{ "paper_id": "2020", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T12:54:21.959586Z" }, "title": "", "authors": [], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "2020", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "the advantages of the parallel sentence-level corpus in training, the two-step training method is used in this paper. Based on the Transformer ,which is trained on a sentence-level corpus, the corpus containing textual information is used for secondary training to help the model gain the ability to capture and understand global context. Experiments on several benchmark corpus data sets show that the proposed model can significantly improve translation quality compared with other strong baseline models. The experiment further shows that combining hierarchical contextual information is more advantageous than word level context. In addition, this paper attempts to combine the global context with the translation model in different ways and observe its influence on the performance of the model, and studies the distribution of the global context in document-level translations.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Keywords: Neural Machine Translation , Document-level Context 1 \u5f15 \u5f15 \u5f15\u8a00 \u8a00 \u8a00 \u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u8fd1\u4e24\u5e74\u4e0d\u65ad\u53d6\u5f97\u9f13\u821e\u4eba\u5fc3\u7684\u8fdb\u5c55\uff0c\u5df2\u7ecf\u6210\u4e3a\u5f53\u524d\u673a\u5668\u7ffb\u8bd1\u6700\u53d7\u5173\u6ce8\u7684\u7814\u7a76\u9886 \u57df\u4e4b\u4e00\u3002\u5728\u8fc7\u53bb\u51e0\u5e74\u4e2d\u7814\u7a76\u8005\u4eec\u901a\u8fc7\u4e00\u7cfb\u7edf\u6a21\u578b\u4e0d\u65ad\u63d0\u9ad8\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u7684\u6027\u80fd (Sutskever et al., 2014; Bahdanau et al., 2015; Vaswani et al., 2017; Gehring et al., 2017) \u3002\u673a\u5668\u7ffb\u8bd1\u548c\u4eba\u5de5\u7ffb\u8bd1\u4e4b\u95f4\u7684 \u8d28\u91cf\u5dee\u8ddd\u88ab\u8fd9\u4e9b\u51fa\u8272\u7684\u5de5\u4f5c\u4e0d\u65ad\u7f29\u5c0f\u3002\u5176\u4e2dTransformer\u6a21\u578bVaswani et al., (2017)\u51ed\u501f\u591a\u5934\u6ce8\u610f \u529b\u673a\u5236\u5728\u53e5\u5b50\u7ea7\u795e\u7ecf\u7ffb\u8bd1\u4efb\u52a1\u4e2d\u8fbe\u5230\u4e86\u5f53\u524d\u6700\u597d\u6210\u7ee9\u3002\u7136\u800cTransformer\u5728\u7bc7\u7ae0\u7ea7\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1 \u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u5374\u5dee\u5f3a\u4eba\u610f\uff0c\u4e3b\u8981\u539f\u56e0\u5728\u4e8e\u5176\u5ffd\u7565\u4e86\u7bc7\u7ae0\u53e5\u5b50\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\u4e5f\u6ca1\u80fd\u6709\u6548\u5229\u7528\u7bc7\u7ae0 \u4e0a\u4e0b\u6587\u3002 \u7814\u7a76\u8005\u4eec\u63d0\u51fa\u5404\u79cd\u83b7\u53d6\u4e0a\u4e0b\u6587\u7684\u65b9\u6cd5\u6539\u5584\u524d\u6587\u6240\u8ff0\u95ee\u9898\uff0c (Maruf and Haffari., 2018; Wang et al., 2017; ", "cite_spans": [ { "start": 150, "end": 174, "text": "(Sutskever et al., 2014;", "ref_id": "BIBREF24" }, { "start": 175, "end": 197, "text": "Bahdanau et al., 2015;", "ref_id": "BIBREF0" }, { "start": 198, "end": 219, "text": "Vaswani et al., 2017;", "ref_id": "BIBREF29" }, { "start": 220, "end": 241, "text": "Gehring et al., 2017)", "ref_id": "BIBREF4" }, { "start": 438, "end": 464, "text": "(Maruf and Haffari., 2018;", "ref_id": "BIBREF17" }, { "start": 465, "end": 483, "text": "Wang et al., 2017;", "ref_id": "BIBREF33" } ], "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": "2 \u5c42 \u5c42 \u5c42\u6b21 \u6b21 \u6b21\u7ed3 \u7ed3 \u7ed3\u6784 \u6784 \u6784\u5168 \u5168 \u5168\u5c40 \u5c40 \u5c40\u4e0a \u4e0a \u4e0a\u4e0b \u4e0b \u4e0b\u6587 \u6587 \u6587\u589e \u589e \u589e\u5f3a \u5f3a \u5f3a\u7684 \u7684 \u7684\u7ffb \u7ffb \u7ffb\u8bd1 \u8bd1 \u8bd1\u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u672c\u6587\u7814\u7a76\u76ee\u6807\u662f\u5c06\u7efc\u5408\u5305\u542b\u53e5\u5b50\u7ea7\u4f9d\u8d56\u5173\u7cfb\u53ca\u5355\u8bcd\u7ea7\u4f9d\u8d56\u5173\u7cfb\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u7ed3\u5408\u8fdb\u7ffb\u8bd1\u6a21 \u578b\uff0c\u4ece\u800c\u63d0\u9ad8\u6a21\u578b\u7684\u7bc7\u7ae0\u7ffb\u8bd1\u8d28\u91cf\u3002\u4e3a\u4e86\u5b9e\u73b0\u8fd9\u4e2a\u76ee\u6807\uff0c\u6211\u4eec\u9996\u5148\u5229\u7528\u6e90\u7aef\u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42 \u8f93\u51fa\u7684\u8bcd\u7ea7\u9690\u85cf\u72b6\u6001\u83b7\u53d6\u53e5\u5b50\u5411\u91cf\u3002\u7136\u540e\u57fa\u4e8e\u8bcd\u7ea7\u9690\u85cf\u72b6\u6001\u53ca\u53e5\u5b50\u5411\u91cf\u5206\u522b\u8ba1\u7b97\u5f53\u524d\u53e5\u5b50\u4e2d\u6bcf \u4e2a\u8bcd\u4e0e\u7bc7\u7ae0\u4e2d\u6240\u6709\u8bcd\u53ca\u6240\u6709\u53e5\u5b50\u4e4b\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\u3002\u6700\u7ec8\uff0c\u6211\u4eec\u901a\u8fc7\u7efc\u5408\u4e86\u4e24\u79cd\u4f9d\u8d56\u5173\u7cfb\u7684\u6743\u91cd \u83b7\u53d6\u5177\u6709\u5c42\u6b21\u5316\u7bc7\u7ae0\u7ed3\u6784\u4fe1\u606f\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\uff0c\u5e76\u5229\u7528\u8fd9\u4e9b\u4e0a\u4e0b\u6587\u8f85\u52a9\u6a21\u578b\u8fdb\u884c\u7bc7\u7ae0\u7ffb\u8bd1\u3002\u4e3a \u4e86\u4fbf\u4e8e\u7406\u89e3\u672c\u6587\u505a\u51fa\u5982\u4e0b\u5b9a\u4e49\uff1a\u542b\u6709N \u4e2a\u8bed\u53e5\u7684\u6587\u6863\u8868\u793a\u4e3a\uff1aX = (X 1 , \u2022 \u2022 \u2022 , X N )\uff0c\u7bc7\u7ae0\u4e2d\u7684\u53e5 \u5b50X i = (x i,1 , \u2022 \u2022 \u2022 , x i,n ) \u542b\u6709n \u4e2a\u8bcd, \u672c\u6587\u4f7f\u7528d m \u8868\u793a\u9690\u85cf\u72b6\u6001\u53ca\u8bcd\u5d4c\u5165\u7684\u7ef4\u5ea6\u3002 2.1 \u8bcd \u8bcd \u8bcd-\u53e5 \u53e5 \u53e5\u7ea7 \u7ea7 \u7ea7\u4f9d \u4f9d \u4f9d\u8d56 \u8d56 \u8d56\u6743 \u6743 \u6743\u91cd \u91cd \u91cd \u4e3a\u4e86\u907f\u514d\u4e4b\u524d\u7684\u7814\u7a76\u5de5\u4f5c\u4e2d\u4ec5\u4f7f\u7528\u5f53\u524d\u8bed\u53e5\u524d\u9762\u7684\u53e5\u5b50\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u5bf9\u7ffb\u8bd1\u8d28\u91cf\u9020\u6210\u7684\u8d1f\u9762 \u5f71\u54cd\u53ca\u9519\u8bef\u7d2f\u79ef\u3002\u672c\u6587\u5c06\u7bc7\u7ae0\u4e2d\u7684\u6240\u6709\u8bed\u53e5\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u6765\u6e90\u3002\u5982\u56fe2(a)\u6240\u793a\uff0c\u8bcd-\u53e5\u7ea7\u4f9d\u8d56\u6743\u91cd \u751f\u6210\u6a21\u5757\u81ea\u4e0b\u800c\u4e0a\u7531\u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\uff0c\u53e5\u5411\u91cf\u5d4c\u5165\u5c42\u53ca\u8bcd-\u53e5\u6743\u91cd\u751f\u6210\u5c42\u7ec4\u6210\u3002\u8be5\u6a21\u5757\u5c06\u6e90\u7aef \u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\u8f93\u51fa\u7684\u9690\u85cf\u72b6\u6001\u4ee5\u53e5\u5b50\u4e3a\u5355\u4f4d\u5d4c\u5165\u4e3a\u53e5\u5411\u91cf\uff0c\u518d\u901a\u8fc7\u5f53\u524d\u53e5\u4e2d\u8bcd\u4e0e\u5168\u6587\u53e5\u5411 \u91cf\u4e4b\u95f4\u7684\u6ce8\u610f\u529b\u51fd\u6570\u83b7\u53d6\u8bcd-\u53e5\u7ea7\u522b\u7684\u6743\u91cd\u3002 \u7f16 \u7f16 \u7f16\u7801 \u7801 \u7801\u5668 \u5668 \u5668\u81ea \u81ea \u81ea\u6ce8 \u6ce8 \u6ce8\u610f \u610f \u610f\u529b \u529b \u529b\u5c42 \u5c42 \u5c42\uff1a \uff1a \uff1a \u672c\u6587\u4f7f\u7528\u591a\u5934\u6ce8\u610f\u529b\u51fd\u6570Vaswani et al., (2017)\u6355\u83b7\u540c\u4e00\u53e5\u5b50\u4e2d\u5355\u8bcd\u95f4 \u7684\u4f9d\u8d56\u5173\u7cfb\u3002\u7bc7\u7ae0\u4e2d\u7684\u6bcf\u4e2a\u53e5\u5b50\u90fd\u4f1a\u4ee5\u8bcd\u4e3a\u5355\u4f4d\u88ab\u7f16\u7801\u5668\u7f16\u7801\uff0c\u4ece\u800c\u83b7\u53d6\u6e90\u7aef\u8bed\u53e5\u7684\u8bcd\u7ea7\u9690\u85cf \u72b6\u6001: S (k) i = MultiHead A (k) i , A (k) i , A (k) i ,", "eq_num": "(1)" } ], "section": "", "sec_num": null }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "MultiHead\u8868\u793a\u591a\u5934\u6ce8\u610f\u529b\u51fd\u6570\uff0c\u901a\u8fc7\u5c06\u8f93\u5165\u6620\u5c04\u5230\u4e0d\u540c\u5b50\u7a7a\u95f4\u5bf9\u8f93\u5165\u5e8f\u5217\u4e4b\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\u8fdb\u884c \u5efa\u6a21\u3002\u5176\u4e2d\u8f93\u51faS (k) i \u7684\u7ef4\u5ea6\u4e3aR n\u00d7dm . \u5bf9\u4e8e\u7f16\u7801\u5668\u7684\u7b2c\u4e00\u5c42\u6765\u8bf4, A", "eq_num": "(1)" } ], "section": "", "sec_num": null }, { "text": "i = X i \u800c\u5bf9\u4e8e\u7f16\u7801\u5668\u7684\u5176\u4ed6 \u5c42\u800c\u8a00A (k) i \u662f\u4e0a\u4e00\u5c42\u7f16\u7801\u5668\u7684\u8f93\u51faA (k\u22121) i \u3002\u5982\u56fe2\u6240\u793a\uff0c\u672c\u6587\u5c06\u8fd9\u90e8\u5206\u53c2\u6570\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u751f\u6210\u5668\u7684 \u5171\u4eab\u53c2\u6570\u3002 \u5728\u751f\u6210\u53e5\u5411\u91cf\u4e4b\u524d\u672c\u6587\u4f7f\u7528\u6b8b\u5dee\u7f51\u7edc\u548c\u5c42\u6807\u51c6\u5316\u5bf9\u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\u7684\u8f93\u51fa\u8fdb\u884c\u89c4\u6574\uff0c\u5f97 \u5230\u7684\u5b9e\u9645\u8f93\u51fa\u5982\u4e0b\uff1a S (k) i = LayerNorm S (k) i + A (k) i .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "(2)", "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": "\u5176\u4e2d\uff0cLayerNorm\u662f\u5c42\u89c4\u8303\u5316\u51fd\u6570\u3002\u51fa\u4e8e\u4fdd\u6301\u6a21\u578b\u7ed3\u6784\u56fe\u7684\u7b80\u6d01\uff0c\u672c\u6587\u5728\u540e\u7eed\u63d2\u56fe\u4e2d\u7701\u7565\u4e86\u6bcf \u4e2a\u6ce8\u610f\u529b\u5c42\u540e\u7684\u6b8b\u5dee\u8fde\u63a5\u548c\u5c42\u6807\u51c6\u5316\u3002 \u53e5 \u53e5 \u53e5\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u5d4c \u5d4c \u5d4c\u5165 \u5165 \u5165\u5c42 \u5c42 \u5c42\uff1a \uff1a \uff1a \u53d7Lin et al., (2017)\u7684\u542f\u53d1\uff0c\u672c\u6587\u4f7f\u7528\u4e00\u4e2a\u7ebf\u6027\u7ed3\u5408\u5c42\u83b7\u53d6\u53e5\u5b50\u5411\u91cf\u3002\u8be5\u5c42 \u901a\u8fc7\u6ce8\u610f\u529b\u673a\u5236\u5c06\u6574\u4e2a\u53e5\u5b50\u4e2d\u6240\u6709\u5355\u8bcd\u4ea7\u751f\u7684\u9690\u85cf\u72b6\u6001\u7ed3\u5408\u5728\u4e00\u8d77\u4ece\u800c\u751f\u6210\u53e5\u5b50\u5411\u91cf\u3002\u53e5\u4e2d\u5355 \u8bcd\u6620\u5c04\u4e3a\u53e5\u5b50\u5411\u91cf\u7684\u6743\u91cd\u8ba1\u7b97\u65b9\u6cd5\u5982\u4e0b\uff1a \u03b1 = softmax W 2 tanh W 1 S (k) i T ,", "eq_num": "(3)" } ], "section": "", "sec_num": null }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "\u5176\u4e2dW 1 \u2208 R dm\u00d7dm \uff0cW 2 \u2208 R dm \u662f\u6a21\u578b\u7684\u53c2\u6570\u77e9\u9635\u3002\u4f7f\u7528\u524d\u6587\u6240\u8ff0\u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\u8f93\u51fa\u7684\u8bcd \u7ea7\u9690\u85cf\u72b6\u6001\u53ca\u8ba1\u7b97\u51fa\u7684\u6620\u5c04\u6743\u91cd\u83b7\u5f97\u7bc7\u7ae0\u4e2d\u7684\u53e5\u5b50\u5411\u91cf: v (k) X i = n j=1 \u03b1 i,j s (k) i,j . (4) \u5176\u4e2dv (k) X i \u8868\u793a\u7bc7\u7ae0\u4e2d\u53e5\u5b50X i \u7ecf\u8fc7\u53e5\u5411\u91cf\u5d4c\u5165\u5c42\u540e\u751f\u6210\u7684\u53e5\u5411\u91cf\uff0c\u03b1 i,j \u8868\u793a\u53e5\u5b50X i \u4e2d\u5404\u5355\u8bcd\u6620\u5c04\u4e3a \u53e5\u5411\u91cf\u7684\u6743\u91cd\uff0cs (k) i,j \u8868\u793a\u53e5\u5b50X i \u4e2d\u7684\u8bcd\u7ecf\u8fc7\u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\u8f93\u51fa\u7684\u9690\u85cf\u72b6\u6001\u3002 \u6743 \u6743 \u6743\u91cd \u91cd \u91cd\u8ba1 \u8ba1 \u8ba1\u7b97 \u7b97 \u7b97\uff1a \uff1a \uff1a \u5229\u7528\u53e5\u5411\u91cf\u5d4c\u5165\u5c42\u7684\u8f93\u51fa\u8ba1\u7b97\u5f53\u524d\u53e5\u4e2d\u7684\u8bcd\u4e0e\u7bc7\u7ae0\u4e2d\u6240\u6709\u53e5\u5b50\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\uff0c \u516c\u5f0f\u5982\u4e0b\uff1a u (k) i,j = softmax s (k) i,j V (k) / d V (k) ,", "eq_num": "(5)" } ], "section": "", "sec_num": null }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "\u5176\u4e2du (k) i,j \u2208 R 1\u00d7N \u8868\u793a\u53e5\u5b50X i \u4e2d\u5355\u8bcdx i,j \u4e0e\u7bc7\u7ae0\u4e2d\u6240\u6709\u53e5\u5b50\u7684\u4f9d\u8d56\u6743\u91cd\u3002V (k) = (v (k) X 1 , \u2022 \u2022 \u2022 , v", "eq_num": "(k)" } ], "section": "", "sec_num": null }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "X N ) \u8868\u793a\u4e00\u4e2a\u7bc7\u7ae0X \u4e2d\u6240\u6709\u53e5\u5b50\u5411\u91cf\u7684\u96c6\u5408\u3002 \u8ba1\u7b97\u8bed\u8a00\u5b66 Word Self-Attention Sentence Embedding Word-Sentence Weights Word Self-Attention Sentence Embedding Word-Sentence Weights Word Self-Attention Word-Word Weights Word Self-Attention Word-Word Weights Word Self-Attention Sentence Embedding Word-Word Weights Word-Sentence Weights Hierarchical Weights Global Context Feed Forward (c) (a) (b) V X V X N \u2026.... \u2026.... \u2026.... \u2026.... Gated Sum X 1 X N S 1 S N U 1 UN X 1 X N S 1 S N Z 1 Z N \u2026.... \u2026.... \u2026.... \u2026.... i \u56fe 2: (a): \u8bcd-\u53e5\u7ea7\u4f9d\u8d56\u6743\u91cd\u7684\u83b7\u53d6\u8fc7\u7a0b; (b): \u8bcd-\u8bcd\u7ea7\u4f9d\u8d56\u6743\u91cd\u7684\u83b7\u53d6\u8fc7\u7a0b\u3002(c): \u901a\u8fc7\u7ed3\u5408\u4e0d\u540c\u5c42 \u6b21\u4f9d\u8d56\u5173\u7cfb\u83b7\u53d6\u5168\u5c40\u4e0a\u4e0b\u6587\u7684\u8fc7\u7a0b\u3002 Context Generator Feed Forward Feed Forward softmax Self-Attention Masked Self-Attention Encoder-Decoder Attention Gated Sum Word Embedding Word Embedding H C \u56fe 3: \u5c42\u6b21\u5316\u7ed3\u6784\u4e0a\u4e0b\u6587\u4e0e\u7ffb\u8bd1\u6a21\u578b\u7684\u7ed3\u5408 2.2 \u8bcd \u8bcd \u8bcd-\u8bcd \u8bcd \u8bcd\u7ea7 \u7ea7 \u7ea7\u4f9d \u4f9d \u4f9d\u8d56 \u8d56 \u8d56\u6743 \u6743 \u6743\u91cd \u91cd \u91cd \u5982\u56fe2(b)\u6240\u793a\uff0c\u5229\u7528\u6e90\u7aef\u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\u8f93\u51fa\u7684\u9690\u85cf\u72b6\u6001\uff0c\u4e3a\u5f53\u524d\u53e5\u7684\u6bcf\u4e00\u4e2a\u5355\u8bcd\u83b7\u53d6 \u5176\u4e0e\u7bc7\u7ae0\u4e2d\u6240\u6709\u5355\u8bcd\u7684\u4f9d\u8d56\u5173\u7cfb\u3002\u8ba1\u7b97\u516c\u5f0f\u5982\u4e0b\uff1a z (k) i,j = softmax s (k) i,j , S (k) / d S (k) ,", "eq_num": "(6)" } ], "section": "", "sec_num": null }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "\u5176\u4e2d\uff0cM = N \u00d7 n\u5373\u7bc7\u7ae0\u4e2d\u6240\u6709\u8bcd\u7684\u4e2a\u6570,z (k) i,j \u2208 R 1\u00d7M \u89c6\u4e3a\u53e5\u5b50X i \u4e2d\u5355\u4e2a\u5355\u8bcd\u4e0e\u5168\u6587\u5355\u8bcd\u7684\u4f9d \u8d56\u5173\u7cfb\u6743\u91cd\u5411\u91cf\u3002 2.3 \u5c42 \u5c42 \u5c42\u6b21 \u6b21 \u6b21\u5316 \u5316 \u5316\u5168 \u5168 \u5168\u5c40 \u5c40 \u5c40\u4e0a \u4e0a \u4e0a\u4e0b \u4e0b \u4e0b\u6587 \u6587 \u6587\u7684 \u7684 \u7684\u83b7 \u83b7 \u83b7\u53d6 \u53d6 \u53d6 \u672c\u6587\u6240\u8ff0\u6a21\u578b\u4f7f\u7528\u6e90\u7aef\u7bc7\u7ae0\u4f5c\u4e3a\u5168\u5c40\u4e0a\u4e0b\u6587\uff0c\u6ca1\u6709\u4f7f\u7528\u989d\u5916\u7684\u4e0a\u4e0b\u6587\u8bed\u6599\u3002\u4e3a\u4e86\u51cf\u5c11\u8ba1\u7b97 \u5f00\u9500\uff0c\u907f\u514d\u6a21\u578b\u53c2\u6570\u589e\u52a0\u8fc7\u591a\uff0c\u6211\u4eec\u5c06\u4e24\u4e2a\u5c42\u6b21\u4f9d\u8d56\u6743\u91cd\u7684\u8ba1\u7b97\u8fc7\u7a0b\u53ca\u5168\u5c40\u4e0a\u4e0b\u6587\u83b7\u53d6\u8fc7\u7a0b\u5efa \u7acb\u5728\u7f16\u7801\u5668\u4e2d\uff0c\u5e76\u5171\u4eab\u7f16\u7801\u5668\u4e2d\u7684\u81ea\u6ce8\u610f\u529b\u5c42\u53c2\u6570\u3002 \u5982\u56fe2(c)\u6240\u793a\uff0c\u672c\u6587\u4f7f\u7528\u901a\u8fc7\u8bcd-\u53e5\u7ea7\u6743\u91cd\u4fee\u6b63\u8bcd-\u8bcd\u7ea7\u6743\u91cd\uff0c\u4f7f\u5f97\u6700\u7ec8\u83b7\u5f97\u7684\u6743\u91cd\u77e9\u9635\u65e2\u542b \u6709\u53e5\u5b50\u5c42\u9762\u7684\u4f9d\u8d56\u5173\u7cfb\uff0c\u53c8\u542b\u6709\u6574\u4e2a\u7bc7\u7ae0\u4e2d\u6bcf\u4e2a\u5355\u8bcd\u4e4b\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\u3002\u516c\u5f0f\u5982\u4e0b\uff1a Q i combine = U i Z i . (7) \u5176\u4e2d\u3002U (k) i \u2208 R n\u00d7N \u89c6\u4e3a\u8868\u793a\u53e5\u5b50X i \u4e2d\u6240\u6709\u5355\u8bcd\u4e0e\u7bc7\u7ae0\u4e2d\u5176\u4ed6\u53e5\u5b50\u4e4b\u95f4\u4f9d\u8d56\u5173\u7cfb\u7684\u6743\u91cd\u5411 \u91cf\uff0cZ (k) i \u2208 R n\u00d7M \u89c6\u4e3a\u7bc7\u7ae0\u4e2d\u53e5\u5b50X i \u6240\u6709\u5355\u8bcd\u5404\u81ea\u4e0e\u7bc7\u7ae0\u6240\u6709\u5355\u8bcd\u7684\u4f9d\u8d56\u5173\u7cfb\u6743\u91cd\u77e9\u9635\u3002 \u5229\u7528\u8574\u542b\u4e24\u5c42\u4f9d\u8d56\u5173\u7cfb\u7684\u6743\u91cd\u77e9\u9635\uff0c\u5c06\u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\u8f93\u51fa\u4ee5\u7bc7\u7ae0\u4e3a\u5355\u4f4d\u7684\u9690\u85cf\u72b6\u6001\u5206 \u914d\u7ed9\u5f53\u524d\u53e5\u7684\u6bcf\u4e2a\u5355\u8bcd\u3002\u81f3\u6b64\uff0c\u5f53\u524d\u53e5\u7684\u6bcf\u4e2a\u5355\u8bcd\u90fd\u5404\u81ea\u83b7\u53d6\u7279\u6709\u7684\u8574\u542b\u81ea\u4e0a\u800c\u4e0b\u4e0d\u540c\u5c42\u9762\u4f9d \u8d56\u5173\u7cfb\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u3002 C hier i = Q i combine S (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": "\u5176\u4e2d\uff0cC hier i \u2208 R n\u00d7dm \u5373\u8bed\u53e5X i \u83b7\u53d6\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u3002 2.4 \u5c42 \u5c42 \u5c42\u6b21 \u6b21 \u6b21\u5316 \u5316 \u5316\u5168 \u5168 \u5168\u5c40 \u5c40 \u5c40\u4e0a \u4e0a \u4e0a\u4e0b \u4e0b \u4e0b\u6587 \u6587 \u6587\u7684 \u7684 \u7684\u7ed3 \u7ed3 \u7ed3\u5408 \u5408 \u5408 \u7f16\u7801\u5668\u81ea\u6ce8\u610f\u529b\u5c42\u8f93\u51fa\u7684\u9690\u85cf\u72b6\u6001\u901a\u8fc7\u524d\u9988\u5168\u8fde\u63a5\u7f51\u7edc\u540e\u5f97\u5230\u7ffb\u8bd1\u6a21\u578b\u7f16\u7801\u5668\u8f93\u51fa\uff0c\u5176\u8868 \u8fbe\u5f62\u5f0f\u5982\u4e0b\uff1a H (k) i = FNN(S (k) i ).", "eq_num": "(9)" } ], "section": "", "sec_num": null }, { "text": "\u6700\u7ec8,\u5c42\u6b21\u5316\u5168\u5c40\u4e0a\u4e0b\u6587\u4e0e\u7f16\u7801\u5668\u8f93\u51fa\u901a\u8fc7\u95e8\u63a7\u5355\u5143\u7ed3\u5408\u3002 Training 47,758 781,524 6,531 180,853 7,491 206,126 Dev 82 1,664 33 887 326 8,967 Test 627 5,833 165 4,706 87 2,271 37.07 36.16 24.00 44.69 HAN-DocNMT (Miculicich et al., 2018) 37 ", "cite_spans": [ { "start": 197, "end": 222, "text": "(Miculicich et al., 2018)", "ref_id": "BIBREF20" } ], "ref_spans": [ { "start": 27, "end": 155, "text": "Training 47,758 781,524 6,531 180,853 7,491 206,126 Dev 82 1,664 33 887 326 8,967 Test 627 5,833 165 4,706 87", "ref_id": null } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "H (k) i = \u03bbH (k) i + (1 \u2212 \u03bb) C hier(k) i . (10) Set ZH-EN ES-EN EN-DE #SubDoc #Sent #SubDoc #Sent #SubDoc #Sent", "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": "\u8868 1: \u8bad\u7ec3\u96c6\uff0c\u5f00\u53d1\u96c6\u53ca\u6d4b\u8bd5\u96c6\u7684\u7edf\u8ba1\u4fe1\u606f \u95e8\u63a7\u5355\u5143\u7cfb\u6570\u7684\u8ba1\u7b97\u65b9\u6cd5\u5982\u4e0b\uff1a \u03bb = sigmoid [H (k) i ; C hier(k) i ]W G ,", "eq_num": "(11)" } ], "section": "", "sec_num": null }, { "text": "\u5176\u4e2dH (k) i \u2208 R n\u00d7dm \u662f\u7f16\u7801\u5668\u7ecf\u8fc7\u5168\u8fde\u63a5\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u5c42\u540e\u7684\u8f93\u51fa\u3002W G \u2208 R 2dm\u00d7dm \u662f\u6a21\u578b\u53c2\u6570 \u77e9\u9635\u3002\u5982\u56fe3\u6240\u793a\uff0c\u5c42\u6b21\u5316\u5168\u5c40\u4e0a\u4e0b\u6587\u4e0e\u7f16\u7801\u5668\u8f93\u51fa\u7ed3\u5408\u540e\u8fdb\u5165\u89e3\u7801\u5668\u3002 3 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c \u672c\u6587\u5c06\u4ec5\u901a\u8fc7\u8bcd-\u8bcd\u7ea7\u4f9d\u8d56\u6743\u91cd\u83b7\u53d6\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u79f0\u4e3a\u8bcd\u7ea7\u4e0a\u4e0b\u6587\u3002\u5c06\u4f7f\u7528\u88ab\u8bcd-\u53e5\u7ea7\u6743\u91cd \u89c4\u6574\u8fc7\u7684\u590d\u5408\u6743\u91cd\u83b7\u53d6\u7684\u542b\u6709\u5c42\u6b21\u5316\u7bc7\u7ae0\u4fe1\u606f\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u79f0\u4e3a\u590d\u5408\u4e0a\u4e0b\u6587\u3002\u672c\u6587\u5206\u522b\u9009\u62e9\u9650 \u5b9a\u4e0a\u4e0b\u6587\u83b7\u53d6\u8303\u56f4\u53ca\u7ed3\u6784\u5316\u4e0a\u4e0b\u6587\u4e24\u7c7b\u4e0a\u4e0b\u6587\u7ed3\u5408\u65b9\u5f0f\u4e2d\u6027\u80fd\u8f83\u597d\u7684\u5f3a\u57fa\u7ebf\u6a21\u578b\u4f5c\u4e3a\u5bf9\u6bd4\u6a21 \u578b\u3002 3.1 \u6570 \u6570 \u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6 \u5728\u4e2d-\u82f1\u5b9e\u9a8c\u4e2d\uff0c\u7bc7\u7ae0\u7ea7\u5e73\u884c\u8bed\u6599\u7684\u8bad\u7ec3\u96c6\u5305\u62ec4.7\u4e07\u4e2a\u6587\u6863\u4e2d\u768478\u4e07\u4e2a\u53e5\u5b50\u5bf9 0 \u3002\u6211\u4eec\u4f7f \u7528NIST MT 2006\u6570\u636e\u96c6\u4f5c\u4e3a\u5f00\u53d1\u96c6\uff0c\u5e76\u4f7f\u7528MT 2002\u30012003\u30012004\u30012005\u30012008\u6570\u636e\u96c6\u4f5c\u4e3a\u6d4b \u8bd5\u96c6\uff0c\u5176\u4e2d\u6d4b\u8bd5\u96c6\u7684\u5408\u96c6\u6807\u8bb0\u4e3aAll\u3002\u672c\u6587\u4f7f\u7528Jieba 1 \u5206\u8bcd\u5c06\u6c49\u8bed\u53e5\u5b50\u6309\u8bcd\u5207\u5206\uff0c\u800c\u82f1\u8bed\u53e5\u5b50\u5219 \u4f7f\u7528Moses \u811a\u672cKoehn et al., (2007)\u8fdb\u884c\u5206\u8bcd\u548c\u5c0f\u5199\u5904\u7406\u3002\u6211\u4eec\u901a\u8fc7BPE Sennrich et al., (2016)\u4f7f \u75283\u4e07\u5927\u5c0f\u7684\u8bcd\u8868\u5206\u522b\u5c06\u6e90\u8bed\u8a00\u548c\u76ee\u6807\u8bed\u8a00\u4e2d\u7684\u5355\u8bcd\u8fdb\u4e00\u6b65\u5206\u5272\u6210\u5b50\u8bcd \u897f \u73ed \u7259-\u82f1 \u7ffb \u8bd1 \u4efb \u52a1 \u4e2d \u7684 \u8bad \u7ec3 \u96c6 \u4e3aIWSLT 2014\u548c2015 Cettolo et al., (2012)\uff0c \u5f00 \u53d1 \u96c6 \u4e3adev2010\uff0c\u6d4b\u8bd5\u96c6\u4e3atst2010\u3001tst2011\u548ctst2012\u3002\u82f1-\u5fb7\u7ffb\u8bd1\u4efb\u52a1\u4e2d\u7684\u8bad\u7ec3\u96c6\u6765\u81eaIWSLT2017\uff0c \u672c\u6587\u4f7f\u7528tst2016\u548ctst2017\u4f5c\u4e3a\u6d4b\u8bd5\u96c6\uff0c\u4f59\u4e0b\u6570\u636e\u96c6\u4f5c\u4e3a\u5f00\u53d1\u96c6\u3002\u6240\u6709\u6570\u636e\u96c6\u5747\u4f7f\u7528Moses\u811a\u672c \u8fdb\u884c\u5206\u8bcd\u548cTruecasing\u5904\u7406\u3002\u5e76\u4f7f\u75283\u4e07\u5927\u5c0f\u7684\u8054\u5408\u8bcd\u8868\u5c06\u6e90\u7aef\u53ca\u76ee\u6807\u7aef\u8bed\u6599\u4e2d\u7684\u5355\u8bcd\u5206\u5272\u6210\u5b50 \u8bcd\u3002\u7531\u4e8e\u672c\u6587\u8bcd\u7ea7\u522b\u4e0a\u4e0b\u6587\u9700\u8981\u8ba1\u7b97\u7bc7\u7ae0\u4e2d\u6240\u6709\u8bcd\u4e4b\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\uff0c\u8ba1\u7b97\u5f00\u9500\u53ca\u663e\u5b58\u5360\u7528\u90fd \u5341\u5206\u53ef\u89c2\u3002\u8003\u8651\u5230\u8bad\u7ec3\u6548\u7387\uff0c\u6211\u4eec\u5c06\u957f\u7bc7\u7ae0\u5207\u5206\u4e3a\u6700\u5927\u957f\u5ea6\u4e3a30\u53e5\u7684\u6bb5\u843d\u3002\u5b9e\u9a8c\u6570\u636e\u96c6\u7684\u7bc7\u7ae0 \u6570\uff0c\u53e5\u5b50\u6570\u53ca\u5e73\u5747\u7bc7\u7ae0\u957f\u5ea6\u7b49\u7edf\u8ba1\u4fe1\u606f\u5982\u88681\u6240\u793a\u3002 3.2 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e \u672c \u6587 \u57fa \u4e8eOpneNMT 2 Klein et al., (2017)\u5b9e \u73b0 \u4ee5 \u5e73 \u884c \u53e5 \u5bf9 \u4e3a \u5355 \u4f4d \u66f4 \u65b0 \u53c2 \u6570 \u7684 \u57fa \u51c6 \u6a21 \u578bTransformer\uff0c\u5e76\u8fdb\u4e00\u6b65\u62d3\u5c55\u4e3a\u4ee5\u7bc7\u7ae0\u4e3a\u5355\u4f4d\u66f4\u65b0\u53c2\u6570\u7684\u7ffb\u8bd1\u6a21\u578b\u3002\u4ee5\u7bc7\u7ae0\u4e3a\u5355\u4f4d\u66f4\u65b0\u4f7f \u5f97\u6a21\u578b\u53ef\u4ee5\u8f7b\u6613\u83b7\u53d6\u8bed\u6599\u7684\u7bc7\u7ae0\u4fe1\u606f\uff0c\u4ece\u800c\u8fdb\u4e00\u6b65\u83b7\u53d6\u5168\u5c40\u4e0a\u4e0b\u6587\u3002\u672c\u6587\u5c06\u6a21\u578b\u9690\u85cf\u72b6\u6001\u7684\u7ef4 \u5ea6\u8bbe\u4e3a512\uff0c\u6bcf\u4e2a\u7f16\u7801\u5668\u89e3\u7801\u5668\u7684\u5c42\u6570\u90fd\u8bbe\u7f6e\u4e3a6\uff0c\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u4e2d\u7684\u4e2a\u6570\u90fd\u8bbe\u7f6e\u4e3a8\uff0c\u67f1\u72b6", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "4 \u5206 \u5206 \u5206\u6790 \u6790 \u6790\u4e0e \u4e0e \u4e0e\u8ba8 \u8ba8 \u8ba8\u8bba \u8bba \u8bba 4.1 \u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u53c2 \u53c2 \u53c2\u6570 \u6570 \u6570\u53ca \u53ca \u53ca\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65f6 \u65f6 \u65f6\u95f4 \u95f4 \u95f4 \u5982\u88684\u7edf\u8ba1\u6570\u636e\u6240\u793a\uff0c\u672c\u6587\u63d0\u51fa\u7684\u4e0a\u4e0b\u6587\u83b7\u53d6\u53ca\u7ed3\u5408\u65b9\u5f0f\u7531\u4e8e\u7f16\u7801\u5668\u591a\u5934\u6ce8\u610f\u529b\u5c42\u53c2\u6570\u5171 \u6a21 \u6a21 \u6a21\u578b \u578b \u578b \u53c2 \u53c2 \u53c2\u6570 \u6570 \u6570( ( (\u767e \u767e \u767e\u4e07 \u4e07 \u4e07) ) ) Transformer 51.3 + \u590d\u5408\u4e0a\u4e0b\u6587", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "57.6 HAN-DocNMT (Miculicich et al., 2018) 63.0 Transformer-DocNMT 96.8 (Gong et al., 2011; Hardmeier et al., 2012; Xiong et al., 2013; Tu et al., 2014; Garcia et al., 2015) \u5728 \u4f7f\u7528\u7bc7\u7ae0\u4fe1\u606f\u63d0\u9ad8\u7edf\u8ba1\u7ffb\u8bd1\u8d28\u91cf\u7684\u7814\u7a76\u9886\u57df\u505a\u4e86\u5927\u91cf\u5de5\u4f5c\u3002\u673a\u5668\u7ffb\u8bd1\u7814\u7a76\u70ed\u70b9\u4ece\u7edf\u8ba1\u7ffb\u8bd1\u8f6c\u5411 \u795e\u7ecf\u7ffb\u8bd1\u540e\u4e0d\u4e45\uff0c\u7bc7\u7ae0\u7ea7\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u7684\u7814\u7a76\u4e5f\u84ec\u52c3\u53d1\u5c55\u8d77\u6765\u3002\u6839\u636e\u83b7\u53d6\u4e0a\u4e0b\u6587\u7684\u8303\u56f4\uff0c\u6211\u4eec \u5c06\u76f8\u5173\u7814\u7a76\u5206\u4e3a\u4e24\u7c7b:(1)\u4f7f\u7528\u90e8\u5206\u8bed\u53e5\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u7684\u7814\u7a76;(2)\u4f7f\u7528\u7bc7\u7ae0\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u7684\u7814\u7a76\u3002 \u5728\u7b2c\u4e00\u7c7b\u7814\u7a76\u4e2d\uff0cTiedemann and Scherrer., (2017)\u57fa\u4e8e\u5faa\u73af\u795e\u7ecf\u7f51\u7edc(RNN)\u76f4\u63a5\u62fc\u63a5\u8bed\u53e5\u4f5c \u4e3a\u4e0a\u4e0b\u6587\u3002\u968f\u540e Wang et al., 2017; Bawden et al., 2018; voita et al., 2019) \u7684\u7814\u7a76\u4e2d\uff0c\u4ee5RNNSearch\u548cTransformer\u4e3a\u57fa\u7840\u4f7f\u7528\u5177\u6709\u4e0d\u540c\u6ce8\u610f\u529b\u673a\u5236\u7684\u591a\u7f16\u7801\u5668\u63d0 \u5347\u7bc7\u7ae0\u7ffb\u8bd1\u8d28\u91cf\u3002Miculicich et al., (2018)\u63d0\u51fa\u4e00\u79cd\u5206\u5c42\u6ce8\u610f\u7f51\u7edc(HAN)\uff0c\u5b83\u901a\u8fc7\u53e5\u8bcd\u7684\u62bd\u8c61\u8868\u793a \u4e3a\u5f53\u524d\u53e5\u4ece\u524d\u9762\u7684\u53e5\u5b50\u4e2d\u63d0\u53d6\u4e0a\u4e0b\u6587\u3002Yang et al., (2019)\u5728HAN\u7684\u57fa\u7840\u4e0a\u63d0\u51fa\u4e00\u79cd\u80f6\u56ca\u7f51\u7edc\u5c06\u4e0a \u4e0b\u6587\u4fe1\u606f\u6309\u4e0d\u540c\u89d2\u5ea6\u8fdb\u884c\u805a\u7c7b\u3002 (Tu et al., 2018; Kuang et al., 2018) \u63d0\u51fa\u7684\u57fa\u4e8e\u7f13\u5b58\u7684\u65b9\u6cd5\u6240\u5b58\u50a8 \u7684\u662f\u524d\u9762\u53e5\u5b50\u4e2d\u7684\u8bcd/\u7ffb\u8bd1\uff0c\u4e5f\u5f52\u4e8e\u8fd9\u4e00\u7c7b\u7814\u7a76\u3002 \u53e6\u4e00\u7c7b\u7814\u7a76\u4ee5\u7bc7\u7ae0\u4e3a\u7ffb\u8bd1\u5355\u5143\uff0c\u9488\u5bf9\u6bcf\u4e2a\u53e5\u5b50\u52a8\u6001\u83b7\u53d6\u6709\u7528\u7684\u7bc7\u7ae0\u7ea7\u4fe1\u606f\u3002Maruf and Haffari., (2018)\u4f7f \u7528 \u989d \u5916 \u7684 \u5b58 \u50a8 \u7f51 \u7edc \u5c06 \u7bc7 \u7ae0 \u8f6c \u6362 \u4e3a \u4e0a \u4e0b \u6587 \u4e0e \u57fa \u4e8eRNN\u7684 \u795e \u7ecf \u673a \u5668 \u7ffb \u8bd1 \u6a21 \u578b \u7ed3 \u5408 \u3002Mace and Servan., (2019)\u5728 \u6bcf \u4e2a \u6e90 \u53e5 \u4e2d \u6dfb \u52a0 \u7bc7 \u7ae0 \u6807 \u7b7e \uff0c \u5e76 \u5c06 \u5176 \u66ff \u6362 \u4e3a \u7bc7 \u7ae0 \u7ea7 \u5d4c \u5165 \u5411 \u91cf\u3002Xiong et al., 2019 ", "cite_spans": [ { "start": 16, "end": 41, "text": "(Miculicich et al., 2018)", "ref_id": "BIBREF20" }, { "start": 71, "end": 90, "text": "(Gong et al., 2011;", "ref_id": "BIBREF3" }, { "start": 91, "end": 114, "text": "Hardmeier et al., 2012;", "ref_id": "BIBREF6" }, { "start": 115, "end": 134, "text": "Xiong et al., 2013;", "ref_id": "BIBREF35" }, { "start": 135, "end": 151, "text": "Tu et al., 2014;", "ref_id": "BIBREF26" }, { "start": 152, "end": 172, "text": "Garcia et al., 2015)", "ref_id": "BIBREF5" }, { "start": 372, "end": 390, "text": "Wang et al., 2017;", "ref_id": "BIBREF33" }, { "start": 391, "end": 411, "text": "Bawden et al., 2018;", "ref_id": "BIBREF1" }, { "start": 412, "end": 431, "text": "voita et al., 2019)", "ref_id": "BIBREF31" }, { "start": 610, "end": 627, "text": "(Tu et al., 2018;", "ref_id": "BIBREF28" }, { "start": 628, "end": 647, "text": "Kuang et al., 2018)", "ref_id": "BIBREF9" } ], "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. Proceedings of ICLR", "authors": [ { "first": "Dzmitry", "middle": [], "last": "Bahdanau", "suffix": "" }, { "first": "Kyunghyun", "middle": [], "last": "Cho", "suffix": "" }, { "first": "Yoshua", "middle": [], "last": "Bengio", "suffix": "" } ], "year": 2015, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Dzmitry Bahdanau and Kyunghyun Cho and Yoshua Bengio. 2015. Neural Machine Translation by Jointly Learn- ing to Align and Translate. Proceedings of ICLR.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Evaluating discourse phenomena in neural machine translation", "authors": [ { "first": "Rachel", "middle": [], "last": "Bawden", "suffix": "" }, { "first": "Rico", "middle": [], "last": "Sennrich", "suffix": "" }, { "first": "Alexandra", "middle": [], "last": "Birch", "suffix": "" }, { "first": "Barry", "middle": [], "last": "Haddow", "suffix": "" } ], "year": 2018, "venue": "Proceedings of NAACL", "volume": "", "issue": "", "pages": "1304--1313", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rachel Bawden and Rico Sennrich and Alexandra Birch and Barry Haddow. 2018. Evaluating discourse phenom- ena in neural machine translation. In Proceedings of NAACL, 1304-1313.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "AllenNLP: A Deep Semantic Natural Language Processing Platform", "authors": [ { "first": "Matt", "middle": [], "last": "Gardner", "suffix": "" }, { "first": "Joel", "middle": [], "last": "Grus", "suffix": "" }, { "first": "Mark", "middle": [], "last": "Neumann", "suffix": "" }, { "first": "Oyvind", "middle": [], "last": "Tafjord", "suffix": "" }, { "first": "Pradeep", "middle": [], "last": "Dasigi", "suffix": "" }, { "first": "Nelson", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Matthew", "middle": [], "last": "Peters", "suffix": "" }, { "first": "Michael", "middle": [], "last": "Schmitz", "suffix": "" }, { "first": "Luke", "middle": [], "last": "Zettlemoyer", "suffix": "" } ], "year": 2017, "venue": "Proceedings of ACL Workshop for Natural Language Processing Open Source Software", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Matt Gardner and Joel Grus and Mark Neumann and Oyvind Tafjord and Pradeep Dasigi and Nelson Liu and Matthew Peters and Michael Schmitz and Luke Zettlemoyer. 2017. AllenNLP: A Deep Semantic Natural Language Processing Platform. In Proceedings of ACL Workshop for Natural Language Processing Open Source Software.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Cache-based Document-level Statistical Machine Translation", "authors": [ { "first": "Zhengxian", "middle": [], "last": "Gong", "suffix": "" }, { "first": "Min", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Guodong", "middle": [], "last": "Zhou", "suffix": "" } ], "year": 2011, "venue": "Proceedings of EMNLP", "volume": "", "issue": "", "pages": "909--919", "other_ids": {}, "num": null, "urls": [], "raw_text": "Zhengxian Gong and Min Zhang and Guodong Zhou. 2011. Cache-based Document-level Statistical Machine Translation. Proceedings of EMNLP, 909-919.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Convolutional sequence to sequence learning", "authors": [ { "first": "Jonas", "middle": [], "last": "Gehring", "suffix": "" }, { "first": "Michael", "middle": [], "last": "Auli", "suffix": "" }, { "first": "David", "middle": [], "last": "Grangier", "suffix": "" }, { "first": "Denis", "middle": [], "last": "Yarats", "suffix": "" }, { "first": "Yann N", "middle": [], "last": "Dauphin", "suffix": "" } ], "year": 2017, "venue": "Proceedings of the 34th International Conference on Machine Learning", "volume": "70", "issue": "", "pages": "1243--1252", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jonas Gehring, Michael Auli, David Grangier, Denis Yarats, and Yann N Dauphin. 2017. Convolutional sequence to sequence learning.. Proceedings of the 34th International Conference on Machine Learning, 70:1243- 1252.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Document-Level Machine Translation with Word Vector Models", "authors": [ { "first": "Mart\u00ednez", "middle": [], "last": "Eva", "suffix": "" }, { "first": "Cristina", "middle": [], "last": "Garcia", "suffix": "" }, { "first": "-", "middle": [], "last": "Espa\u00f1a", "suffix": "" }, { "first": "Llu\u00eds", "middle": [], "last": "Bonet", "suffix": "" }, { "first": "", "middle": [], "last": "M\u00e0rquez", "suffix": "" } ], "year": 2015, "venue": "Proceedings of EAMT", "volume": "", "issue": "", "pages": "59--66", "other_ids": {}, "num": null, "urls": [], "raw_text": "Eva Mart\u00ednez Garcia and Cristina Espa\u00f1a-Bonet and Llu\u00eds M\u00e0rquez. 2015. Document-Level Machine Translation with Word Vector Models. Proceedings of EAMT, 59-66.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Document-Wide Decoding for Phrase-Based Statistical Machine Translation", "authors": [ { "first": "Christian", "middle": [], "last": "Hardmeier", "suffix": "" }, { "first": "Joakim", "middle": [], "last": "Nivre", "suffix": "" }, { "first": "J\u00f6rg", "middle": [], "last": "Tiedemann", "suffix": "" } ], "year": 2012, "venue": "Proceedings of EMNLP-CoNLL", "volume": "", "issue": "", "pages": "1179--1190", "other_ids": {}, "num": null, "urls": [], "raw_text": "Christian Hardmeier and Joakim Nivre and J\u00f6rg Tiedemann. 2012. Document-Wide Decoding for Phrase-Based Statistical Machine Translation. Proceedings of EMNLP-CoNLL, 1179-1190.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Achieving Human Parity on Automatic Chinese to English News Translation. Computing Research Repository", "authors": [], "year": 2018, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1803.05567" ] }, "num": null, "urls": [], "raw_text": "Hany Hassan and Anthony Aue and Chang Chen and Vishal Chowdhary and Jonathan Clark and Christian Fed- ermann and Xuedong Huang and others. 2018. Achieving Human Parity on Automatic Chinese to English News Translation. Computing Research Repository, arXiv:1803.05567.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Does neural machinetranslation benefit from larger context", "authors": [ { "first": "Sebastien", "middle": [], "last": "Jean", "suffix": "" }, { "first": "Stanislas", "middle": [], "last": "Lauly", "suffix": "" }, { "first": "Orhan", "middle": [], "last": "Firat", "suffix": "" }, { "first": "Kyunghyun", "middle": [], "last": "Cho", "suffix": "" } ], "year": 2017, "venue": "Computing Research Repository", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1704.05135" ] }, "num": null, "urls": [], "raw_text": "Sebastien Jean and Stanislas Lauly and Orhan Firat and Kyunghyun Cho. 2017. Does neural machinetranslation benefit from larger context?. In Computing Research Repository, arXiv:1704.05135.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Modeling Coherence for Neural Machine Translation with Dynamic and Topic Caches", "authors": [ { "first": "Shaohui", "middle": [], "last": "Kuang", "suffix": "" }, { "first": "Deyi", "middle": [], "last": "Xiong", "suffix": "" }, { "first": "Weihua", "middle": [], "last": "Luo", "suffix": "" }, { "first": "Guodong", "middle": [], "last": "Zhou", "suffix": "" } ], "year": 2018, "venue": "Proceedings of COLING", "volume": "", "issue": "", "pages": "596--606", "other_ids": {}, "num": null, "urls": [], "raw_text": "Shaohui Kuang and Deyi Xiong and Weihua Luo and Guodong Zhou. 2018. Modeling Coherence for Neural Machine Translation with Dynamic and Topic Caches. Proceedings of COLING, 596-606.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Statistical significance tests for machine translation evaluation", "authors": [ { "first": "Philipp", "middle": [], "last": "Koehn", "suffix": "" } ], "year": 2004, "venue": "Proceedings of EMNLP", "volume": "", "issue": "", "pages": "388--395", "other_ids": {}, "num": null, "urls": [], "raw_text": "Philipp Koehn. 2004. Statistical significance tests for machine translation evaluation. Proceedings of EMNLP, 388--395.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Moses: Open Source Toolkit for Statistical Machine Translation. Proceedings of ACL", "authors": [ { "first": "Philipp", "middle": [], "last": "Koehn", "suffix": "" }, { "first": "Hieu", "middle": [], "last": "Hoang", "suffix": "" }, { "first": "Alexandra", "middle": [], "last": "Birch", "suffix": "" }, { "first": "Chris", "middle": [], "last": "Callison-Burch", "suffix": "" }, { "first": "Marcello", "middle": [], "last": "Federico", "suffix": "" }, { "first": "Nicola", "middle": [], "last": "Bertoldi", "suffix": "" }, { "first": "Brooke", "middle": [], "last": "Cowan", "suffix": "" }, { "first": "Wade", "middle": [], "last": "Shen", "suffix": "" }, { "first": "Christine", "middle": [], "last": "Moran", "suffix": "" }, { "first": "Richard", "middle": [], "last": "Zens", "suffix": "" }, { "first": "Chris", "middle": [], "last": "Dyer", "suffix": "" }, { "first": "Ond\u0159ej", "middle": [], "last": "Bojar", "suffix": "" }, { "first": "Alexandra", "middle": [], "last": "Constantin", "suffix": "" }, { "first": "Evan", "middle": [], "last": "Herbst", "suffix": "" } ], "year": 2007, "venue": "", "volume": "", "issue": "", "pages": "177--180", "other_ids": {}, "num": null, "urls": [], "raw_text": "Koehn, Philipp and Hoang, Hieu and Birch, Alexandra and Callison-Burch, Chris and Federico, Marcello and Bertoldi, Nicola and Cowan, Brooke and Shen, Wade and Moran, Christine and Zens, Richard and Dyer, Chris and Bojar, Ond\u0159ej and Constantin, Alexandra and Herbst, Evan. 2007. Moses: Open Source Toolkit for Statistical Machine Translation. Proceedings of ACL, (Jun):177-180.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "OpenNMT: Open-Source Toolkit for Neural Machine Translation", "authors": [ { "first": "Guillaume", "middle": [], "last": "Klein", "suffix": "" }, { "first": "Yoon", "middle": [], "last": "Kim", "suffix": "" }, { "first": "Yuntian", "middle": [], "last": "Deng", "suffix": "" }, { "first": "Jean", "middle": [], "last": "Senellart", "suffix": "" }, { "first": "Alexander", "middle": [], "last": "Rush", "suffix": "" } ], "year": 2017, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "67--72", "other_ids": {}, "num": null, "urls": [], "raw_text": "Klein, Guillaume and Kim, Yoon and Deng, Yuntian and Senellart, Jean and Rush, Alexander. 2017. OpenNMT: Open-Source Toolkit for Neural Machine Translation. Proceedings of ACL, 67-72.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Adam: A method for stochastic optimization", "authors": [ { "first": "P", "middle": [], "last": "Diederik", "suffix": "" }, { "first": "Jimmy", "middle": [], "last": "Kingma", "suffix": "" }, { "first": "", "middle": [], "last": "Ba", "suffix": "" } ], "year": 2015, "venue": "Proceedings of ICLR", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. Proceedings of ICLR.", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "A Structured Self-attentive Sentence Embedding", "authors": [ { "first": "Zhouhan", "middle": [], "last": "Lin", "suffix": "" }, { "first": "Minwei", "middle": [], "last": "Feng", "suffix": "" }, { "first": "Cicero", "middle": [], "last": "Nogueira", "suffix": "" }, { "first": "Mo", "middle": [], "last": "Santos", "suffix": "" }, { "first": "Bing", "middle": [], "last": "Yu", "suffix": "" }, { "first": "Bowen", "middle": [], "last": "Xiang", "suffix": "" }, { "first": "Yoshua", "middle": [], "last": "Zhou", "suffix": "" }, { "first": "", "middle": [], "last": "Bengio", "suffix": "" } ], "year": 2017, "venue": "Proceedings of ICLR", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Zhouhan Lin and Minwei Feng and Cicero Nogueira dos Santos and Mo Yu and Bing Xiang and Bowen Zhou and Yoshua Bengio. 2017. A Structured Self-attentive Sentence Embedding. Proceedings of ICLR.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "METEOR: An Automatic Metric for MT Evaluation with High Levels of Correlation with Human Judgments", "authors": [ { "first": "Alon", "middle": [], "last": "Lavie", "suffix": "" }, { "first": "Abhaya", "middle": [], "last": "Agarwal", "suffix": "" } ], "year": 2007, "venue": "Proceedings of WMT", "volume": "", "issue": "", "pages": "228--231", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lavie, Alon and Agarwal, Abhaya. 2007. METEOR: An Automatic Metric for MT Evaluation with High Levels of Correlation with Human Judgments. Proceedings of WMT, (Jun):228-231.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "Using whole document context in neural machine translation", "authors": [ { "first": "Valentin", "middle": [], "last": "Mace", "suffix": "" }, { "first": "Christophe", "middle": [], "last": "Servan", "suffix": "" } ], "year": 2019, "venue": "Proceedings of IWSLT", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Valentin Mace and Christophe Servan. 2019. Using whole document context in neural machine translation. In Proceedings of IWSLT.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "Document Context Neural Machine Translation with Memory Networks", "authors": [ { "first": "Sameen", "middle": [], "last": "Maruf", "suffix": "" }, { "first": "Gholamreza", "middle": [], "last": "Haffari", "suffix": "" } ], "year": 2018, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "1275--1284", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sameen Maruf and Gholamreza Haffari. 2018. Document Context Neural Machine Translation with Memory Networks. Proceedings of ACL, 1275-1284.", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "Selective Attention for Context-aware Neural Machine Translation", "authors": [ { "first": "Sameen", "middle": [], "last": "Maruf", "suffix": "" }, { "first": "F", "middle": [ "T" ], "last": "Andr\u00e9", "suffix": "" }, { "first": "Gholamreza", "middle": [], "last": "Martins", "suffix": "" }, { "first": "", "middle": [], "last": "Haffari", "suffix": "" } ], "year": 2019, "venue": "Proceedings of NAACL", "volume": "", "issue": "", "pages": "3092--3102", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sameen Maruf and Andr\u00e9 F. T. Martins and Gholamreza Haffari. 2019. Selective Attention for Context-aware Neural Machine Translation. Proceedings of NAACL, 3092-3102.", "links": null }, "BIBREF19": { "ref_id": "b19", "title": "Validation of an automatic metric for the ac-curacy of pronoun translation (APT)", "authors": [ { "first": "Lesly", "middle": [], "last": "Miculicich Werlen", "suffix": "" }, { "first": "Andrei", "middle": [], "last": "Popescu-Belis", "suffix": "" } ], "year": 2017, "venue": "Proceedings of the Third Workshop on Discourse in Machine Translation", "volume": "", "issue": "", "pages": "17--25", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lesly Miculicich Werlen and Andrei Popescu-Belis. 2017. Validation of an automatic metric for the ac-curacy of pronoun translation (APT). Proceedings of the Third Workshop on Discourse in Machine Translation, 17-25.", "links": null }, "BIBREF20": { "ref_id": "b20", "title": "Document-Level Neural Machine Translation with Hierarchical Attention Networks", "authors": [ { "first": "Lesly", "middle": [], "last": "Miculicich", "suffix": "" }, { "first": "Dhananjay", "middle": [], "last": "Ram", "suffix": "" }, { "first": "Nikolaos", "middle": [], "last": "Pappas", "suffix": "" }, { "first": "James", "middle": [], "last": "Henderson", "suffix": "" } ], "year": 2018, "venue": "Proceedings of EMNLP", "volume": "", "issue": "", "pages": "2947--2954", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lesly Miculicich and Dhananjay Ram and Nikolaos Pappas and James Henderson. 2018. Document-Level Neural Machine Translation with Hierarchical Attention Networks. Proceedings of EMNLP, 2947-2954.", "links": null }, "BIBREF21": { "ref_id": "b21", "title": "WIT3: Web Inventory of Transcribed and Translated Talks", "authors": [ { "first": "Mauro", "middle": [], "last": "Cettolo", "suffix": "" }, { "first": "Christian", "middle": [], "last": "Girardi", "suffix": "" }, { "first": "Marcello", "middle": [], "last": "Federico", "suffix": "" } ], "year": 2012, "venue": "Proceedings of EAMT", "volume": "", "issue": "", "pages": "261--268", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mauro Cettolo and Christian Girardi and Marcello Federico. 2012. WIT3: Web Inventory of Transcribed and Translated Talks. Proceedings of EAMT, 261-268.", "links": null }, "BIBREF22": { "ref_id": "b22", "title": "BLEU: a method for automatic evaluation of machine translation", "authors": [ { "first": "Kishore", "middle": [], "last": "Papineni", "suffix": "" }, { "first": "Salim", "middle": [], "last": "Roukos", "suffix": "" }, { "first": "Ward", "middle": [], "last": "Todd", "suffix": "" }, { "first": "Wei-Jing", "middle": [], "last": "Zhu", "suffix": "" } ], "year": 2002, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "311--318", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kishore Papineni and Salim Roukos and Ward Todd and Wei-Jing Zhu. 2002. BLEU: a method for automatic evaluation of machine translation. Proceedings of ACL, 311-318.", "links": null }, "BIBREF23": { "ref_id": "b23", "title": "Neural Machine Translation of Rare Words with Subword Units", "authors": [ { "first": "Rico", "middle": [], "last": "Sennrich", "suffix": "" }, { "first": "Barry", "middle": [], "last": "Haddow", "suffix": "" }, { "first": "Alexandra", "middle": [], "last": "Birch", "suffix": "" } ], "year": 2016, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "1715--1725", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rico Sennrich and Barry Haddow and Alexandra Birch. 2016. Neural Machine Translation of Rare Words with Subword Units. In Proceedings of ACL, 1715-1725.", "links": null }, "BIBREF24": { "ref_id": "b24", "title": "Sequence to sequence learning with neural networks", "authors": [ { "first": "Ilya", "middle": [], "last": "Sutskever", "suffix": "" }, { "first": "Oriol", "middle": [], "last": "Vinyals", "suffix": "" }, { "first": "Quoc V", "middle": [], "last": "Le", "suffix": "" } ], "year": 2014, "venue": "Advances in neural information processing systems", "volume": "", "issue": "", "pages": "3104--3112", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ilya Sutskever, Oriol Vinyals, and Quoc V Le. 2014. Sequence to sequence learning with neural networks. Advances in neural information processing systems, 3104--3112.", "links": null }, "BIBREF25": { "ref_id": "b25", "title": "Hierarchical Modeling of Global Context for Document-Level Neural Machine Translation", "authors": [ { "first": "Xin", "middle": [], "last": "Tan", "suffix": "" }, { "first": "Longyin", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Deyi", "middle": [], "last": "Xiong", "suffix": "" }, { "first": "Guodong", "middle": [], "last": "Zhou", "suffix": "" } ], "year": 2019, "venue": "Proceedings of EMNLP-IJCNLP", "volume": "", "issue": "", "pages": "1576--1585", "other_ids": {}, "num": null, "urls": [], "raw_text": "Xin Tan and Longyin Zhang and Deyi Xiong and Guodong Zhou. 2019. Hierarchical Modeling of Global Context for Document-Level Neural Machine Translation. In Proceedings of EMNLP-IJCNLP, 1576-1585.", "links": null }, "BIBREF26": { "ref_id": "b26", "title": "Enhancing Grammatical Cohesion: Generating Transitional Expressions for SMT", "authors": [ { "first": "Mei", "middle": [], "last": "Tu", "suffix": "" }, { "first": "Yu", "middle": [], "last": "Zhou", "suffix": "" }, { "first": "Chengqing", "middle": [], "last": "Zong", "suffix": "" } ], "year": 2014, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "850--860", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mei Tu and Yu Zhou and Chengqing Zong. 2014. Enhancing Grammatical Cohesion: Generating Transitional Expressions for SMT. Proceedings of ACL, 850-860.", "links": null }, "BIBREF27": { "ref_id": "b27", "title": "Neural Machine Translation with Extended Context", "authors": [ { "first": "J\u00f6rg", "middle": [], "last": "Tiedemann", "suffix": "" }, { "first": "Yves", "middle": [], "last": "Scherrer", "suffix": "" } ], "year": 2017, "venue": "Proceedings of the Third Workshop on Discourse in Machine Translation", "volume": "", "issue": "", "pages": "82--92", "other_ids": {}, "num": null, "urls": [], "raw_text": "Tiedemann, J\u00f6rg and Scherrer, Yves. 2017. Neural Machine Translation with Extended Context. In Proceedings of the Third Workshop on Discourse in Machine Translation, \"82-92.", "links": null }, "BIBREF28": { "ref_id": "b28", "title": "Transactions of the Association for Computational Linguistics", "authors": [ { "first": "Zhaopeng", "middle": [], "last": "Tu", "suffix": "" }, { "first": "Yang", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Shuming", "middle": [], "last": "Shi", "suffix": "" }, { "first": "Tong", "middle": [], "last": "Zhang", "suffix": "" } ], "year": 2018, "venue": "", "volume": "", "issue": "6", "pages": "407--420", "other_ids": {}, "num": null, "urls": [], "raw_text": "Zhaopeng Tu and Yang Liu and Shuming Shi and Tong Zhang. 2018. Transactions of the Association for Compu- tational Linguistics. Transactions of the Association for Computational Linguistics, (6):407-420.", "links": null }, "BIBREF29": { "ref_id": "b29", "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": "Proceedings of NIPS", "volume": "", "issue": "", "pages": "5998--6008", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ashish Vaswani and Noam Shazeer and Niki Parmar and Jakob Uszkoreit and Llion Jones and Aidan N.Gomez and Lukasz Kaiser and Illia Polosukhin. 2017. Attention Is All You Need. In Proceedings of NIPS, 5998-6008.", "links": null }, "BIBREF30": { "ref_id": "b30", "title": "Context-Aware Neural Machine Translation Learns Anaphora Resolution", "authors": [ { "first": "Elena", "middle": [], "last": "Voita", "suffix": "" }, { "first": "Pavel", "middle": [], "last": "Serdyukov", "suffix": "" }, { "first": "Rico", "middle": [], "last": "Sennrich", "suffix": "" }, { "first": "Ivan", "middle": [], "last": "Titov", "suffix": "" } ], "year": 2018, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "1264--1274", "other_ids": {}, "num": null, "urls": [], "raw_text": "Elena Voita and Pavel Serdyukov and Rico Sennrich and Ivan Titov. 2018. Context-Aware Neural Machine Translation Learns Anaphora Resolution. Proceedings of ACL, 1264-1274.", "links": null }, "BIBREF31": { "ref_id": "b31", "title": "When a Good Translation is Wrong in Context: Context-Aware Machine Translation Improves on Deixis, Ellipsis, and Lexical Cohesion", "authors": [ { "first": "Elena", "middle": [], "last": "Voita", "suffix": "" }, { "first": "Rico", "middle": [], "last": "Sennrich", "suffix": "" }, { "first": "Ivan", "middle": [], "last": "Titov", "suffix": "" } ], "year": 2019, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "1198--1212", "other_ids": {}, "num": null, "urls": [], "raw_text": "Elena Voita and Rico Sennrich and Ivan Titov. 2019. When a Good Translation is Wrong in Context: Context- Aware Machine Translation Improves on Deixis, Ellipsis, and Lexical Cohesion. Proceedings of ACL, 1198- 1212.", "links": null }, "BIBREF32": { "ref_id": "b32", "title": "Contextual Neural Machine Translation Improves Translation of Cataphoric Pronouns", "authors": [ { "first": "Kayyen", "middle": [], "last": "Wong", "suffix": "" }, { "first": "Sameen", "middle": [], "last": "Maruf", "suffix": "" }, { "first": "Gholamreza", "middle": [], "last": "Haffari", "suffix": "" } ], "year": 2020, "venue": "Proceedings of ACL", "volume": "", "issue": "", "pages": "2826--2831", "other_ids": {}, "num": null, "urls": [], "raw_text": "KayYen Wong and Sameen Maruf and Gholamreza Haffari. 2020. Contextual Neural Machine Translation Im- proves Translation of Cataphoric Pronouns. In Proceedings of ACL, 2826-2831.", "links": null }, "BIBREF33": { "ref_id": "b33", "title": "Exploiting cross-sentence context for neural machine translation", "authors": [ { "first": "Longyue", "middle": [], "last": "Wang", "suffix": "" }, { "first": "Zhaopeng", "middle": [], "last": "Tu", "suffix": "" }, { "first": "Andy", "middle": [], "last": "Way", "suffix": "" }, { "first": "Qun", "middle": [], "last": "Liu", "suffix": "" } ], "year": 2017, "venue": "Proceedings of EMNLP", "volume": "", "issue": "", "pages": "2826--2831", "other_ids": {}, "num": null, "urls": [], "raw_text": "Longyue Wang and Zhaopeng Tu and Andy Way and Qun Liu. 2017. Exploiting cross-sentence context for neural machine translation. In Proceedings of EMNLP, 2826-2831.", "links": null }, "BIBREF34": { "ref_id": "b34", "title": "Validation of an Automatic Metric for the Accuracy of Pronoun Translation (APT)", "authors": [ { "first": "Lesly", "middle": [], "last": "Miculicich Werlen", "suffix": "" }, { "first": "Andrei", "middle": [], "last": "Popescu-Belis", "suffix": "" } ], "year": 2017, "venue": "Proceedings of Workshop on Discourse in Machine Translation", "volume": "", "issue": "", "pages": "17--25", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lesly Miculicich Werlen and Andrei Popescu-Belis. 2017. Validation of an Automatic Metric for the Accuracy of Pronoun Translation (APT). Proceedings of Workshop on Discourse in Machine Translation, 17-25.", "links": null }, "BIBREF35": { "ref_id": "b35", "title": "Lexical Chain Based Cohesion Models for Document-Level Statistical Machine Translation", "authors": [ { "first": "Deyi", "middle": [], "last": "Xiong", "suffix": "" }, { "first": "Yang", "middle": [], "last": "Ding", "suffix": "" }, { "first": "Min", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "", "middle": [], "last": "Chew Lim Tan", "suffix": "" } ], "year": 2013, "venue": "Proceedings of EMNLP", "volume": "", "issue": "", "pages": "1563--1573", "other_ids": {}, "num": null, "urls": [], "raw_text": "Deyi Xiong and Yang Ding and Min Zhang and Chew Lim Tan. 2013. Lexical Chain Based Cohesion Models for Document-Level Statistical Machine Translation. Proceedings of EMNLP, 1563-1573.", "links": null }, "BIBREF36": { "ref_id": "b36", "title": "Modeling coherence for discourse neural machine translation", "authors": [ { "first": "Hao", "middle": [], "last": "Xiong", "suffix": "" }, { "first": "Zhongjun", "middle": [], "last": "He", "suffix": "" }, { "first": "Hua", "middle": [], "last": "Wu", "suffix": "" }, { "first": "Haifeng", "middle": [], "last": "Wang", "suffix": "" } ], "year": 2019, "venue": "Proceedings of AAAI", "volume": "", "issue": "", "pages": "7338--7345", "other_ids": {}, "num": null, "urls": [], "raw_text": "Hao Xiong and Zhongjun He and Hua Wu and Haifeng Wang. 2019. Modeling coherence for discourse neural machine translation. In Proceedings of AAAI, 7338-7345.", "links": null }, "BIBREF37": { "ref_id": "b37", "title": "Enhancing Context Modeling with a Query-Guided Capsule Network for Document-level Translation. Proceedings of EMNLP", "authors": [ { "first": "Zhengxin", "middle": [], "last": "Yang", "suffix": "" }, { "first": "Jinchao", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Fandong", "middle": [], "last": "Meng", "suffix": "" }, { "first": "Shuhao", "middle": [], "last": "Gu", "suffix": "" }, { "first": "Yang", "middle": [], "last": "Feng", "suffix": "" }, { "first": "Jie", "middle": [], "last": "Zhou", "suffix": "" } ], "year": 2019, "venue": "", "volume": "", "issue": "", "pages": "1527--1537", "other_ids": {}, "num": null, "urls": [], "raw_text": "Zhengxin Yang and Jinchao Zhang and Fandong Meng and Shuhao Gu and Yang Feng and Jie Zhou. 2019. Enhancing Context Modeling with a Query-Guided Capsule Network for Document-level Translation. Pro- ceedings of EMNLP, 1527-1537.", "links": null }, "BIBREF38": { "ref_id": "b38", "title": "Improving the Transformer Translation Model with Document-Level Context", "authors": [ { "first": "Jiacheng", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Huanbo", "middle": [], "last": "Luan", "suffix": "" }, { "first": "Maosong", "middle": [], "last": "Sun", "suffix": "" }, { "first": "Feifei", "middle": [], "last": "Zhai", "suffix": "" }, { "first": "Jingfang", "middle": [], "last": "Xu", "suffix": "" }, { "first": "Min", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Yang", "middle": [], "last": "Liu", "suffix": "" } ], "year": 2018, "venue": "Proceedings of EMNLP", "volume": "", "issue": "", "pages": "533--542", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jiacheng Zhang and Huanbo Luan and Maosong Sun and Feifei Zhai and Jingfang Xu and Min Zhang and Yang Liu. 2018. Improving the Transformer Translation Model with Document-Level Context. Proceedings of EMNLP, 533-542.", "links": null } }, "ref_entries": { "TABREF1": { "num": null, "text": "\u8bcd\u7ea7\u4e0a\u4e0b\u6587 37.05 \u2021 43.79 \u2021 44.57 \u2021 41.98 \u2020 42.10 \u2021 32.49 \u2021 40.61 \u2021 + \u590d\u5408\u4e0a\u4e0b\u6587 37.46 \u2021 44.08 \u2021 44.86 \u2021 42.87 \u2021 42.16 \u2021 32.74 \u2021 41.10 \u2021", "html": null, "type_str": "table", "content": "
\u6a21 \u6a21 \u6a21\u578b \u578b \u578bMT06 MT02 MT03 MT04 MT05 MT08 All
Transformer36.2742.7143.5141.2541.0731.5439.64
+ Transformer(Zhang et al., 2018)36.2042.4143.1241.0240.9331.4939.53
Transformer-DocNMT(Zhang et al., 2018) 37.1243.2943.7041.4241.8432.3640.22
\u8868 2: \u672c\u6587\u6a21\u578b\u4e2d-\u82f1\u7ffb\u8bd1\u4efb\u52a1\u7684\u6027\u80fd(BLEU). \u2020 \u548c \u2021\u8868\u793a\u4e0eTransformer\u57fa\u51c6\u6a21\u578b\u76f8\u6bd4\u663e\u8457\u6027p\u503c\u5c0f
\u4e8e0.05/0.01
\u6a21 \u6a21 \u6a21\u578b \u578b \u578b\u897f \u897f \u897f-\u82f1 \u82f1 \u82f1\u82f1 \u82f1 \u82f1-\u5fb7 \u5fb7 \u5fb7
BLEU Meteor BLEU Meteor
Transformer35.5034.6023.0243.66
+ \u8bcd\u7ea7\u4e0a\u4e0b\u658737.5936.5024.4045.19
+ \u590d\u5408\u4e0a\u4e0b\u658737.7536.8324.9845.70
Transformer-DocNMT
\u641c\u7d22\u7684\u5927\u5c0f\u8bbe\u7f6e\u4e3a5\uff0cdropout\u8bbe\u7f6e\u4e3a0.1\u3002\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u5c06\u6279\u5927\u5c0f\u8bbe\u7f6e\u4e3a8192\u4e2a\u5b57\u7b26\u5e76\u4f7f
\u7528\u03b2 1 = 0.1\u7684Adam\u4f18\u5316\u5668\u5bf9\u6a21\u578b\u8fdb\u884c\u4f18\u5316Kingma and Ba., (2015)\u3002
3.3 \u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65b9 \u65b9 \u65b9\u5f0f \u5f0f \u5f0f
\u53d7Zhang et al., (2018)\u542f\u53d1\uff0c\u672c\u6587\u4f7f\u7528\u4e24\u6b65\u8bad\u7ec3\u7b56\u7565\u5145\u5206\u5229\u7528\u53e5\u5b50\u7ea7\u5e73\u884c\u8bed\u6599\u5728\u8bad\u7ec3\u901f\u5ea6\u53ca
\u8ba1\u7b97\u5f00\u9500\u7b49\u65b9\u9762\u7684\u4f18\u52bf\u3002\u5728\u7b2c\u4e00\u6b65\u8bad\u7ec3\u4e2d\u4f7f\u7528\u5e73\u884c\u53e5\u5bf9\u8bed\u6599\u5bf9\u53e5\u5b50\u7ea7\u522b\u53c2\u6570\u8fdb\u884c\u8bad\u7ec3,\u5728\u7b2c\u4e8c\u6b65
\u8bad\u7ec3\u4e2d\u4f7f\u7528\u542b\u6709\u7bc7\u7ae0\u4fe1\u606f\u7684\u8bed\u6599\u8bad\u7ec3\u7bc7\u7ae0\u7ea7\u53c2\u6570,\u8be5\u90e8\u5206\u53c2\u6570\u5305\u62ec\u4e0d\u540c\u5c42\u6b21\u4f9d\u8d56\u6743\u91cd\u7684\u83b7\u53d6\uff0c\u542b
\u6709\u5206\u5c42\u7ed3\u6784\u4fe1\u606f\u5168\u5c40\u4e0a\u4e0b\u6587\u7684\u83b7\u53d6\uff0c\u53ca\u7ed3\u5408\u4e0a\u4e0b\u6587\u4e0e\u7f16\u7801\u5668\u8f93\u51fa\u7684\u95e8\u63a7\u7b49\u3002\u4e24\u6b65\u8bad\u7ec3\u4f7f\u7528\u7684\u5e73
0 \u8bad\u7ec3\u96c6\u7531LDC2002T01, LDC2004T07, LDC2005T06, LDC2005T10, LDC2009T02, LDC2009T15, LDC2010T03\u7ec4
\u6210\u3002
1 https://github.com/fxsjy/jieba
2 https://github.com/OpenNMT/OpenNMT-py
" }, "TABREF3": { "num": null, "text": "\u4e0d \u4e0d \u4e0d\u540c \u540c \u540c\u4e0a \u4e0a \u4e0a\u4e0b \u4e0b \u4e0b\u6587 \u6587 \u6587\u5229 \u5229 \u5229\u7528 \u7528 \u7528\u65b9 \u65b9 \u65b9\u5f0f \u5f0f \u5f0f \u65b0\u534e \u534e \u534e\u793e \u793e \u793e\u7ebd \u7ebd \u7ebd\u7ea6 \u7ea6 \u7ea62 \u6708 \u6708 \u670812 \u65e5 \u65e5 \u65e5\u7535 \u7535 \u7535( \u8bb0 \u8bb0 \u8bb0\u8005 \u8005 \u8005\u8303\u8303 \u8303\u5c0f \u5c0f \u5c0f\u6797 \u6797 \u6797) \u968f \u968f \u968f\u7740 \u7740 \u7740\u9ad8 \u9ad8 \u9ad8\u5206 \u5206 \u5206\u8fa8 \u8fa8 \u8fa8\u7387 \u7387 \u7387\u6570 \u6570 \u6570\u5b57 \u5b57 \u5b57\u7535 \u7535 \u7535\u89c6 \u89c6 \u89c6\u673a \u673a \u673a( HDTV ) \u7684 \u7684 \u7684\u552e \u552e \u552e\u4ef7 \u4ef7 \u4ef7. . . . . . \u521a\u843d\u5e55\u7684\u7f8e\u56fd\" \u8d85\u7ea7\u7897\" \u7f8e\u5f0f\u6a44\u6984\u7403\u8d5b\u5df2\u7ecf\u6210\u4e3a\u7f8e\u56fd\u5e7f\u64ad\u4e1a\u8005\u8fce\u63a5\u7535\u89c6\u9ad8\u6e05. . . . . . \u3002 \u636e\u300a\u91d1\u878d\u65f6\u62a5\u300b\u65e5\u524d\u62a5\u9053, \u7f8e\u56fd\u56fd\u4f1a\u6700\u8fd1\u901a\u8fc7\u4e00\u9879\u6cd5\u6848, \u89c4\u5b9a\u7f8e\u56fd\u5e7f\u64ad\u4e1a\u8005\u5fc5\u987b\u5728. . . . . . \u3002 \u8fd9\u610f\u5473\u7740\u7f8e\u56fd\u7535\u89c6\u8282\u76ee\u64ad\u653e\u7684\u5168\u9762\u6570\u5b57\u5316\u5df2\u7ecf\u6709\u4e86\u660e\u786e\u7684\u65f6\u95f4\u8868\u3002 \u9ad8\u5206\u8fa8\u7387\u6570\u5b57\u7535\u89c6\u7684\u666e\u53ca\u4e00\u65b9\u9762\u8981\u9760\u6d88\u8d39\u8005\u8d2d\u4e70\u7535\u89c6\u673a, \u4e00\u65b9\u9762. . . . . . \u3002 \u6570\u5b57\u7535\u89c6\u867d\u7136\u5728\u8fc7\u53bb\u51e0\u5e74\u88ab\u4e00\u518d\u63d0\u53ca, \u4f46\u59cb\u7ec8\u90fd\u672a\u6210\u4e3a\u73b0\u5b9e\u3002 \u8fd9\u4e2a\u95ee\u9898\u5728\u4eca\u5e74\u51fa\u73b0\u8f6c\u673a\u3002 \u6839\u636e\u7f8e\u56fd\u6d88\u8d39\u7535\u5b50\u534f\u4f1a\u7684\u9884\u6d4b, \u6570\u5b57\u7535\u89c6\u673a\u4eca\u5e74\u5728\u7f8e\u9500\u552e\u91cf\u5c06\u9996\u6b21\u8d85\u8fc7\u4f20\u7edf\u7535\u89c6\u673a, . . . . . . \u3002 \u4ee3\u8bcd\u7ffb\u8bd1\u8d28\u91cf(APT)\u5bf9\u6bd4 \u6e90\u7aef . . . . . . \u4eca\u5929\u665a\u4e0a\u7684\u5341\u4e00\u4e8c\u70b9\u949f\u5de6\u53f3\u5427\u3002 \u53c2\u8003\u7ffb\u8bd1 . . . . . . itwill arrive around 11 : 00 or 12 : 00 tonight . Transformer . . . . . . that about 11.2 pm today . + \u8bcd\u7ea7\u4e0a\u4e0b\u6587 . . . . . . it will be around 11.2 pm today . + \u590d\u5408\u4e0a\u4e0b\u6587 . . . . . . it will be around 12 : 00 tonight . \u6e90\u7aef . . . . . . \u4e00\u6b3e\u975e\u5e38\u4f18\u79c0\u7684\u57fa\u4e8ePHP \u548cMySQL \u6570\u636e\u5e93\u7684\u793e\u533a\u7a0b \u7a0b \u7a0b\u5e8f \u5e8f \u5e8f \u3002 \u53c2\u8003\u7ffb\u8bd1 . . . . . . an excellent community software based on php and mysql database . Transformer . . . . . . an extremely outstanding community procedure based on the php. . . . . . . + \u8bcd\u7ea7\u4e0a\u4e0b\u6587 . . . . . . an extremely outstanding community process based on php a. . . . . . . + \u590d\u5408\u4e0a\u4e0b\u6587 . . . . . . an extremely outstanding communityprograme based on php a. . . . . . .", "html": null, "type_str": "table", "content": "
\u4eab\uff0c\u53c2\u6570\u589e\u52a0\u6570\u91cf\u8f83\u5c11\u3002\u7efc\u5408\u8003\u8651\u4e86\u6a21\u578b\u6027\u80fd\u4e0e\u53c2\u6570\u53ca\u8ba1\u7b97\u5f00\u9500\u4e4b\u95f4\u7684\u5e73\u8861\u3002\u540c\u65f6\uff0c\u672c\u6587\u5145\u5206 \u57284.2\u7684\u5b9e\u9a8c\u4e2d\uff0c\u672c\u6587\u4f7f\u7528\u6765\u81ea\u6574\u4e2a\u7bc7\u7ae0\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u4e0e\u7ffb\u8bd1\u6a21\u578b\u7ed3\u5408\uff0c\u5176\u7ffb\u8bd1\u8d28\u91cf\u6bd4\u4ec5\u4f7f
\u5229\u7528\u53e5\u5b50\u7ea7\u5e73\u884c\u8bed\u6599\u5728\u8bad\u7ec3\u65f6\u95f4\u4e0a\u7684\u4f18\u52bf\uff0c\u901a\u8fc7\u4e24\u6b65\u8bad\u7ec3\u6cd5\u4f7f\u5f97\u8bad\u7ec3\u65f6\u95f4\u76f8\u6bd4\u5176\u4ed6\u6a21\u578b\u6ca1\u6709\u660e \u7528\u524d\u6587\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u53d6\u5f97\u4e86\u663e\u8457\u63d0\u5347\u3002\u51fa\u4e8e\u63a2\u7d22\u524d\u540e\u4e0a\u4e0b\u6587\u5bf9\u7bc7\u7ae0\u7ffb\u8bd1\u8d28\u91cf\u5f71\u54cd\u7684\u76ee\u7684\uff0c\u6211\u4eec\u5bf9
\u663e\u589e\u52a0\u3002 \u5168\u5c40\u4e0a\u4e0b\u6587\u7684\u5206\u5e03\u5c55\u5f00\u5982\u4e0b\u5206\u6790\u4e0e\u5b9e\u9a8c\u3002\u672c\u6587\u5229\u7528\u53e5\u5b50\u7ea7\u4f9d\u8d56\u6743\u91cd\u5bf9\u5f00\u53d1\u96c6\u4e2d\u7684\u7bc7\u7ae0\u53e5\u5b50\u8fdb\u884c
\u7edf\u8ba1\uff0c\u83b7\u53d6\u7bc7\u7ae0\u4e2d\u5bf9\u6240\u6709\u53e5\u5b50\u800c\u8a00\u90fd\u6700\u91cd\u8981\u7684\u53e5\u5b50\uff0c\u5e76\u5c06\u8be5\u53e5\u4f7f\u7528\u52a0\u7c97\u5b57\u4f53\u8868\u793a\u3002\u88687\u4e2d\u7684\u6837\u4f8b \u53ef\u4ee5\u76f4\u89c2\u8868\u660e\uff0c\u5f53\u524d\u53e5\u7684\u4e0a\u4e0b\u6587\u4e0d\u4e00\u5b9a\u53ea\u5b58\u5728\u4e8e\u90bb\u8fd1\u8bed\u53e5\u3002\u8be5\u73b0\u8c61\u4e0d\u4ec5\u5b58\u5728\u4e8e\u672c\u6587\u6240\u4e3e\u6837\u4f8b\uff0c 4.2 \u5982\u56fe4\u6240\u793a\uff0c\u4e3a\u4e86\u89c2\u5bdf\u4e0d\u540c\u65b9\u5f0f\u5229\u7528\u4e0a\u4e0b\u6587\u5bf9\u7ffb\u8bd1\u8d28\u91cf\u7684\u5f71\u54cd\uff0c\u6211\u4eec\u5c1d\u8bd5\u5728\u89e3\u7801\u5668\u7aef\u589e\u52a0\u4e13 \u4e5f\u5b58\u5728\u4e8e\u672c\u6587\u5b9e\u9a8c\u6240\u4f7f\u7528\u7684\u5176\u4ed6\u7bc7\u7ae0\u8bed\u6599\u4e2d\u3002
\u95e8\u9488\u5bf9\u5168\u5c40\u4e0a\u4e0b\u6587\u7684\u6ce8\u610f\u529b\u5c42\u7ed3\u6784\u3002\u5b9e\u9a8c\u7ed3\u679c\u5bf9\u6bd4\u5982\u88685\uff0c\u76f8\u6bd4\u672c\u6587\u4f7f\u7528\u7684\u76f4\u63a5\u878d\u5408\u4e0d\u540c\u5c42\u6b21\u4f9d \u8d56\u5173\u7cfb\u7684\u65b9\u6cd5\uff0c\u589e\u52a0\u6ce8\u610f\u529b\u5c42\u7684\u65b9\u6cd5\u589e\u52a0\u4e86\u6a21\u578b\u53c2\u6570\u548c\u8ba1\u7b97\u5f00\u9500\uff0c\u5728\u7ffb\u8bd1\u6027\u80fd\u65b9\u9762\u6ca1\u6709\u53d6\u5f97\u6709 4.4 \u540d \u540d \u540d\u8bcd \u8bcd \u8bcd\u4e0e \u4e0e \u4e0e\u4ee3 \u4ee3 \u4ee3\u8bcd \u8bcd \u8bcd\u7ffb \u7ffb \u7ffb\u8bd1 \u8bd1 \u8bd1
\u610f\u4e49\u7684\u63d0\u5347\u3002 \u4e3a\u4e86\u89c2\u5bdf\u672c\u6587\u63d0\u51fa\u7684\u5c42\u6b21\u5316\u7ed3\u6784\u5168\u5c40\u4e0a\u4e0b\u6587\u6a21\u578b\u662f\u5982\u4f55\u63d0\u9ad8\u7ffb\u8bd1\u8d28\u91cf\u7684\uff0c\u6211\u4eec\u5bf9\u4ee3\u8bcd\u548c
\u540d\u8bcd\u7684\u7ffb\u8bd1\u8fdb\u884c\u8fdb\u4e00\u6b65\u7684\u5b9e\u9a8c\u4e0e\u5206\u6790\u3002\u5728\u4ee3\u8bcd\u7ffb\u8bd1\u4e2d\uff0c\u6211\u4eec\u4f7f\u7528Miculicich et al., (2017)\u63d0\u51fa
softmax \u7684APT\u5ea6\u91cf\u6807\u51c6\u8bc4\u4ef7\u4e2d-\u82f1\u7ffb\u8bd1\u5b9e\u9a8c\u4ee3\u8bcd\u7ffb\u8bd1\u7684\u51c6\u786e\u6027\u3002\u5982\u88688\u6240\u793a\uff0c\u7ed3\u679c\u8868\u660e\u672c\u6587\u63d0\u51fa\u7684\u591a\u5c42
\u7ed3\u6784\u5168\u5c40\u4e0a\u4e0b\u6587\u6a21\u578b\u80fd\u591f\u66f4\u597d\u5730\u6355\u6349\u5230\u6bcf\u4e2a\u8bcd\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\uff0c\u4ece\u800c\u63d0\u5347\u4e2d-\u82f1\u7ffb\u8bd1\u5b9e\u9a8c\u5728\u4ee3\u8bcd\u7ffb
\u8bd1\u7684\u51c6\u786e\u7387\u3002Feed Forward
Model Transformer + \u590d\u5408\u4e0a\u4e0b\u6587 70.24 74.22 69.02 65.45 68.29 71.91 69.40 MT06 MT02 MT03 MT04 MT05 MT08 All Encoder-Decoder 69.54 73.67 68.41 65.32 67.71 71.60 68.68 Attention
Masked Self-Attention Global Context Attention Word Embedding \u56fe 4: \u5728\u89e3\u7801\u5668\u4e2d\u589e\u52a0\u5168\u5c40\u4e0a\u4e0b\u6587\u6ce8\u610f\u529b\u5c42\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u7ed3\u5408\u65b9\u5f0f Encoder Output Global Context \u8868 8: \u8868 9: \u4ee3\u8bcd\u7ffb\u8bd1\u6837\u4f8b
\u7ed3 \u7ed3 \u7ed3\u5408 \u5408 \u5408\u65b9 \u65b9 \u65b9\u5f0f \u5f0f \u5f0f \u76f4\u63a5\u878d\u5408 \u589e\u52a0\u6ce8\u610f\u529b\u5c42 41.09 BLEU 41.10 \u672c\u6587\u5728\u88689\u4e2d\u5217\u4e3e\u4e86\u4e00\u4e2a\u7ffb\u8bd1\u4f8b\u5b50\u8fdb\u4e00\u6b65\u89c2\u5bdf\u5c42\u6b21\u5316\u5168\u5c40\u4e0a\u4e0b\u6587\u5bf9\u4ee3\u8bcd\u7ffb\u8bd1\u7684\u5e2e\u52a9\u3002\u901a\u8fc7\u5b9e \u4e0a \u4e0a \u4e0a\u4e0b \u4e0b \u4e0b\u6587 \u6587 \u6587\u6765 \u6765 \u6765\u6e90 \u6e90 \u6e90 BLEU \u5f53\u524d\u53e5\u524d\u6587 40.31 \u4f8b\u53ef\u4ee5\u770b\u51fa\u672c\u6587\u63d0\u51fa\u7684\u6a21\u578b\u53ef\u4ee5\u8f83\u597d\u5730\u63a8\u65ad\u51fa\u6f5c\u5728\u4ee3\u8bcd\uff0c\u4ece\u800c\u9a8c\u8bc1\u4e86\u8be5\u6a21\u578b\u7684\u4ee3\u8bcd\u7ffb\u8bd1\u6027\u80fd\u3002 \u5f53\u524d\u53e5\u540e\u6587 40.49 \u65e0\u4e0a\u4e0b\u6587 39.64 \u5bf9\u4e8e\u540d\u8bcd\u7ffb\u8bd1\u7684\u5206\u6790\uff0c\u672c\u6587\u5c06\u5c55\u793a\u53e6\u4e00\u4e2a\u6837\u4f8b\u3002
\u8868 5: \u4e0d\u540c\u4e0a\u4e0b\u6587\u7ed3\u5408\u65b9\u5f0f\u7ffb\u8bd1\u6027\u80fd\u5bf9\u6bd4.\u8868 6: \u4e0d\u540c\u6765\u6e90\u4e0a\u4e0b\u6587\u5bf9\u7ffb\u8bd1\u6027\u80fd\u5f71\u54cd.
4.3 \u4e0a \u4e0a \u4e0a\u4e0b \u4e0b \u4e0b\u6587 \u6587 \u6587\u5206 \u5206 \u5206\u5e03 \u5e03 \u5e03
\u9ad8\u6e05\u6570\u5b57\u7535\u89c6\u5728\u7f8e\u6e10\u6210\u4e3b\u6d41
\u8868 10: \u540d\u8bcd\u7ffb\u8bd1\u6837\u4f8b \u886810\u7684\u6837\u4f8b\u53ef\u4ee5\u89c2\u5bdf\u5230\u672c\u6587\u63d0\u51fa\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u6bd4\u5176\u4ed6\u5bf9\u6bd4\u6a21\u578b\u66f4\u597d\u7684\u7ffb\u8bd1\u4e86\u6613\u6df7\u6dc6\u7684\u540d \u8bcd\u3002\u540c\u65f6\u4e0d\u96be\u770b\u51fa\u76f8\u6bd4\u4ec5\u4f7f\u7528\u5355\u8bcd\u7ea7\u522b\u4e0a\u4e0b\u6587\uff0c\u4f7f\u7528\u5c42\u6b21\u5316\u5168\u5c40\u4e0a\u4e0b\u6587\u5bf9\u63d0\u5347\u540d\u8bcd\u7ffb\u8bd1\u8d28\u91cf\u7684 \u6548\u679c\u66f4\u597d\u3002 \u65b0 \u65b0 \u8868 7: \u5f00\u53d1\u96c6\u7bc7\u7ae0\u4e0a\u4e0b\u6587\u5206\u5e03\u6837\u4f8b 5 \u76f8 \u76f8 \u76f8\u5173 \u5173 \u5173\u5de5 \u5de5 \u5de5\u4f5c \u4f5c \u4f5c
\u672c\u6587\u9488\u5bf9\u6587\u4e2d\u4f7f\u7528\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u8fdb\u884c\u524d\u540e\u6587\u5c4f\u853d\u5b9e\u9a8c\u4ee5\u89c2\u5bdf\u5f53\u524d\u53e5\u524d\u6587\u53ca\u540e\u6587\u5404\u81ea\u5bf9\u7ffb
\u8bd1\u7684\u5f71\u54cd\u3002\u5b9e\u9a8c\u7ed3\u679c\u5982\u88686\u6240\u793a,\u4e8c\u8005\u65e0\u663e\u8457\u5dee\u5f02\uff0c\u5c4f\u853d\u524d\u6587\u7684\u7ffb\u8bd1\u6027\u80fd\u7565\u4f4e\u4e8e\u5c4f\u853d\u540e\u6587\uff0c\u8fd9
\u4e0eWong et al., (2020)\u7684\u7814\u7a76\u76f8\u7b26\uff0c\u6211\u4eec\u63a8\u6d4b\u524d\u540e\u6587\u7684\u91cd\u8981\u6027\u53ef\u80fd\u968f\u8bed\u79cd\u548c\u8bed\u6599\u7c7b\u578b\u53d1\u751f\u53d8\u5316\u3002
\u672c\u6587\u8ba4\u4e3a\u8be5\u5b9e\u9a8c\u867d\u7136\u4e0d\u80fd\u76f4\u63a5\u5f97\u51fa\u5f53\u524d\u53e5\u524d\u540e\u6587\u91cd\u8981\u6027\u5b70\u8f7b\u5b70\u91cd\u7684\u7ed3\u8bba\uff0c\u4f46\u53ef\u4ee5\u8868\u660e\u540e\u6587\u4f5c\u4e3a
\u4e0a\u4e0b\u6587\u5bf9\u7bc7\u7ae0\u7ffb\u8bd1\u7684\u91cd\u8981\u6027\u3002\u8868 4: \u4e0d\u540c\u6a21\u578b\u53c2\u6570\u6bd4\u8f83.
" } } } }