ba1 \u65b9 \u6cd5UNQE", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "BERT+QE \"\u5728 \u5fb7 \u82f1 \u3001 \u4e2d \u82f1 \u548c \u82f1 \u4e2d \u4e09 \u4e2a \u8bed \u8a00 \u5bf9 \u4e0a \uff0c \u4e0e \u4eba \u5de5 \u8bc4 \u4ef7 \u7684 \u53e5 \u5b50 \u7ea7 \u522b \u76f8 \u5173 \u6027 \u5747 \u503c \u5206 \u522b \u9ad8 \u4e8e \u4f7f \u7528 \u8bed \u5883 \u8bcd \u5411 \u91cf \u7684 \u65b9 \u6cd5\"(Bi-LSTM+attention) BERT \"\u548c\"(ESIM) BERT \"\u3002\"(Bi- LSTM+attention) BERT+QE \"\u76f8 \u5bf9 \u4e8e\"(Bi-LSTM+attention) BERT \"\u5728 \u5fb7 \u82f1 \u3001 \u4e2d \u82f1 \u3001 \u82f1 \u4e2d \u4e09 \u4e2a \u4efb \u52a1 \u4e0a \u5206 \u522b \u63d0 \u5347 \u4e864.6%\u30013.2%\u548c3.8%\uff0c\"(ESIM) BERT+QE \"\u76f8 \u5bf9 \u4e8e\"(ESIM) BERT \"\u65b9 \u6cd5 \u5206 \u522b \u63d0 \u5347 \u4e867.5%\u30012.8%\u548c6.3%\u3002\u5176\u4e2d\"(Bi-LSTM+attention) BERT+QE \"", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "4.3 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u5206 \u5206 \u5206\u6790 \u6790 \u6790 \u4e3a\u4e86\u8fdb\u4e00\u6b65\u5206\u6790\u878d\u5408\u6e90\u7aef\u4fe1\u606f\u7684\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u7684\u7279\u70b9\uff0c\u6211\u4eec\u5728\u5f00\u53d1\u96c6\u4e0a\u5206\u522b\u62bd\u53d6\u4e86\u4e2d \u82f1\u548c\u82f1\u4e2d\u7ffb\u8bd1\u81ea\u52a8\u8bc4\u4ef7\u7684\u5b9e\u4f8b\u8fdb\u884c\u5206\u6790\u3002\u88685\u7ed9\u51fa\u4e86\u5bf9\u4e24\u4e2a\u8bd1\u6587\u8fdb\u884c\u6253\u5206\u7684\u5b9e\u4f8b\uff0c\u5176\u4e2dHTER\u662f \u6307\u5c06\u673a\u5668\u8bd1\u6587mt\u8f6c\u6362\u6210\u4eba\u5de5\u540e\u7f16\u8f91\u7684\u53c2\u8003\u8bd1\u6587ref \u9700\u8981\u7684\u6700\u5c11\u7f16\u8f91\u6b21\u6570\u4e0e\u8bd1\u6587\u957f\u5ea6\u7684\u6bd4\u503c\uff0c\u5b83 \u53ef\u4ee5\u770b\u4f5c\u662f\u8bd1\u6587\u4eba\u5de5\u6253\u5206\u7684\u7ed3\u679c\u3002\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u5bf9\u673a\u5668\u8bd1\u6587\u7684\u6253\u5206\u8d8a\u63a5\u8fd1\u4eba\u5de5\u6253\u5206(1-HTER)\uff0c \u8868\u660e\u8be5\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u5bf9\u8bd1\u6587\u7684\u8bc4\u4ef7\u8d8a\u51c6\u786e\u3002 \u5728\u7b2c\u4e00\u4e2a\u5b9e\u4f8b\u4e2d\uff0c\u6e90\u8bed\u8a00\u53e5\u5b50\u4e2d\"\u5bf9 \u5bf9 \u5bf9\u57ce \u57ce \u57ce\u5e02 \u5e02 \u5e02\u4ea4 \u4ea4 \u4ea4\u901a \u901a \u901a\u6765 \u6765 \u6765\u8bf4 \u8bf4 \u8bf4\"\u5728\u673a\u5668\u8bd1\u6587\u4e2d\u7f3a\u4e4f\u5bf9\u5e94\u7ffb\u8bd1\uff0c\u5b58\u5728 \u6f0f\u8bd1\u7684\u60c5\u51b5\uff0c\u4f46(Bi-LSTM+attention) BERT \u548c(ESIM) BERT \u5374\u7ed9\u4e86\u5f88\u9ad8\u7684\u5206\u503c\uff0c\u800c\u672c\u6587\u7684\u65b9\u6cd5 \u6253\u5206\u5747\u66f4\u63a5\u8fd1\u4eba\u5de5HTER\u5206\u503c\u3002\u8bf4\u660e(Bi-LSTM+attention) BERT+QE \u548c(ESIM) BERT+QE \u65b9\u6cd5\u7ed3\u5408 \u4e86\u6e90\u8bed\u8a00\u53e5\u5b50\u4fe1\u606f\u5bf9\u8bd1\u6587\u8fdb\u884c\u8bc4\u4ef7\uff0c\u80fd\u66f4\u51c6\u786e\u5730\u63cf\u8ff0\u8bd1\u6587\u7684\u5b8c\u6574\u5ea6\u7279\u5f81\uff0c\u56e0\u6b64\uff0c\u76f8\u6bd4\u4e8e\u4ec5 \u7ed3\u5408\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u4fe1\u606f\u6253\u5206\u7684(Bi-LSTM+attention) BERT \u548c(ESIM) BERT \u65b9\u6cd5\uff0c\u5f15\u5165\u6e90\u7aef\u4fe1\u606f\u7684 \u65b9\u6cd5\u5176\u8bc4\u4ef7\u66f4\u51c6\u786e\u3002\u5728\u7b2c\u4e8c\u4e2a\u5b9e\u4f8b\u4e2d\uff0c\u673a\u5668\u8bd1\u6587\u4e2d\u5b58\u5728\u591a\u8bd1\u3001\u8fc7\u5ea6\u7ffb\u8bd1\u7684\u60c5\u51b5\uff0c\u6e90\u8bed\u8a00\u53e5 \u5b50\u4e2d\"T okyo, Japan\"\u88ab\u8fc7\u5ea6\u7ffb\u8bd1\u6210\"\u4e1c \u4e1c \u4e1c\u4eac \u4eac \u4eac\"\u548c\"\u65e5 \u65e5 \u65e5\u672c \u672c \u672c\"\u4e24\u4e2a\u5730\u65b9\u3002\u5bf9\u4e8e\u8fd9\u79cd\u60c5\u51b5\uff0c\u672c\u6587\u65b9\u6cd5\u4f9d\u7136 \u6bd4(Bi-LSTM+attention) BERT \u548c(ESIM) BERT \u66f4\u63a5\u8fd1\u4eba\u5de5\u6253\u5206\u7ed3\u679cHTER\u3002\u8fd9\u5b9a\u6027\u7684\u8bf4\u660e\u4e86\u7ed3\u5408", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Neural machine translation by jointly learning to align and translate", |
| "authors": [ |
| { |
| "first": "Dzmitry", |
| "middle": [], |
| "last": "Bahdanau", |
| "suffix": "" |
| }, |
| { |
| "first": "Kyunghyun", |
| "middle": [], |
| "last": "Cho", |
| "suffix": "" |
| }, |
| { |
| "first": "Yoshua", |
| "middle": [], |
| "last": "Bengio", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proceedings of the ICLR", |
| "volume": "", |
| "issue": "", |
| "pages": "1--15", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. In Proceedings of the ICLR, pages 1-15.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Meteor: An automatic metric for mt evaluation with improved correlation with human judgments", |
| "authors": [ |
| { |
| "first": "Satanjeev", |
| "middle": [], |
| "last": "Banerjee", |
| "suffix": "" |
| }, |
| { |
| "first": "Alon", |
| "middle": [], |
| "last": "Lavie", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization", |
| "volume": "", |
| "issue": "", |
| "pages": "65--72", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Satanjeev Banerjee and Alon Lavie. 2005. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. In Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65-72.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Findings of the 2015 workshop on statistical machine translation", |
| "authors": [ |
| { |
| "first": "Ond\u0159ej", |
| "middle": [], |
| "last": "Bojar", |
| "suffix": "" |
| }, |
| { |
| "first": "Christian", |
| "middle": [], |
| "last": "Buck", |
| "suffix": "" |
| }, |
| { |
| "first": "Christian", |
| "middle": [], |
| "last": "Federmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Barry", |
| "middle": [], |
| "last": "Haddow", |
| "suffix": "" |
| }, |
| { |
| "first": "Philipp", |
| "middle": [], |
| "last": "Koehn", |
| "suffix": "" |
| }, |
| { |
| "first": "Johannes", |
| "middle": [], |
| "last": "Leveling", |
| "suffix": "" |
| }, |
| { |
| "first": "Christof", |
| "middle": [], |
| "last": "Monz", |
| "suffix": "" |
| }, |
| { |
| "first": "Pavel", |
| "middle": [], |
| "last": "Pecina", |
| "suffix": "" |
| }, |
| { |
| "first": "Matt", |
| "middle": [], |
| "last": "Post", |
| "suffix": "" |
| }, |
| { |
| "first": "Herve", |
| "middle": [], |
| "last": "Saint-Amand", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the WMT", |
| "volume": "", |
| "issue": "", |
| "pages": "1--46", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ond\u0159ej Bojar, Christian Buck, Christian Federmann, Barry Haddow, Philipp Koehn, Johannes Leveling, Christof Monz, Pavel Pecina, Matt Post, Herve Saint-Amand, et al. 2015. Findings of the 2015 workshop on statistical machine translation. In Proceedings of the WMT, pages 1-46.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Findings of the 2016 conference on machine translation", |
| "authors": [ |
| { |
| "first": "Ond\u0159ej", |
| "middle": [], |
| "last": "Bojar", |
| "suffix": "" |
| }, |
| { |
| "first": "Rajen", |
| "middle": [], |
| "last": "Chatterjee", |
| "suffix": "" |
| }, |
| { |
| "first": "Christian", |
| "middle": [], |
| "last": "Federmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Yvette", |
| "middle": [], |
| "last": "Graham", |
| "suffix": "" |
| }, |
| { |
| "first": "Barry", |
| "middle": [], |
| "last": "Haddow", |
| "suffix": "" |
| }, |
| { |
| "first": "Matthias", |
| "middle": [], |
| "last": "Huck", |
| "suffix": "" |
| }, |
| { |
| "first": "Antonio", |
| "middle": [ |
| "Jimeno" |
| ], |
| "last": "Yepes", |
| "suffix": "" |
| }, |
| { |
| "first": "Philipp", |
| "middle": [], |
| "last": "Koehn", |
| "suffix": "" |
| }, |
| { |
| "first": "Varvara", |
| "middle": [], |
| "last": "Logacheva", |
| "suffix": "" |
| }, |
| { |
| "first": "Christof", |
| "middle": [], |
| "last": "Monz", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Proceedings of the WMT", |
| "volume": "", |
| "issue": "", |
| "pages": "131--198", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ond\u0159ej Bojar, Rajen Chatterjee, Christian Federmann, Yvette Graham, Barry Haddow, Matthias Huck, Antonio Jimeno Yepes, Philipp Koehn, Varvara Logacheva, Christof Monz, et al. 2016. Findings of the 2016 conference on machine translation. In Proceedings of the WMT, pages 131-198.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Representation based translation evaluation metrics", |
| "authors": [ |
| { |
| "first": "Boxing", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Hongyu", |
| "middle": [], |
| "last": "Guo", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the ACL and IJCNLP", |
| "volume": "", |
| "issue": "", |
| "pages": "150--155", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Boxing Chen and Hongyu Guo. 2015. Representation based translation evaluation metrics. In Proceedings of the ACL and IJCNLP, pages 150-155.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Enhanced lstm for natural language inference", |
| "authors": [ |
| { |
| "first": "Qian", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Xiaodan", |
| "middle": [], |
| "last": "Zhu", |
| "suffix": "" |
| }, |
| { |
| "first": "Zhenhua", |
| "middle": [], |
| "last": "Ling", |
| "suffix": "" |
| }, |
| { |
| "first": "Si", |
| "middle": [], |
| "last": "Wei", |
| "suffix": "" |
| }, |
| { |
| "first": "Hui", |
| "middle": [], |
| "last": "Jiang", |
| "suffix": "" |
| }, |
| { |
| "first": "Diana", |
| "middle": [], |
| "last": "Inkpen", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Proceedings of the ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "1657--1668", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Qian Chen, Xiaodan Zhu, Zhenhua Ling, Si Wei, Hui Jiang, and Diana Inkpen. 2016. Enhanced lstm for natural language inference. In Proceedings of the ACL, page 1657-1668.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Bert: Pre-training of deep bidirectional transformers for language understanding", |
| "authors": [ |
| { |
| "first": "Jacob", |
| "middle": [], |
| "last": "Devlin", |
| "suffix": "" |
| }, |
| { |
| "first": "Ming-Wei", |
| "middle": [], |
| "last": "Chang", |
| "suffix": "" |
| }, |
| { |
| "first": "Kenton", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| }, |
| { |
| "first": "Kristina", |
| "middle": [], |
| "last": "Toutanova", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the NAACL", |
| "volume": "", |
| "issue": "", |
| "pages": "4171--4186", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. Proceedings of the NAACL, page 4171-4186.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Automatic evaluation of machine translation quality using n-gram cooccurence statistics", |
| "authors": [ |
| { |
| "first": "George", |
| "middle": [], |
| "last": "Doddington", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proceedings of the HLT", |
| "volume": "", |
| "issue": "", |
| "pages": "138--145", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "George Doddington. 2002. Automatic evaluation of machine translation quality using n-gram co- occurence statistics. In Proceedings of the HLT, pages 138-145.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "bilingual expert\" can find translation errors", |
| "authors": [ |
| { |
| "first": "Kai", |
| "middle": [], |
| "last": "Fan", |
| "suffix": "" |
| }, |
| { |
| "first": "Jiayi", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Bo", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Fengming", |
| "middle": [], |
| "last": "Zhou", |
| "suffix": "" |
| }, |
| { |
| "first": "Boxing", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Luo", |
| "middle": [], |
| "last": "Si", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the AAAI Conference on Artificial Intelligence", |
| "volume": "33", |
| "issue": "", |
| "pages": "6367--6374", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kai Fan, Jiayi Wang, Bo Li, Fengming Zhou, Boxing Chen, and Luo Si. 2019. \"bilingual expert\" can find translation errors. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 6367-6374.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Reval: A simple and effective machine translation evaluation metric based on recurrent neural networks", |
| "authors": [ |
| { |
| "first": "Rohit", |
| "middle": [], |
| "last": "Gupta", |
| "suffix": "" |
| }, |
| { |
| "first": "Constantin", |
| "middle": [], |
| "last": "Orasan", |
| "suffix": "" |
| }, |
| { |
| "first": "Josef", |
| "middle": [], |
| "last": "Van Genabith", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "1066--1072", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Rohit Gupta, Constantin Orasan, and Josef van Genabith. 2015. Reval: A simple and effective ma- chine translation evaluation metric based on recurrent neural networks. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1066-1072.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Pairwise neural machine translation evaluation", |
| "authors": [ |
| { |
| "first": "Francisco", |
| "middle": [], |
| "last": "Guzm\u00e1n", |
| "suffix": "" |
| }, |
| { |
| "first": "Shafiq", |
| "middle": [], |
| "last": "Joty", |
| "suffix": "" |
| }, |
| { |
| "first": "Llu\u00eds", |
| "middle": [], |
| "last": "M\u00e0rquez", |
| "suffix": "" |
| }, |
| { |
| "first": "Preslav", |
| "middle": [], |
| "last": "Nakov", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the ACL and IJCNLP", |
| "volume": "", |
| "issue": "", |
| "pages": "805--814", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Francisco Guzm\u00e1n, Shafiq Joty, Llu\u00eds M\u00e0rquez, and Preslav Nakov. 2019. Pairwise neural machine translation evaluation. In Proceedings of the ACL and IJCNLP, pages 805-814.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Predictorestimator: Neural quality estimation based on target word prediction for machine translation", |
| "authors": [ |
| { |
| "first": "Hyun", |
| "middle": [], |
| "last": "Kim", |
| "suffix": "" |
| }, |
| { |
| "first": "Hun-Young", |
| "middle": [], |
| "last": "Jung", |
| "suffix": "" |
| }, |
| { |
| "first": "Hongseok", |
| "middle": [], |
| "last": "Kwon", |
| "suffix": "" |
| }, |
| { |
| "first": "Jong-Hyeok", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| }, |
| { |
| "first": "Seung-Hoon", |
| "middle": [], |
| "last": "Na", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "ACM Transactions on Asian and Low-Resource Language Information Processing", |
| "volume": "17", |
| "issue": "1", |
| "pages": "1--22", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hyun Kim, Hun-Young Jung, Hongseok Kwon, Jong-Hyeok Lee, and Seung-Hoon Na. 2017. Predictor- estimator: Neural quality estimation based on target word prediction for machine translation. ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP), 17(1):1-22.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "A unified neural network for quality estimation of machine translation", |
| "authors": [ |
| { |
| "first": "Maoxi", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Qingyu", |
| "middle": [], |
| "last": "Xiang", |
| "suffix": "" |
| }, |
| { |
| "first": "Zhiming", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Mingwen", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "IEICE TRANSACTIONS on Information and Systems", |
| "volume": "101", |
| "issue": "9", |
| "pages": "2417--2421", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Maoxi Li, Qingyu Xiang, Zhiming Chen, and Mingwen Wang. 2018. A unified neural network for quality estimation of machine translation. IEICE TRANSACTIONS on Information and Systems, 101(9):2417-2421.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Meant 2.0: Accurate semantic mt evaluation for any output language", |
| "authors": [ |
| { |
| "first": "Chi-Kiu", |
| "middle": [], |
| "last": "Lo", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of the second conference on machine translation", |
| "volume": "", |
| "issue": "", |
| "pages": "589--597", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Chi-kiu Lo. 2017. Meant 2.0: Accurate semantic mt evaluation for any output language. In Proceedings of the second conference on machine translation, pages 589-597.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Results of the wmt19 metrics shared task: Segment-level and strong mt systems pose big challenges", |
| "authors": [ |
| { |
| "first": "Qingsong", |
| "middle": [], |
| "last": "Ma", |
| "suffix": "" |
| }, |
| { |
| "first": "Johnny", |
| "middle": [], |
| "last": "Wei", |
| "suffix": "" |
| }, |
| { |
| "first": "Ond\u0159ej", |
| "middle": [], |
| "last": "Bojar", |
| "suffix": "" |
| }, |
| { |
| "first": "Yvette", |
| "middle": [], |
| "last": "Graham", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the Fourth Conference on Machine Translation", |
| "volume": "2", |
| "issue": "", |
| "pages": "62--90", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Qingsong Ma, Johnny Wei, Ond\u0159ej Bojar, and Yvette Graham. 2019. Results of the wmt19 metrics shared task: Segment-level and strong mt systems pose big challenges. In Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages 62-90.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Putting evaluation in context: Contextual embeddings improve machine translation evaluation", |
| "authors": [ |
| { |
| "first": "Nitika", |
| "middle": [], |
| "last": "Mathur", |
| "suffix": "" |
| }, |
| { |
| "first": "Timothy", |
| "middle": [], |
| "last": "Baldwin", |
| "suffix": "" |
| }, |
| { |
| "first": "Trevor", |
| "middle": [], |
| "last": "Cohn", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "2799--2808", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Nitika Mathur, Timothy Baldwin, and Trevor Cohn. 2019. Putting evaluation in context: Contextual embeddings improve machine translation evaluation. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2799-2808.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Distributed representations of words and phrases and their compositionality", |
| "authors": [ |
| { |
| "first": "Tomas", |
| "middle": [], |
| "last": "Mikolov", |
| "suffix": "" |
| }, |
| { |
| "first": "Ilya", |
| "middle": [], |
| "last": "Sutskever", |
| "suffix": "" |
| }, |
| { |
| "first": "Kai", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Greg", |
| "middle": [ |
| "S" |
| ], |
| "last": "Corrado", |
| "suffix": "" |
| }, |
| { |
| "first": "Jeff", |
| "middle": [], |
| "last": "Dean", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Advances in neural information processing systems", |
| "volume": "", |
| "issue": "", |
| "pages": "3111--3119", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representa- tions of words and phrases and their compositionality. In Advances in neural information processing systems, pages 3111-3119.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Natural language inference by tree-based convolution and heuristic matching", |
| "authors": [ |
| { |
| "first": "Lili", |
| "middle": [], |
| "last": "Mou", |
| "suffix": "" |
| }, |
| { |
| "first": "Rui", |
| "middle": [], |
| "last": "Men", |
| "suffix": "" |
| }, |
| { |
| "first": "Ge", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Yan", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| }, |
| { |
| "first": "Lu", |
| "middle": [], |
| "last": "Zhang", |
| "suffix": "" |
| }, |
| { |
| "first": "Rui", |
| "middle": [], |
| "last": "Yan", |
| "suffix": "" |
| }, |
| { |
| "first": "Zhi", |
| "middle": [], |
| "last": "Jin", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "130--136", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lili Mou, Rui Men, Ge Li, Yan Xu, Lu Zhang, Rui Yan, and Zhi Jin. 2015. Natural language inference by tree-based convolution and heuristic matching. In Proceedings of the ACL, pages 130-136.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "Findings of the 2017 conference on machine translation (wmt17)", |
| "authors": [ |
| { |
| "first": "Bojar", |
| "middle": [], |
| "last": "Ondrej", |
| "suffix": "" |
| }, |
| { |
| "first": "Rajen", |
| "middle": [], |
| "last": "Chatterjee", |
| "suffix": "" |
| }, |
| { |
| "first": "Federmann", |
| "middle": [], |
| "last": "Christian", |
| "suffix": "" |
| }, |
| { |
| "first": "Graham", |
| "middle": [], |
| "last": "Yvette", |
| "suffix": "" |
| }, |
| { |
| "first": "Haddow", |
| "middle": [], |
| "last": "Barry", |
| "suffix": "" |
| }, |
| { |
| "first": "Huck", |
| "middle": [], |
| "last": "Matthias", |
| "suffix": "" |
| }, |
| { |
| "first": "Koehn", |
| "middle": [], |
| "last": "Philipp", |
| "suffix": "" |
| }, |
| { |
| "first": "Logacheva", |
| "middle": [], |
| "last": "Liu Qun", |
| "suffix": "" |
| }, |
| { |
| "first": "Monz", |
| "middle": [], |
| "last": "Varvara", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Christof", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of the WMT", |
| "volume": "", |
| "issue": "", |
| "pages": "169--214", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bojar Ondrej, Rajen Chatterjee, Federmann Christian, Graham Yvette, Haddow Barry, Huck Matthias, Koehn Philipp, Liu Qun, Logacheva Varvara, Monz Christof, et al. 2017. Findings of the 2017 conference on machine translation (wmt17). In Proceedings of the WMT, pages 169-214.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Bleu: a method for automatic evaluation of machine translation", |
| "authors": [ |
| { |
| "first": "Kishore", |
| "middle": [], |
| "last": "Papineni", |
| "suffix": "" |
| }, |
| { |
| "first": "Salim", |
| "middle": [], |
| "last": "Roukos", |
| "suffix": "" |
| }, |
| { |
| "first": "Todd", |
| "middle": [], |
| "last": "Ward", |
| "suffix": "" |
| }, |
| { |
| "first": "Wei-Jing", |
| "middle": [], |
| "last": "Zhu", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proceedings of the ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "311--318", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the ACL, pages 311-318.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Syntax-oriented evaluation measures for machine translation output", |
| "authors": [ |
| { |
| "first": "Maja", |
| "middle": [], |
| "last": "Popovi\u0107", |
| "suffix": "" |
| }, |
| { |
| "first": "Hermann", |
| "middle": [], |
| "last": "Ney", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "Proceedings of the Fourth Workshop on Statistical Machine Translation", |
| "volume": "", |
| "issue": "", |
| "pages": "29--32", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Maja Popovi\u0107 and Hermann Ney. 2009. Syntax-oriented evaluation measures for machine translation output. In Proceedings of the Fourth Workshop on Statistical Machine Translation, pages 29-32.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "chrf: character n-gram f-score for automatic mt evaluation", |
| "authors": [ |
| { |
| "first": "Maja", |
| "middle": [], |
| "last": "Popovi\u0107", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the Tenth Workshop on Statistical Machine Translation", |
| "volume": "", |
| "issue": "", |
| "pages": "392--395", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Maja Popovi\u0107. 2015. chrf: character n-gram f-score for automatic mt evaluation. In Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 392-395.", |
| "links": null |
| }, |
| "BIBREF22": { |
| "ref_id": "b22", |
| "title": "Ruse: Regressor using sentence embeddings for automatic machine translation evaluation", |
| "authors": [ |
| { |
| "first": "Hiroki", |
| "middle": [], |
| "last": "Shimanaka", |
| "suffix": "" |
| }, |
| { |
| "first": "Tomoyuki", |
| "middle": [], |
| "last": "Kajiwara", |
| "suffix": "" |
| }, |
| { |
| "first": "Mamoru", |
| "middle": [], |
| "last": "Komachi", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the Third Conference on Machine Translation: Shared Task Papers", |
| "volume": "", |
| "issue": "", |
| "pages": "751--758", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hiroki Shimanaka, Tomoyuki Kajiwara, and Mamoru Komachi. 2018. Ruse: Regressor using sentence embeddings for automatic machine translation evaluation. In Proceedings of the Third Conference on Machine Translation: Shared Task Papers, pages 751-758.", |
| "links": null |
| }, |
| "BIBREF23": { |
| "ref_id": "b23", |
| "title": "Terp system description", |
| "authors": [ |
| { |
| "first": "Matthew", |
| "middle": [], |
| "last": "Snover", |
| "suffix": "" |
| }, |
| { |
| "first": "Nitin", |
| "middle": [], |
| "last": "Madnani", |
| "suffix": "" |
| }, |
| { |
| "first": "Bonnie", |
| "middle": [], |
| "last": "Dorr", |
| "suffix": "" |
| }, |
| { |
| "first": "Richard", |
| "middle": [], |
| "last": "Schwartz", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "MetricsMATR workshop at AMTA", |
| "volume": "", |
| "issue": "", |
| "pages": "104--108", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Matthew Snover, Nitin Madnani, Bonnie Dorr, and Richard Schwartz. 2008. Terp system description. In MetricsMATR workshop at AMTA, pages 104-108.", |
| "links": null |
| }, |
| "BIBREF24": { |
| "ref_id": "b24", |
| "title": "Findings of the wmt 2018 shared task on quality estimation", |
| "authors": [ |
| { |
| "first": "Lucia", |
| "middle": [], |
| "last": "Specia", |
| "suffix": "" |
| }, |
| { |
| "first": "Fr\u00e9d\u00e9ric", |
| "middle": [], |
| "last": "Blain", |
| "suffix": "" |
| }, |
| { |
| "first": "Varvara", |
| "middle": [], |
| "last": "Logacheva", |
| "suffix": "" |
| }, |
| { |
| "first": "Ram\u00f3n", |
| "middle": [], |
| "last": "Astudillo", |
| "suffix": "" |
| }, |
| { |
| "first": "Andr\u00e9 Ft", |
| "middle": [], |
| "last": "Martins", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Proceedings of the Third Conference on Machine Translation: Shared Task Papers", |
| "volume": "", |
| "issue": "", |
| "pages": "689--709", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lucia Specia, Fr\u00e9d\u00e9ric Blain, Varvara Logacheva, Ram\u00f3n Astudillo, and Andr\u00e9 FT Martins. 2018. Find- ings of the wmt 2018 shared task on quality estimation. In Proceedings of the Third Conference on Machine Translation: Shared Task Papers, pages 689-709.", |
| "links": null |
| }, |
| "BIBREF25": { |
| "ref_id": "b25", |
| "title": "Beer: Better evaluation as ranking", |
| "authors": [ |
| { |
| "first": "Milo\u0161", |
| "middle": [], |
| "last": "Stanojevi\u0107", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proceedings of the Ninth Workshop on Statistical Machine Translation", |
| "volume": "", |
| "issue": "", |
| "pages": "414--419", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Milo\u0161 Stanojevi\u0107 and Khalil Sima'an. 2014. Beer: Better evaluation as ranking. In Proceedings of the Ninth Workshop on Statistical Machine Translation, pages 414-419.", |
| "links": null |
| }, |
| "BIBREF26": { |
| "ref_id": "b26", |
| "title": "Improved semantic representations from tree-structured long short-term memory networks", |
| "authors": [ |
| { |
| "first": "Kai Sheng", |
| "middle": [], |
| "last": "Tai", |
| "suffix": "" |
| }, |
| { |
| "first": "Richard", |
| "middle": [], |
| "last": "Socher", |
| "suffix": "" |
| }, |
| { |
| "first": "Christopher D", |
| "middle": [], |
| "last": "Manning", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the ACL and IJCNLP", |
| "volume": "", |
| "issue": "", |
| "pages": "1556--1566", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kai Sheng Tai, Richard Socher, and Christopher D Manning. 2015. Improved semantic representations from tree-structured long short-term memory networks. In Proceedings of the ACL and IJCNLP, page 1556-1566.", |
| "links": null |
| }, |
| "BIBREF27": { |
| "ref_id": "b27", |
| "title": "Attention is all you need", |
| "authors": [ |
| { |
| "first": "Ashish", |
| "middle": [], |
| "last": "Vaswani", |
| "suffix": "" |
| }, |
| { |
| "first": "Noam", |
| "middle": [], |
| "last": "Shazeer", |
| "suffix": "" |
| }, |
| { |
| "first": "Niki", |
| "middle": [], |
| "last": "Parmar", |
| "suffix": "" |
| }, |
| { |
| "first": "Jakob", |
| "middle": [], |
| "last": "Uszkoreit", |
| "suffix": "" |
| }, |
| { |
| "first": "Llion", |
| "middle": [], |
| "last": "Jones", |
| "suffix": "" |
| }, |
| { |
| "first": "Aidan", |
| "middle": [ |
| "N" |
| ], |
| "last": "Gomez", |
| "suffix": "" |
| }, |
| { |
| "first": "Lukasz", |
| "middle": [], |
| "last": "Kaiser", |
| "suffix": "" |
| }, |
| { |
| "first": "Illia", |
| "middle": [], |
| "last": "Polosukhin", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Advances in neural information processing systems", |
| "volume": "", |
| "issue": "", |
| "pages": "5998--6008", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems, pages 5998-6008.", |
| "links": null |
| }, |
| "BIBREF28": { |
| "ref_id": "b28", |
| "title": "Niutrans submission for ccmt19 quality estimation task", |
| "authors": [ |
| { |
| "first": "Ziyang", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Hui", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| }, |
| { |
| "first": "Hexuan", |
| "middle": [], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Kai", |
| "middle": [], |
| "last": "Feng", |
| "suffix": "" |
| }, |
| { |
| "first": "Zeyang", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Bei", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Chen", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| }, |
| { |
| "first": "Tong", |
| "middle": [], |
| "last": "Xiao", |
| "suffix": "" |
| }, |
| { |
| "first": "Jingbo", |
| "middle": [], |
| "last": "Zhu", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "China Conference on Machine Translation", |
| "volume": "", |
| "issue": "", |
| "pages": "82--92", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ziyang Wang, Hui Liu, Hexuan Chen, Kai Feng, Zeyang Wang, Bei Li, Chen Xu, Tong Xiao, and Jingbo Zhu. 2019. Niutrans submission for ccmt19 quality estimation task. In China Conference on Machine Translation, pages 82-92.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF1": { |
| "content": "<table><tr><td>4 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c</td></tr><tr><td>4.1</td></tr><tr><td>score \u4e0e\u4eba\u5de5\u8bc4\u4ef7\u5f97\u5206h\u7684\u5747\u65b9\u5dee\uff0c\u540c\u5f0f(12)\u6240</td></tr><tr><td>\u793a\u3002</td></tr><tr><td>\u83b7\u53d6\u4e86\u673a\u5668\u8bd1\u6587\u53e5\u5b50\u7ea7\u522b\u5206\u503c\u540e\uff0c\u6211\u4eec\u5bf9\u6574\u4e2a\u6d4b\u8bd5\u96c6(\u6216\u6587\u6863\u96c6)\u4e2d\u673a\u5668\u8bd1\u6587\u7684\u53e5\u5b50\u7ea7\u522b</td></tr><tr><td>\u5f97\u5206\u53d6\u5e73\u5747\u503c\u4f5c\u4e3a\u7ffb\u8bd1\u7cfb\u7edf\u7684\u7cfb\u7edf\u7ea7\u522b(\u6216\u6587\u6863\u7ea7\u522b)\u5f97\u5206\u3002</td></tr></table>", |
| "num": null, |
| "html": null, |
| "text": ") \u5176\u4e2d\u7b26\u53f7i\u3001j\u5747\u8868\u793a\u8bcd\u5e8f\u53f7\uff0cF \u8868\u793a\u6fc0\u6d3b\u51fd\u6570\u4e3aReLU \u7684\u5355\u5c42\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u5c42\uff1b\u5f0f(15)\u4e2d \u7684\u1e7d t,avg \u548c\u1e7d t,max \u5411\u91cf\u5206\u522b\u662fm t \u5e73\u5747\u6c60\u5316\u548c\u6700\u5927\u6c60\u5316\u7684\u5411\u91cf\uff0c\u1e7d r,avg \u548c\u1e7d r,max \u5206\u522b\u662fm r \u5e73\u5747\u6c60\u5316\u548c \u6700\u5927\u6c60\u5316\u7684\u5411\u91cf\uff1b\u5f0f(16)\u4e2d\u7684w\uff0cW \uff0cb\uff0cb \u5747\u4e3a\u8be5\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u7684\u53c2\u6570\u3002\u540c\u6837\uff0c\u6a21\u578b\u7684\u4f18 \u5316\u76ee\u6807\u4e5f\u5728\u8bad\u7ec3\u96c6\u4e0a\u6700\u5c0f\u5316\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u5f97\u5206y \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e \u4e3a\u4e86\u9a8c\u8bc1\u5f15\u5165\u6e90\u7aef\u4fe1\u606f\u7684\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u7684\u6548\u679c\uff0c\u6211\u4eec\u5728WMT'19 Metrics Task", |
| "type_str": "table" |
| }, |
| "TABREF4": { |
| "content": "<table><tr><td>\u8868 5. \u4e0d\u540c\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u5bf9\u673a\u5668\u8bd1\u6587\u6253\u5206\u5b9e\u4f8b</td></tr><tr><td>5 \u7ed3 \u7ed3 \u7ed3\u8bba \u8bba \u8bba</td></tr><tr><td>\u672c\u6587\u63d0\u51fa\u5f15\u5165\u6e90\u7aef\u4fe1\u606f\u7684\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u3002\u4e0e\u4f20\u7edf\u7684BLEU\u3001BEER\u3001chrF\u7b49\u8bc4\u4ef7</td></tr><tr><td>\u6307\u6807\u76f8\u6bd4\uff0c\u5f15\u5165\u6e90\u7aef\u4fe1\u606f\u7684\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\uff0c\u878d\u5408\u4e86\u6e90\u8bed\u8a00\u53e5\u5b50\u3001\u4eba\u5de5\u53c2\u8003\u8bd1\u6587\u3001\u673a\u5668</td></tr><tr><td>\u8bd1\u6587\u4e09\u8005\u7684\u4fe1\u606f\uff0c\u80fd\u66f4\u5168\u9762\u66f4\u6709\u6548\u5730\u63cf\u8ff0\u8bd1\u6587\u8d28\u91cf\u3002\u5728\u672a\u6765\u7684\u5de5\u4f5c\u4e2d\uff0c\u6211\u4eec\u5c06\u5c1d\u8bd5\u5728\u66f4\u5927\u7684\u8bed</td></tr><tr><td>\u6599\u5e93\u3001\u66f4\u591a\u7684\u8bed\u8a00\u5bf9\u4e0a\u8fdb\u884c\u5b9e\u9a8c\uff0c\u4ee5\u53ca\u5f15\u5165\u66f4\u5148\u8fdb\u7684\u6a21\u578b\u548c\u65b9\u6cd5\u6765\u6316\u6398\u6e90\u7aef\u4fe1\u606f\uff0c\u4ee5\u63d0\u9ad8\u673a\u5668</td></tr><tr><td>\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u7684\u6027\u80fd\u3002</td></tr></table>", |
| "num": null, |
| "html": null, |
| "text": "\u6e90\u7aef\u4fe1\u606f\u7684\u673a\u5668\u8bd1\u6587\u81ea\u52a8\u8bc4\u4ef7\u65b9\u6cd5\u80fd\u66f4\u5145\u5206\u5229\u7528\u6e90\u8bed\u8a00\u53e5\u5b50\u7684\u4fe1\u606f\u5bf9\u8bd1\u6587\u8d28\u91cf\u8fdb\u884c\u8bc4\u4ef7\u3002 src\uff1a\u5982\u6b64\u89c4\u6a21\u7684\u57ce\u5e02\u53d1\u5c55\u5bf9 \u5bf9 \u5bf9\u57ce \u57ce \u57ce\u5e02 \u5e02 \u5e02\u4ea4 \u4ea4 \u4ea4\u901a \u901a \u901a\u6765 \u6765 \u6765\u8bf4 \u8bf4 \u8bf4\u65e2\u662f\u6311\u6218\uff0c\u4e5f\u662f\u673a\u9047\u3002 mt\uff1aThis scale of urban development urban traffic is both a challenge and an opportunity. ref \uff1aThis scale of urban development urban traffic is both a challenge and an opportunity to urban transportation. \u4eba\u5de5\u6253\u5206(1-HTER)\uff1a0.833 (Bi-LSTM+attention) BERT \u5f97\u5206\uff1a0.883 (ESIM) BERT \u5f97\u5206\uff1a0.862 (Bi-LSTM+attention) BERT+QE \u5f97\u5206\uff1a0.833 (ESIM) BERT+QE \u5f97\u5206\uff1a0.845 src: The African Development Conference was dominated by Japan, and the previous five meetings were held in Tokyo, Japan or Yokohama, so this meeting will be the first move to Africa. mt: \u975e\u6d32\u53d1\u5c55\u4f1a\u8bae\u7531\u65e5\u672c\u4e3b\u5bfc\uff0c\u524d\u4e94\u6b21\u4f1a\u8bae\u5206\u522b\u5728\u4e1c\u4eac\u3001\u65e5 \u65e5 \u65e5\u672c \u672c \u672c\u6216\u6a2a\u6ee8\u4e3e\u884c\uff0c\u56e0\u6b64\u8fd9 \u6b21\u4f1a\u8bae\u5c06\u662f\u7b2c\u4e00\u6b21\u5230\u975e\u6d32\u7684\u4f1a\u8bae\u3002 ref : \u975e\u6d32\u5f00\u53d1\u4f1a\u8bae\u7531\u65e5\u672c\u4e3b\u5bfc\uff0c\u6b64\u524d\u7684\u4e94\u6b21\u4f1a\u8bae\u5747\u662f\u5728\u65e5\u672c\u4e1c\u4eac\u6216\u8005\u6a2a\u6ee8\u4e3e\u884c\uff0c\u56e0 \u6b64\uff0c\u672c\u6b21\u4f1a\u8bae\u4e5f\u5c06\u662f\u9996\u6b21\u79fb\u5e08\u975e\u6d32\u3002 \u4eba\u5de5\u6253\u5206(1-HTER)\uff1a0.836 (Bi-LSTM+attention) BERT \u5f97\u5206\uff1a0.705 (ESIM) BERT \u5f97\u5206\uff1a0.904 (Bi-LSTM+attention) BERT+QE \u5f97\u5206\uff1a0.888 (ESIM) BERT+QE \u5f97\u5206\uff1a0.879", |
| "type_str": "table" |
| } |
| } |
| } |
| } |
