| { | |
| "paper_id": "2020", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T12:53:19.443934Z" | |
| }, | |
| "title": "AMR-to-Text Generation with Target Syntax", | |
| "authors": [ | |
| { | |
| "first": "Jie", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou, Jiangsu" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Junhui", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Soochow University", | |
| "location": { | |
| "settlement": "Suzhou, Jiangsu" | |
| } | |
| }, | |
| "email": "[email protected]" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "The task of AMR-to-text generation is to generate text with the same semantic representation given an AMR graph. This task can be viewed as a translation task from the source AMR graph to the target sentence. Some existing methods are currently exploring how to better model the graph structure. However, they all have an unrestricted problem, because many syntactic decisions in the generation phase are not constrained by the semantic graph, thus ignoring the syntactic information hidden within the sentence. In order to clearly consider this shortcoming, this paper proposes a direct and effective method, which shows the integration of syntactic information in the task generated by AMR-to-Text, and has conducted experiments on Transformer and the current model of the optimal performance of the task. The experimental results show that on the two existing standard English data sets LDC2018E86 and LDC2017T10, both have achieved significant improvements and reached new state-of-the-art.", | |
| "pdf_parse": { | |
| "paper_id": "2020", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "The task of AMR-to-text generation is to generate text with the same semantic representation given an AMR graph. This task can be viewed as a translation task from the source AMR graph to the target sentence. Some existing methods are currently exploring how to better model the graph structure. However, they all have an unrestricted problem, because many syntactic decisions in the generation phase are not constrained by the semantic graph, thus ignoring the syntactic information hidden within the sentence. In order to clearly consider this shortcoming, this paper proposes a direct and effective method, which shows the integration of syntactic information in the task generated by AMR-to-Text, and has conducted experiments on Transformer and the current model of the optimal performance of the task. The experimental results show that on the two existing standard English data sets LDC2018E86 and LDC2017T10, both have achieved significant improvements and reached new state-of-the-art.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "\u56fe 1. \"It begins and ends with Romneycare .\" \u62bd\u8c61\u6210AMR\u56fe\u7684\u4e00\u4e2a\u4f8b\u5b50 \u62bd\u8c61\u8bed\u4e49\u8868\u793a(Abstract Meaning Representation\uff0c\u7b80\u79f0AMR) (Banarescu et al., 2013) \u662f \u4e00\u79cd\u65b0\u578b\u7684\u8bed\u4e49\u8868\u793a\u65b9\u6cd5\uff0c\u5b83\u662f\u4ece\u6587\u672c\u4e2d\u62bd\u8c61\u51fa\u6765\u6355\u6349\u6838\u5fc3\u7684\"\u8c01\u5bf9\u8c01\u505a\u4e86\u4ec0\u4e48\"\u7684\u8bed\u4e49\u7ed3 \u6784\uff0c\u5f62\u5f0f\u4e0a\u662f\u4e00\u79cd\u5355\u6839\u6709\u5411\u65e0\u73af\u56fe\u7684\u7ed3\u6784\u3002\u56fe 1\u7ed9\u51fa\u4e86\u4e00\u4e2aAMR\u56fe\u793a\u4f8b\uff0c\u5b83\u662f\u7531\u53e5\u5b50\"It begins and ends with Romneycare .\"\u62bd\u8c61\u800c\u6210\u7684\u3002\u6587\u672c\u4e2d\u7684\u5b9e\u8bcd\u88ab\u62bd\u8c61\u6210AMR\u56fe\u4e2d\u7684\u6982\u5ff5\u8282 \u70b9(concept)\uff0c\u5982\u56fe\u4e2d\"begin-01\"\u548c\"thing\"\u7b49\u8282\u70b9\u79f0\u4f5c\u4e3a\u6982\u5ff5\u3002\u6982\u5ff5\u4e4b\u95f4\u7684\u76f8\u4e92\u5173\u7cfb\u5219\u88ab\u62bd \u8c61\u4e3a\u8fb9(edge)\uff0c\u8868\u793a\u4e24\u4e2a\u6982\u5ff5\u4e4b\u95f4\u5b58\u5728\u7684\u8bed\u4e49\u5173\u7cfb\uff0c\u6bd4\u5982\":ARG0\"\u548c\":op1\"\u7b49\u3002AMR\u56fe\u5728\u8bed \u4e49\u8868\u793a\u4e2d\u5df2\u7ecf\u5f97\u5230\u4e86\u5e7f\u6cdb\u7684\u5e94\u7528\uff0c\u5e76\u4e14\u5728\u673a\u5668\u7ffb\u8bd1 (Tamchyna et al., 2015) \uff0c\u95ee\u7b54\u7cfb\u7edf (Mitra and Baral, 2016) \uff0c\u4e8b\u4ef6\u62bd\u53d6 (Li et al., 2015) \u7b49\u81ea\u7136\u8bed\u8a00\u5904\u7406\u76f8\u5173\u4efb\u52a1\u4e5f\u5f97\u5230\u4e86\u5b9e\u8df5\u3002\u4e0e\u6b64\u540c \u65f6\uff0cAMR-to-Text\u751f\u6210\u5728\u8fd1\u5e74\u6765\u4e5f\u53d7\u5230\u4e86\u8d8a\u6765\u8d8a\u591a\u7684\u5173\u6ce8\u3002 AMR-to-Text\u751f\u6210\u662f\u5728\u7ed9\u5b9aAMR\u56fe\u7684\u6761\u4ef6\u4e0b\uff0c\u81ea\u52a8\u751f\u6210\u76f8\u540c\u8bed\u4e49\u7684\u6587\u672c\u3002\u8be5\u4efb\u52a1\u73b0\u5b58\u7684\u4e00 \u4e9b\u65b9\u6cd5 (Flanigan et al., 2016; Konstas et al., 2017; Song et al., 2016; Song et al., 2018; Beck et al., 2018; Damonte and Cohen, 2019; \u90fd\u7740\u91cd\u5728\u8003\u8651\u5982\u4f55\u5bf9\u56fe\u5173\u7cfb\u8fdb\u884c\u5efa\u6a21\uff0c\u4ece \u800c\u5ffd\u7565\u4e86\u751f\u6210\u65f6\u5b58\u5728\u7684\u53e5\u6cd5\u7ea6\u675f\u3002 \u6700\u521d\u7684\u5de5\u4f5c\u662f\u91c7\u7528\u57fa\u4e8e\u7edf\u8ba1\u7684\u65b9\u6cd5 (Pourdamghani et al., 2016; Song et al., 2017; Flanigan et al., 2016) (Li et al., 2017) \u3002 \u9488\u5bf9\u4e0a\u8ff0\u5b58\u5728\u7684\u95ee\u9898\uff0c\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u663e\u793a\u7684\u65b9\u6cd5\u6765\u878d\u5165\u53e5\u6cd5\u4fe1\u606f\uff0c\u4ece\u800c\u7ed9\u5b9a\u751f\u6210\u65f6\u4e00\u4e9b\u53e5 \u6cd5\u7ea6\u675f,\u5e76\u4e14\u4e0d\u9700\u8981\u5bf9\u6a21\u578b\u672c\u8eab\u8fdb\u884c\u4efb\u4f55\u4fee\u6539\u3002\u4e3a\u4e86\u66f4\u597d\u7684\u8bc1\u660e\u672c\u6587\u65b9\u6cd5\u7684\u6709\u6548\u6027\uff0c\u672c\u6587\u9009\u53d6 \u4e86S2S\u4e2d\u6700\u4f18\u7684Transformer\u6a21\u578b\u548cG2S\u4e2d\u73b0\u5b58\u6700\u4f18\u7684\u6a21\u578b \u8fdb\u884c\u4e86\u5b9e\u9a8c\u3002\u6700\u7ec8\uff0c \u5728\u4e24\u4efd\u6807\u51c6\u7684\u82f1\u6587\u6570\u636e\u96c6LDC2015E86\u548cLDC2017T10\u4e0a\u90fd\u53d6\u5f97\u4e86\u663e\u8457\u7684\u63d0\u5347\u3002 c 2020 \u4e2d\u56fd\u8ba1\u7b97\u8bed\u8a00\u5b66\u5927\u4f1a \u6839\u636e\u300aCreative Commons Attribution 4.0 International License\u300b\u8bb8\u53ef\u51fa\u7248", | |
| "cite_spans": [ | |
| { | |
| "start": 103, | |
| "end": 127, | |
| "text": "(Banarescu et al., 2013)", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 402, | |
| "end": 425, | |
| "text": "(Tamchyna et al., 2015)", | |
| "ref_id": "BIBREF28" | |
| }, | |
| { | |
| "start": 432, | |
| "end": 455, | |
| "text": "(Mitra and Baral, 2016)", | |
| "ref_id": "BIBREF18" | |
| }, | |
| { | |
| "start": 462, | |
| "end": 479, | |
| "text": "(Li et al., 2015)", | |
| "ref_id": "BIBREF15" | |
| }, | |
| { | |
| "start": 584, | |
| "end": 607, | |
| "text": "(Flanigan et al., 2016;", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 608, | |
| "end": 629, | |
| "text": "Konstas et al., 2017;", | |
| "ref_id": "BIBREF13" | |
| }, | |
| { | |
| "start": 630, | |
| "end": 648, | |
| "text": "Song et al., 2016;", | |
| "ref_id": "BIBREF24" | |
| }, | |
| { | |
| "start": 649, | |
| "end": 667, | |
| "text": "Song et al., 2018;", | |
| "ref_id": "BIBREF26" | |
| }, | |
| { | |
| "start": 668, | |
| "end": 686, | |
| "text": "Beck et al., 2018;", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 687, | |
| "end": 711, | |
| "text": "Damonte and Cohen, 2019;", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 763, | |
| "end": 790, | |
| "text": "(Pourdamghani et al., 2016;", | |
| "ref_id": "BIBREF21" | |
| }, | |
| { | |
| "start": 791, | |
| "end": 809, | |
| "text": "Song et al., 2017;", | |
| "ref_id": "BIBREF25" | |
| }, | |
| { | |
| "start": 810, | |
| "end": 832, | |
| "text": "Flanigan et al., 2016)", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 833, | |
| "end": 850, | |
| "text": "(Li et al., 2017)", | |
| "ref_id": "BIBREF16" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\uff0c\u968f\u540eKonstas(2017)\u5c06\u8be5\u4efb\u52a1\u5f15\u5165\u5230\u4e86\u5e8f\u5217\u5230\u5e8f\u5217(sequence-to-sequence\uff0c \u7b80 \u79f0S2S) \u6a21 \u578b \u4e0a \uff0c \u4f7f \u7528 \u53cc \u5411 \u957f \u77ed \u65f6 \u8bb0 \u5fc6 \u7f51 \u7edc (Bi-LSTM) \u8fdb \u884c \u7f16 \u7801 \u3002 \u4f46 \u662fS2S\u6a21 \u578b \u9700 \u8981 \u5c06AMR\u56fe\u8fdb\u884c\u5e8f\u5217\u5316\u53bb\u9002\u5e94\u6a21\u578b\u7684\u8f93\u5165\uff0c\u8fd9\u6837\u4f1a\u635f\u5931\u5927\u91cf\u7684\u56fe\u7ed3\u6784\u4fe1\u606f\u3002\u56e0\u6b64\uff0c\u4e3a\u4e86\u66f4\u597d\u7684\u5bf9 \u56fe \u5173 \u7cfb \u8fdb \u884c \u5efa \u6a21 \uff0cBeck\u7b49(2018),Song\u7b49(2018),Damonte\u7b49(2019),Guo\u7b49(2019),Zhu\u7b49(2019)\u63d0 \u51fa\u4e86\u56fe\u5230\u5e8f\u5217(graph-to-seq\uff0c\u7b80\u79f0G2S)\u7684\u6846\u67b6\uff0c\u4f7f\u7528\u56fe\u6a21\u578b\u6765\u5bf9AMR\u56fe\u8fdb\u884c\u5efa\u6a21\u3002\u7136\u800c\uff0c \u4ed6\u4eec\u7684\u5de5\u4f5c\u90fd\u5c06\u53e5\u5b50\u8868\u793a\u4e3a\u5355\u8bcd\u5e8f\u5217\uff0c\u5e76\u6ca1\u6709\u8003\u8651\u5230\u53e5\u5b50\u4e2d\u6f5c\u5728\u7684\u53e5\u6cd5\u4fe1\u606f\u3002\u6700\u8fd1\u7684\u4e00\u4e9b\u7814\u7a76 \u4e5f\u8868\u660e\uff0c\u5373\u4f7f\u767e\u4e07\u7ea7\u7684\u5e73\u884c\u8bed\u6599\uff0c\u6a21\u578b\u4ecd\u7136\u65e0\u6cd5\u4ece\u4e2d\u6355\u83b7\u6df1\u5c42\u7684\u53e5\u6cd5\u4fe1\u606f", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "It begins and ends with Romneycare .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cb\u53e5\u5b50\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(a / and :op1 (b / begin-01 :ARG1 (i / it) :ARG2 (t / thing :wiki \"Massachusetts_health_care_reform\" :name (n / name :op1 \"Romneycare\"))) :op2(e / end-01 :ARG1 i :ARG2 t))", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u7ebf\u6027\u5316\u4e4b\u540e\u7684AMR\u56fe\uff1a and :op1 ( begin :arg1 it :arg2 ( thing :name ( name :op1 romneycare ) ) ) :op2 ( end :arg1 it :arg2 thing )", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u56fe 2. \u4e00\u4e2aAMR\u56fe\u7ebf\u6027\u5316\u793a\u4f8b 2 \u76f8 \u76f8 \u76f8\u5173 \u5173 \u5173\u5de5 \u5de5 \u5de5\u4f5c \u4f5c \u4f5c \u76ee\u524dAMR-to-Text\u751f\u6210\u7684\u4efb\u52a1\u5927\u81f4\u53ef\u4ee5\u5206\u4e3a\u4e24\u7c7b\uff1a\u57fa\u4e8e\u7edf\u8ba1\u7684\u65b9\u6cd5\u548c\u57fa\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u65b9 \u6cd5\u3002\u800c\u57fa\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u65b9\u6cd5\uff0c\u73b0\u5728\u53c8\u53ef\u4ee5\u5206\u4e3aseq2seq\u548cgraph2seq\u4e24\u7c7b\u3002 2.1 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u7edf \u7edf \u7edf\u8ba1 \u8ba1 \u8ba1\u7684 \u7684 \u7684\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 \u65e9 \u671f \u795e \u7ecf \u7f51 \u7edc \u672a \u666e \u53ca \u65f6 \u5019 \uff0c \u5728AMR-to-Text\u751f \u6210 \u4e0a \u7684 \u5de5 \u4f5c \u5927 \u90fd \u4f7f \u7528 \u57fa \u4e8e \u7edf \u8ba1 \u7684 \u65b9 \u6cd5\u3002Flanigan\u7b49(2016)\u5c06AMR\u56fe\u8f6c\u6362\u4e3a\u5408\u9002\u7684\u751f\u6210\u6811\uff0c\u5e76\u5e94\u7528\u6811-\u4e32(tree-to-string)\u8f6c\u6362\u5668 \u751f\u6210\u6587\u672c\u3002Song\u7b49(2016)\u5c06\u4e00\u4e2aAMR\u56fe\u62c6\u5206\u6210\u4e86\u8bb8\u591a\u5c0f\u7684\u7247\u6bb5\uff0c\u5e76\u751f\u6210\u6240\u6709\u7247\u6bb5\u7684\u7ffb\u8bd1\uff0c\u6700\u7ec8 \u901a\u8fc7\u91c7\u7528\u975e\u5bf9\u79f0\u5e7f\u4e49\u65c5\u884c\u5546\u95ee\u9898\u89e3\u6cd5\u6765\u5bf9\u7247\u6bb5\u786e\u5b9a\u5176\u987a\u5e8f\u3002Song\u7b49(2017)\u4f7f\u7528\u540c\u6b65\u8282\u70b9\u66ff\u6362\u8bed \u6cd5\u6765\u5bf9AMR\u56fe\u8fdb\u884c\u89e3\u6790\uff0c\u5e76\u751f\u6210\u76f8\u5e94\u7684\u53e5\u5b50\u3002Pourdamghani\u7b49(2016)\u91c7\u7528\u57fa\u4e8e\u77ed\u8bed\u7684\u673a\u5668\u7ffb\u8bd1 \u6a21\u578b\u6765\u5bf9\u7ebf\u6027\u5316AMR\u56fe\u8fdb\u884c\u5efa\u6a21\u3002 2.2 \u57fa \u57fa \u57fa\u4e8e \u4e8e \u4e8e\u795e \u795e \u795e\u7ecf \u7ecf \u7ecf\u7f51 \u7f51 \u7f51\u7edc \u7edc \u7edc\u7684 \u7684 \u7684\u65b9 \u65b9 \u65b9\u6cd5 \u6cd5 \u6cd5 \u968f\u7740\u795e\u7ecf\u7f51\u7edc\u7684\u5174\u8d77\uff0c\u6700\u8fd1\u7684\u7814\u7a76\u90fd\u662f\u4f7f\u7528\u795e\u7ecf\u7f51\u7edc\u6765\u751f\u6210\u3002\u5728Sutskever\u7b49(2014)\u8bc1\u660e \u4e86\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u4f18\u8d8a\u6027\u4e4b\u540e\uff0cKonstas\u7b49(2017)\u63d0\u51fa\u4f7f\u7528\u5e8f\u5217\u5230\u5e8f\u5217(S2S)\u6a21\u578b\u6765\u751f\u6210\u6587 \u672c\uff0c\u5229\u7528\u53cc\u5411LSTM\u6765\u5bf9\u7ebf\u6027\u5316\u7684AMR\u56fe\u8fdb\u884c\u7f16\u7801\u3002\u4e3a\u4e86\u9650\u5236\u751f\u6210\u7684\u6587\u672c\u5177\u6709\u66f4\u5408\u7406\u7684\u53e5 \u6cd5\uff0cCao\u7b49(2019)\u5c06AMR-to-Text\u751f\u6210\u7684\u4efb\u52a1\u62c6\u5206\u6210\u4e24\u4e2a\u6b65\u9aa4\uff0c\u5148\u4f7f\u7528\u53e5\u6cd5\u6a21\u578b\u53bb\u9884\u6d4b\u6700\u4f18\u7684\u76ee \u6807\u7aef\u53e5\u6cd5\u7ed3\u6784\uff0c\u518d\u5229\u7528\u9884\u6d4b\u7684\u53e5\u6cd5\u4fe1\u606f\u53bb\u8f85\u52a9\u751f\u6210\u6a21\u578b\u66f4\u597d\u7684\u751f\u6210\u53e5\u5b50\u3002\u4f46\u662f\u4e5f\u76f8\u5e94\u7684\u635f\u5931\u4e86 \u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7aef\u5230\u7aef\u7684\u7279\u6027\uff0c\u5e76\u4e14\u548c\u672c\u6587\u65b9\u6cd5\u76f8\u6bd4\u66f4\u52a0\u590d\u6742\uff0c\u589e\u52a0\u4e86\u7f51\u7edc\u7684\u590d\u6742\u5ea6\u548c\u53c2\u6570\u3002 \u968f\u540e\uff0c\u4e3a\u4e86\u89e3\u51b3seq2seq\u6a21\u578b\u5c06AMR\u56fe\u7ebf\u6027\u5316\u4e4b\u540e\u4fe1\u606f\u635f\u5931\u7684\u95ee\u9898\uff0c\u5927\u5bb6\u7684\u7814\u7a76\u70ed\u70b9\u90fd \u7740\u91cd\u5728\u7814\u7a76\u56fe\u795e\u7ecf\u7f51\u7edc\u4e0a\u3002\u56fe\u5230\u5e8f\u5217(Graph-to-Sequence)\u6a21\u578b\u5e38\u4f18\u4e8e\u5e8f\u5217\u5230\u5e8f\u5217(S2S) \u6a21\u578b\uff0c\u5305\u62ec\u56fe\u72b6\u6001LSTM(2018),GGNN(2018)\u7b49\u3002\u56fe\u72b6\u6001LSTM\u901a\u8fc7\u6bcf\u6b65\u7684\u8fed\u4ee3\u4ea4\u6362\u76f8\u90bb\u8282\u70b9 \u7684\u4fe1\u606f\u6765\u66f4\u65b0\u8282\u70b9\u3002\u540c\u65f6\u4e5f\u5bf9\u6bcf\u4e2a\u8282\u70b9\u589e\u52a0\u4e00\u4e2a\u5411\u91cf\u5355\u5143\u4fdd\u5b58\u5386\u53f2\u4fe1\u606f\u3002GGNN\u662f\u4e00\u4e2a\u57fa \u4e8e\u95e8\u63a7\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\uff0c\u5c06AMR\u56fe\u7ed3\u6784\u5b8c\u6574\u7684\u878d\u5165\u6a21\u578b\u4e2d\uff0c\u5e76\u4e14\u5c06\u8fb9\u4fe1\u606f\u4e5f\u8f6c\u5316\u4e3a\u8282\u70b9\uff0c\u89e3 \u51b3\u4e86\u53c2\u6570\u7206\u70b8\u95ee\u9898\u7684\u540c\u65f6\uff0c\u4e5f\u7ed9\u4e86\u89e3\u7801\u5668\u66f4\u4e30\u5bcc\u7684\u4fe1\u606f\u3002\u7740\u91cd\u4e8e\u89e3\u51b3AMR\u56fe\u4e2d\u91cd\u5165\u8282\u70b9 \u7684\u95ee\u9898\uff0cDamonte\u7b49(2019)\u63d0\u51fa\u4e86\u4e00\u79cd\u5806\u6808\u5f0f\u7684\u7f16\u7801\u5668\uff0c\u7531\u56fe\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u548c\u53cc\u5411LSTM\u5806 \u6808\u800c\u6210\u3002Guo\u7b49(2019)\u63d0\u51fa\u4e86\u4e00\u79cd\u6df1\u5ea6\u8fde\u63a5\u56fe\u5377\u79ef\u7f51\u7edc(GCN)\u66f4\u597d\u7684\u83b7\u53d6\u5c40\u90e8\u4e0e\u975e\u5c40\u90e8\u4fe1 \u606f\u3002Zhu\u7b49(2019) \u5728Transformer\u7684\u57fa\u7840\u4e0a\uff0c\u53d7\u5230(Shaw et al., 2018)\u5bf9\u76f8\u5bf9\u4f4d\u7f6e\u5efa\u6a21\u7684\u542f\u53d1\uff0c\u63d0 \u51fa\u4e86\u4e00\u79cdStructure-Aware Self-Attention\u7684\u7f16\u7801\u65b9\u6cd5\u53ef\u4ee5\u5bf9\u56fe\u7ed3\u6784\u4e2d\u4efb\u610f\u4e24\u4e24\u8282\u70b9\u8fdb\u884c\u5b8c\u6574\u7684 \u5efa\u6a21(\u4e0d\u8bba\u8282\u70b9\u4e4b\u95f4\u662f\u5426\u76f4\u63a5\u76f8\u8fde)\uff0c\u5728\u8be5\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u6700\u9ad8\u7684\u6027\u80fd\u3002 \u672c\u6587\u91c7\u7528\u4e86\u4e24\u79cd\u65b9\u6cd5\u4f5c\u4e3a\u57fa\u51c6\u6a21\u578b(Baseline)\u3002 1. Transformer\uff0c\u6700\u5148\u8fdb\u7684seq2seq\u6a21\u578b\uff0c\u6700\u521d\u4f7f\u7528\u4e8e\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u548c\u53e5\u6cd5\u5206\u6790\u4efb\u52a1(Vaswani et al., 2017)\u3002 2. Zhu\u7b49(2019)\u63d0\u51fa\u7684Structure-Aware Self-Attention\u6a21\u578b\uff0c\u76ee\u524d\u5728AMR-to-Text\u751f\u6210\u7684\u4efb\u52a1\u4e0a \u53d6\u5f97\u4e86\u6700\u9ad8\u7684\u6027\u80fd\u3002 3.1 \u57fa \u57fa \u57fa\u51c6 \u51c6 \u51c6\u6a21 \u6a21 \u6a21\u578b \u578b \u578b1( ( (Baseline1) ) ) 3.1.1 Trasnformer Baseline1\u662f\u4f7f\u7528\u7684Transformer\u6a21\u578b\uff0c\u5b83\u91c7\u7528\u4e86\u7f16\u7801\u5668-\u89e3\u7801\u5668(Encoder-Decoder)\u7684\u67b6 \u6784\uff0c\u7531\u8bb8\u591a\u7f16\u7801\u5668\u548c\u89e3\u7801\u5668\u5806\u6808\u7ec4\u6210\u3002\u6bcf\u4e00\u4e2a\u7f16\u7801\u5668\u90fd\u5b58\u5728\u4e24\u4e2a\u5b50\u5c42\uff1a\u81ea\u6ce8\u610f\u529b\u673a\u5236\u5c42(self- attention)\u540e\u9762\u7d27\u63a5\u7740\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u5c42(position-wise feed forward)\u3002\u81ea\u6ce8\u610f\u529b\u5c42\u4f7f\u7528\u4e86\u591a \u4e2a\u6ce8\u610f\u529b\u5934(attention head)\uff0c\u5c06\u6bcf\u4e2a\u6ce8\u610f\u529b\u5934\u7684\u7ed3\u679c\u8fdb\u884c\u8fde\u63a5\u548c\u8f6c\u6362\u4e4b\u540e\uff0c\u5f62\u6210\u81ea\u6ce8\u610f \u673a\u5236\u5c42\u7684\u8f93\u51fa\u3002\u6bcf\u4e2a\u6ce8\u610f\u529b\u5934\u4f7f\u7528\u70b9\u4e58\u6ce8\u610f\u529b\u673a\u5236(scaled dot-product)\u6765\u8ba1\u7b97\u8f93\u5165\u4e00\u4e2a\u5e8f \u5217x = (x 1 , \u2022 \u2022 \u2022 , x n )\uff0c\u5f97\u5230\u4e00\u4e2a\u540c\u6837\u957f\u5ea6\u7684\u65b0\u7684\u5e8f\u5217z = (z 1 , \u2022 \u2022 \u2022 , z n ): z = Attention (x) (1) \u5176\u4e2dx i \u2208 R dx \uff0cz \u2208 R n\u00d7dz \u3002\u6bcf\u4e00\u4e2a\u8f93\u51fa\u5143\u7d20z i \u662f\u8f93\u5165\u5143\u7d20\u7684\u7ebf\u6027\u53d8\u6362\u7684\u52a0\u6743\u548c: z i = n j=1 \u03b1 ij x j W V", | |
| "eq_num": "(2)" | |
| } | |
| ], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u5176\u4e2dW V \u2208 R dx\u00d7dz \u662f\u4e00\u4e2a\u53ef\u5b66\u4e60\u7684\u53c2\u6570\u77e9\u9635. \u516c\u5f0f2\u4e2d\u7684\u5411\u91cf\u03b1 i = (\u03b1 i1 , \u2022 \u2022 \u2022 , \u03b1 in ) \u662f\u901a\u8fc7\u81ea\u6ce8\u610f \u529b\u673a\u5236\u6a21\u578b\u5f97\u5230\u7684\uff0c\u8be5\u673a\u5236\u6355\u83b7\u4e86x i \u548c\u5176\u5b83\u5143\u7d20\u4e4b\u95f4\u7684\u5bf9\u5e94\u5173\u7cfb\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u6bcf\u4e2a\u5143\u7d20x j \u7684\u81ea \u6ce8\u610f\u529b\u6743\u91cd\u03b1 ij \u662f\u901a\u8fc7\u4e00\u4e2asoftmax\u51fd\u6570\u8ba1\u7b97\u5f97\u5230\uff1a \u03b1 ij = exp(e ij ) n k=1 exp(e ik )", | |
| "eq_num": "(3)" | |
| } | |
| ], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u5176\u4e2d", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "e ij = x i W Q x j W K T \u221a d z (4) \u662f\u4e00\u4e2a\u5bf9\u9f50\u51fd\u6570\uff0c\u5b83\u7528\u6765\u5ea6\u91cf\u8f93\u5165\u5143\u7d20x i \u548cx j \u7684\u5339\u914d\u7a0b\u5ea6. W Q , W K \u2208 R dx\u00d7dz \u662f\u53ef\u5b66\u4e60\u7684\u53c2 \u6570\u77e9\u9635\u3002 3.1.2 \u7ebf \u7ebf \u7ebf\u6027 \u6027 \u6027\u5316 \u5316 \u5316\u9884 \u9884 \u9884\u5904 \u5904 \u5904\u7406 \u7406 \u7406 \u56e0\u4e3aTransformer\u662fseq2seq\u6a21\u578b\uff0c\u8f93\u5165\u53ea\u652f\u6301\u5e8f\u5217\u5316\u7684\u8f93\u5165\uff0c\u6240\u4ee5\u9700\u8981\u5bf9AMR\u56fe\u8fdb\u884c\u7ebf\u6027 \u5316\u7684\u9884\u5904\u7406\u3002\u672c\u6587\u91c7\u7528Konstas\u7b49(2017)\u63d0\u51fa\u7684\u6df1\u5ea6\u4f18\u5148\u904d\u5386\u7684\u7ebf\u6027\u5316\u65b9\u6cd5\u6765\u5bf9AMR\u56fe\u8fdb\u884c\u9884 \u5904\u7406\uff0c\u4ece\u800c\u5f97\u5230\u7b80\u5316\u7248\u7684AMR\u56fe\u3002\u5728\u7ebf\u6027\u5316\u4e4b\u524d\uff0c\u9996\u5148\u79fb\u9664\u4e86\u56fe\u4e2d\u7684\u53d8\u91cf\u3001wiki\u94fe\u63a5\u548c\u8bed\u4e49\u6807 \u7b7e\u3002\u56fe2\u5c55\u793a\u4e86\u4e00\u4e2aAMR\u56fe\u7ebf\u6027\u5316\u793a\u4f8b\u3002 3.2 \u57fa \u57fa \u57fa\u51c6 \u51c6 \u51c6\u6a21 \u6a21 \u6a21\u578b \u578b \u578b2( ( (Baseline2) ) ) 3.2.1 Structure-Aware Self-Attention Zhu\u7b49(2019)\u6269\u5c55\u4e86\u4f20\u7edf\u7684\u81ea\u6ce8\u610f\u529b\u673a\u5236\u6846\u67b6,\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u9896\u7684\u7ed3\u6784\u5316\u7684\u6ce8\u610f\u529b\u673a\u5236\uff0c\u5728 \u5bf9\u9f50\u51fd\u6570\u4e2d\u663e\u5f0f\u5730\u5bf9\u5143\u7d20\u5bf9(x i , x j )\u4e4b\u95f4\u7684\u5173\u7cfb\u8fdb\u884c\u7f16\u7801\uff0c\u7528\u516c\u5f0f5\u66ff\u6362\u516c\u5f0f4\u3002 e ij = x i W Q x j W K + r ij W R T \u221a d z", | |
| "eq_num": "(5)" | |
| } | |
| ], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "x i x j \u7ed3 \u7ed3 \u7ed3\u6784 \u6784 \u6784\u6807 \u6807 \u6807\u7b7e \u7b7e \u7b7e\u5e8f \u5e8f \u5e8f\u5217 \u5217 \u5217 begin-01 and :ARG1\u2191 begin-01 Romneycare :ARG2\u2193 :name\u2193 begin-01 begin-01 None ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "\u8868 1. \u56fe1\u4e2d\u4e00\u4e9b\u6982\u5ff5\u5bf9\u4e4b\u95f4\u7684\u7ed3\u6784\u8def\u5f84\u793a\u4f8b\u3002 \u5176\u4e2dW R \u2208 R dz\u00d7dz \u662f\u4e00\u4e2a\u53c2\u6570\u77e9\u9635. \u7136\u540e, \u518d\u76f8\u5e94\u5730\u66f4\u65b0\u516c\u5f0f2\uff0c\u5c06\u7ed3\u6784\u4fe1\u606f\u4f20\u64ad\u5230\u5b50\u5c42\u7684\u8f93 \u51fa\u3002 z i = n j=1 \u03b1 ij x j W V + r ij W F (6) \u5176\u4e2d\uff0cW F \u2208 R dz\u00d7dz \u662f\u4e00\u4e2a\u53c2\u6570\u77e9\u9635\u3002r ij \u2208 R dz \u4ee3\u8868\u4e86\u5143\u7d20\u5bf9(x i , x j )\u4e4b\u95f4\u7684\u5173\u7cfb\uff0c\u5b83\u662f\u901a \u8fc73.2.2\u5b66\u4e60\u5230\u7684\u4e00\u4e2a\u5411\u91cf\u8868\u793a\u3002 3.2.2 \u5b66 \u5b66 \u5b66\u4e60 \u4e60 \u4e60\u56fe \u56fe \u56fe\u6982 \u6982 \u6982\u5ff5 \u5ff5 \u5ff5( ( (concept) ) )\u5bf9 \u5bf9 \u5bf9\u4e4b \u4e4b \u4e4b\u95f4 \u95f4 \u95f4\u7684 \u7684 \u7684\u5411 \u5411 \u5411\u91cf \u91cf \u91cf\u8868 \u8868 \u8868\u793a \u793a \u793ar \u4e0a\u8ff0\u7684Structure-Aware Self-Attention\u673a\u5236\u53ef\u4ee5\u7528\u6765\u83b7\u53d6\u5230\u56fe\u4e2d\u4efb\u610f\u4e24\u4e24\u6982\u5ff5\u5bf9(concept pairs)\u4e4b\u95f4\u7684\u56fe\u7ed3\u6784\u5173\u7cfb\u3002\u5b9a\u4e49\u4f7f\u7528\u6cbf\u7740\u6982\u5ff5x i \u5230x j \u4e4b\u95f4\u8fb9\u6807\u7b7e(edge label)\u7ec4\u6210\u7684\u4e00\u6761\u8def\u5f84 \u5f53\u4f5c\u6982\u5ff5\u5bf9\u4e4b\u95f4\u7684\u56fe\u7ed3\u6784\u5173\u7cfb 0 \u3002\u540c\u65f6\uff0c\u4e3a\u4e86\u533a\u5206\u65b9\u5411\uff0c\u4e5f\u7ed9\u6bcf\u6761\u8fb9\u6807\u7b7e\u76f8\u5e94\u7684\u589e\u52a0\u4e86\u65b9\u5411\u7b26 \u53f7\u3002Table 1\u5c55\u793a\u4e86\u56fe 1\u4e2d\u7684\u51e0\u4e2a\u6982\u5ff5\u5bf9\u4e4b\u95f4\u7684\u7ed3\u6784\u6807\u7b7e\u5e8f\u5217\u3002 \u73b0\u5728\u5df2\u7ecf\u7ed9\u5b9a\u4e86\u4e00\u4e2a\u7ed3\u6784\u6807\u7b7e\u8def\u5f84s = s 1 , \u2022 \u2022 \u2022 , s k \uff0c\u7136\u540e\u83b7\u53d6\u5230\u5b83\u7684\u5411\u91cf\u8868\u793al = l 1 , \u2022 \u2022 \u2022 , l k , \u6700\u540e\u672c\u6587\u4f7f\u7528\u57fa\u4e8e\u5377\u79ef\u795e\u7ecf\u7f51\u7edc(Kalchbrenner et al., 2014)(CNN-based) 1 \u7684\u65b9\u6cd5\u6765\u83b7\u5f97\u516c \u5f0f 5\u548c\u516c\u5f0f 6\u4e2d\u7684\u5411\u91cf\u8868\u793a r ij \u3002 CNN-based \u4f7f\u7528CNN\u6765\u5377\u79ef\u6807\u7b7e\u5e8f\u5217l\u83b7\u5f97\u4e00\u4e2a\u5411\u91cfr\uff1a conv = Conv1D(kernel size = (m), strides = 1, f ilters = d z , input shape = d z activation = relu ) (7) r = conv (l)", | |
| "eq_num": "(8)" | |
| } | |
| ], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u8868 2. \u672c\u6587\u65b9\u6cd5\u5728LDC2015E86\u548cLDC2017T10\u6d4b\u8bd5\u96c6\u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u53ca\u4e0e\u5176\u5b83\u6a21\u578b\u7684\u6bd4\u8f83\u3002 * \u4ee3 \u8868seq2seq\u6a21\u578b\uff0c \u2020 \u4ee3\u8868graph2seq\u6a21\u578b\uff0c \u2021 \u4ee3\u8868\u5176\u5b83\u6a21\u578b\u3002 \u957f\u5ea6\u4e3a 4\u3002\u89e3\u7801\u65f6\uff0c\u9ed8\u8ba4\u7684\u989d\u5916\u957f\u5ea6\u4ece50\u589e\u52a0\u81f3150\uff0c\u8be5\u503c\u8868\u793a\u6a21\u578b\u89e3\u7801\u65f6\u5141\u8bb8\u751f\u6210\u53e5\u5b50\u957f\u5ea6\u662f", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\u6e90\u7aef\u6700\u5927\u957f\u5ea6\u52a0150\u3002\u5728\u6240\u6709\u5b9e\u9a8c\u4e2d\uff0c\u4ee5\u5b66\u4e60\u7387 0.5 \u5728 Tesla P40 GPU\u4e0a\u8bad\u7ec3 300K \u6b65\u505c\u6b62\u3002\u672c \u6587\u5b9e\u9a8c\u4ee3\u7801\u5df2\u5f00\u6e90\u516c\u5e03\u81f3https://github.com/Amazing-J/structural-transformer\u3002 \u4e3a\u4e86\u66f4\u597d\u7684\u4f53\u73b0\u672c\u6587\u65b9\u6cd5\u7684\u6709\u6548\u6027\uff0c\u91c7\u7528\u4e86BLEU (Papineni et al., 2002) \uff0cMeteor (Banerjee and Lavie, 2005; Denkowski and Lavie, 2014) \uff0cchrF++ (Popovi\u0107, 2017) \u4e09\u79cd\u8bc4\u6d4b\u6307\u6807\u3002BLEU\u662f \u57fa\u4e8e\u8bed\u6599\u7ea7\u7684\u8bc4\u4f30\uff0c\u540e\u4e24\u8005\u662f\u57fa\u4e8e\u53e5\u5b50\u7ea7\u7684\u8bc4\u4f30\u3002\u76f8\u5bf9\u6765\u8bf4\uff0c\u540e\u4e24\u8005\u7684\u5206\u6570\u66f4\u63a5\u8fd1\u4e8e\u4eba\u5de5\u8bc4 \u6d4b\u3002 4.3 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u7ed3 \u7ed3 \u7ed3\u679c \u679c \u679c \u88682\u7ed9\u51fa\u4e86\u672c\u6587\u4e24\u4e2a\u57fa\u51c6\u6a21\u578b\u5728\u878d\u5408\u4e86\u76ee\u6807\u7aef\u53e5\u6cd5\u4fe1\u606f\u524d\u540eAMR-to-Text\u751f\u6210\u7684\u6027\u80fd\u5bf9\u6bd4\u3002 \u4ece\u88682\u53ef\u4ee5\u770b\u51fa\uff0c\u878d\u5408\u76ee\u6807\u7aef\u53e5\u6cd5\u4fe1\u606f\u4e4b\u540e\uff0cAMR-to-Text\u751f\u6210\u7684\u6027\u80fd\u6709\u7740\u663e\u8457\u7684\u63d0\u5347\u3002\u5728\u4e24\u4e2a \u57fa\u51c6\u6a21\u578b\u4e0a\uff0c\u5206\u522b\u63d0\u9ad8\u4e861.26\u548c1.07(LDC2015E86)\uff0c1.15\u548c1.18(LDC2017T10)BLEU\u3002\u8fd9 \u4e5f\u6709\u529b\u7684\u8bc1\u660e\uff0c\u5728\u76ee\u6807\u7aef\u878d\u5165\u53e5\u6cd5\u4fe1\u606f\uff0c\u53ef\u4ee5\u5e2e\u52a9\u6a21\u578b\u5b66\u4e60\u5230\u53e5\u5b50\u4e2d\u6f5c\u85cf\u7684\u4e00\u4e9b\u77e5\u8bc6\uff0c\u4ece\u800c\u5728 \u751f\u6210\u65f6\u8003\u8651\u5230\u53e5\u6cd5\u4fe1\u606f\u7684\u7ea6\u675f\u3002\u8be5\u65b9\u6cd5\u4e0e\u878d\u5408\u6e90\u7aef\u53e5\u6cd5\u548c\u8bed\u4e49\u89d2\u8272\u4fe1\u606f\u7684\u673a\u5668\u7ffb\u8bd1\u65b9\u6cd5\u7c7b\u4f3c (Li et al., 2013) ", | |
| "cite_spans": [ | |
| { | |
| "start": 145, | |
| "end": 168, | |
| "text": "(Papineni et al., 2002)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 177, | |
| "end": 203, | |
| "text": "(Banerjee and Lavie, 2005;", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 204, | |
| "end": 230, | |
| "text": "Denkowski and Lavie, 2014)", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 239, | |
| "end": 254, | |
| "text": "(Popovi\u0107, 2017)", | |
| "ref_id": "BIBREF20" | |
| }, | |
| { | |
| "start": 575, | |
| "end": 592, | |
| "text": "(Li et al., 2013)", | |
| "ref_id": "BIBREF14" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "\uff0c\u672c\u6587\u8fdb\u4e00\u6b65\u8bc1\u660e\u4e86\u5728\u751f\u6210\u4efb\u52a1\u4e2d\u76ee\u6807\u7aef\u878d\u5165\u53e5\u6cd5\u4fe1\u606f\u540c\u6837\u53ef\u4ee5\u6709\u7740\u663e\u8457\u7684\u63d0\u5347\u3002 \u8868 2\u4e5f \u7ed9 \u51fa \u4e86 \u4e0e \u5176 \u5b83 \u73b0 \u5b58 \u6a21 \u578b \u5728 \u8be5 \u4efb \u52a1 \u4e0a \u7684 \u6027 \u80fd \u6bd4 \u8f83 \u3002 \u503c \u5f97 \u6ce8 \u610f \u7684 \u662f\uff0cLDC2015E86\u548cLDC2017T10\u7684\u9a8c\u8bc1\u96c6\u548c\u6d4b\u8bd5\u96c6\u662f\u76f8\u540c\u7684\uff0c\u533a\u522b\u53ea\u662f\u8bad\u7ec3\u96c6\u7684\u6570\u91cf\u76f8\u5dee \u4e86\u4e00\u500d\u5de6\u53f3\u3002\u4ece\u88682\u53ef\u4ee5\u770b\u5230\uff0c\u4e0eseq2seq\u6a21\u578b\u76f8\u6bd4\uff0c\u672c\u6587\u7684baseline1\u5c31\u5df2\u7ecf\u663e\u8457\u7684\u8d85\u8d8a\u4e86\u5b83 \u4eec\uff0c\u5e76\u4e14\u5728\u878d\u5165\u53e5\u6cd5\u4fe1\u606f(+Syntax)\u4e4b\u540e\uff0c\u6027\u80fd\u4f9d\u7136\u6709\u7740\u660e\u663e\u7684\u63d0\u5347\u3002\u76ee\u524d\u6700\u9ad8\u7684\u6027\u80fd \u662fZhu\u7b49(2019)\u63d0\u51fa\u7684Structure-Aware Self-Attention\u6a21\u578b\uff0c\u672c\u6587\u5728\u5b83\u4eec\u7684\u57fa\u7840\u4e4b\u4e0a\u4e5f\u540c\u6837\u6709\u7740 \u6709\u6548\u7684\u63d0\u9ad8\uff0c\u521b\u9020\u4e86\u65b0\u7684\u6700\u9ad8\u7684\u6027\u80fd(SOTA)\u3002\u53ef\u4ee5\u8bc1\u660e\u672c\u6587\u7684\u65b9\u6cd5\u65e0\u8bba\u7528\u5728seq2seq\u6a21\u578b\u6216 \u8005graph2seq\u6a21\u578b\u4e0a\u90fd\u6709\u6548\u3002 4.4 \u53c2 \u53c2 \u53c2\u6570 \u6570 \u6570\u6570 \u6570 \u6570\u91cf \u91cf \u91cf\u548c \u548c \u548c\u8bad \u8bad \u8bad\u7ec3 \u7ec3 \u7ec3\u65f6 \u65f6 \u65f6\u95f4 \u95f4 \u95f4 \u672c\u6587\u878d\u5408\u76ee\u6807\u7aef\u53e5\u6cd5\u4fe1\u606f\u7684\u65b9\u6cd5\u662f\u5c06\u76ee\u6807\u7aef\u53e5\u5b50\u66ff\u6362\u4e3a\u7ebf\u6027\u5316\u7ed3\u6784\u53e5\u6cd5\u6811\uff0c\u4e0d\u4f1a\u5bf9\u6a21\u578b\u8fdb \u884c\u4efb\u4f55\u4fee\u6539\uff0c\u4e5f\u5c31\u610f\u5473\u7740\u5e76\u4e0d\u4f1a\u7ed9\u6a21\u578b\u589e\u52a0\u53c2\u6570\uff0c\u8fd9\u4e5f\u662f\u672c\u6587\u65b9\u6cd5\u7684\u4e00\u5927\u4f18\u70b9\u3002\u4f46\u662f\uff0c\u76ee\u6807\u7aef \u53e5\u5b50\u66ff\u6362\u6210\u7ebf\u6027\u5316\u7ed3\u6784\u53e5\u6cd5\u6811\u4e4b\u540e\uff0c\u5b83\u7684\u5e8f\u5217\u957f\u5ea6\u4f1a\u76f8\u5e94\u7684\u53d8\u957f\uff0c\u8fd9\u5c31\u4f1a\u5bfc\u81f4\u8bad\u7ec3\u7684\u65f6\u95f4\u7565\u5fae \u589e\u52a0\u3002\u636e\u7edf\u8ba1\uff0c\u672c\u6587baseline1\u57fa\u51c6\u6a21\u578b\u5728LDC2015E86\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u5b8c\u573a\u4e00\u8f6e\u8bad\u7ec3\u7684\u65f6\u95f4\u5927\u6982 \u9700\u8981288\u79d2(\u7ea64.80\u5206\u949f)\uff0c\u800c\u878d\u5408\u76ee\u6807\u7aef\u53e5\u6cd5\u4fe1\u606f\u4e4b\u540e\uff0c\u5927\u6982\u82b1\u8d39345\u79d2(\u7ea65.75\u5206\u949f)\u3002", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "4.5 \u878d \u878d \u878d\u5408 \u5408 \u5408\u4e0d \u4e0d \u4e0d\u540c \u540c \u540c\u5f62 \u5f62 \u5f62\u5f0f \u5f0f \u5f0f\u53e5 \u53e5 \u53e5\u6cd5 \u6cd5 \u6cd5\u4fe1 \u4fe1 \u4fe1\u606f \u606f \u606f\u7684 \u7684 \u7684\u5f71 \u5f71 \u5f71\u54cd \u54cd \u54cd \u4ece\u5b9e\u9a8c\u7ed3\u679c\u53ef\u4ee5\u5f97\u5230\u878d\u5408\u76ee\u6807\u7aef\u53e5\u5b50\u7684\u53e5\u6cd5\u4fe1\u606f\u53ef\u4ee5\u663e\u8457\u7684\u63d0\u5347AMR-to-Text\u751f\u6210\u7684\u6027 \u80fd\uff0c\u4f46\u662f\u4e3a\u4e86\u63a2\u7a76\u54ea\u79cd\u5f62\u5f0f\u7684\u53e5\u6cd5\u4fe1\u606f\u5bf9\u751f\u6210\u6027\u80fd\u6700\u4e3a\u6709\u6548\uff0c\u672c\u6587\u505a\u4e86\u8fdb\u4e00\u6b65\u7684\u5b9e\u9a8c\u5206\u6790\u3002 \u4e00 \u4e00 \u4e00 (S (NP (DT the )DT (NN girl )NN (VP (VBD took )VBD (NP (NNS apples )NNS (PP (P from )P (NP (DT a )DT (NN bag )NN )NP )PP )NP )VP )NP )S \u4e8c \u4e8c \u4e8c S NP the girl VP took NP apples PP from NP a bag \u4e09 \u4e09 \u4e09 S NP DT the NN girl VP VBD took NP NNS apples PP P from NP DT a NN bag ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u539f\u59cbAMR\u56fe\uff1a", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "https://nlp.stanford.edu/software/lex-parser.html 3 https://github.com/OpenNMT/OpenNMT-py", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Abstract meaning representation for sembanking", | |
| "authors": [ | |
| { | |
| "first": "Laura", | |
| "middle": [], | |
| "last": "Banarescu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Claire", | |
| "middle": [], | |
| "last": "Bonial", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shu", | |
| "middle": [], | |
| "last": "Cai", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Madalina", | |
| "middle": [], | |
| "last": "Georgescu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kira", | |
| "middle": [], | |
| "last": "Griffitt", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ulf", | |
| "middle": [], | |
| "last": "Hermjakob", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kevin", | |
| "middle": [], | |
| "last": "Knight", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Philipp", | |
| "middle": [], | |
| "last": "Koehn", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Martha", | |
| "middle": [], | |
| "last": "Palmer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nathan", | |
| "middle": [], | |
| "last": "Schneider", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proceedings of 7th Linguistic Annotation Workshop & Interoperability with Discourse", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "178--186", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Laura Banarescu, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. 2013. Abstract meaning repre- sentation for sembanking. In Proceedings of 7th Linguistic Annotation Workshop & Interoperability with Discourse, pages 178-186.", | |
| "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 ACL", | |
| "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 ACL, pages 65-72.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Graph-to-sequence learning using gated graph neural networks", | |
| "authors": [ | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Beck", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Gholamreza", | |
| "middle": [], | |
| "last": "Haffari", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Trevor", | |
| "middle": [], | |
| "last": "Cohn", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "273--283", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Daniel Beck, Gholamreza Haffari, and Trevor Cohn. 2018. Graph-to-sequence learning using gated graph neural networks. In Proceedings of ACL, pages 273-283.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Factorising amr generation through syntax", | |
| "authors": [ | |
| { | |
| "first": "Kris", | |
| "middle": [], | |
| "last": "Cao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of NAACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2157--2163", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kris Cao and Stephen Clark. 2019. Factorising amr generation through syntax. In Proceedings of NAACL, pages 2157-2163.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Structural neural encoders for AMR-to-text generation", | |
| "authors": [ | |
| { | |
| "first": "Marco", | |
| "middle": [], | |
| "last": "Damonte", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shay", | |
| "middle": [ | |
| "B" | |
| ], | |
| "last": "Cohen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of NAACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "3649--3658", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Marco Damonte and Shay B. Cohen. 2019. Structural neural encoders for AMR-to-text generation. In Proceedings of NAACL, pages 3649-3658.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Meteor universal: Language specific translation evaluation for any target language", | |
| "authors": [ | |
| { | |
| "first": "Michael", | |
| "middle": [], | |
| "last": "Denkowski", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Alon", | |
| "middle": [], | |
| "last": "Lavie", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of WMT", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "376--380", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Michael Denkowski and Alon Lavie. 2014. Meteor universal: Language specific translation evaluation for any target language. In Proceedings of WMT, pages 376-380.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Generation from abstract meaning representation using tree transducers", | |
| "authors": [ | |
| { | |
| "first": "Jeffrey", | |
| "middle": [], | |
| "last": "Flanigan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chris", | |
| "middle": [], | |
| "last": "Dyer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Noah", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Smith", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jaime", | |
| "middle": [], | |
| "last": "Carbonell", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of NAACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "731--739", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jeffrey Flanigan, Chris Dyer, Noah A. Smith, and Jaime Carbonell. 2016. Generation from abstract meaning representation using tree transducers. In Proceedings of NAACL, pages 731-739.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Modeling source syntax and semantics for neural amr parsing", | |
| "authors": [ | |
| { | |
| "first": "Donglai", | |
| "middle": [], | |
| "last": "Ge", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Junhui", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Muhua", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shoushan", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of IJCAI", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4975--4981", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "DongLai Ge, Junhui Li, Muhua Zhu, and Shoushan Li. 2019. Modeling source syntax and semantics for neural amr parsing. In Proceedings of IJCAI, pages 4975-4981.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Pointing the unknown words", | |
| "authors": [ | |
| { | |
| "first": "Caglar", | |
| "middle": [], | |
| "last": "Gulcehre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sungjin", | |
| "middle": [], | |
| "last": "Ahn", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ramesh", | |
| "middle": [], | |
| "last": "Nallapati", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bowen", | |
| "middle": [], | |
| "last": "Zhou", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yoshua", | |
| "middle": [], | |
| "last": "Bengio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "140--149", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Caglar Gulcehre, Sungjin Ahn, Ramesh Nallapati, Bowen Zhou, and Yoshua Bengio. 2016. Pointing the unknown words. In Proceedings of ACL, pages 140-149.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Densely connected graph convolutional networks for graph-to-sequence learning", | |
| "authors": [ | |
| { | |
| "first": "Zhijiang", | |
| "middle": [], | |
| "last": "Guo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yan", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhiyang", | |
| "middle": [], | |
| "last": "Teng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Wei", | |
| "middle": [], | |
| "last": "Lu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Transactions of the Association of Computational Linguistics", | |
| "volume": "7", | |
| "issue": "", | |
| "pages": "297--312", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Zhijiang Guo, Yan Zhang, Zhiyang Teng, and Wei Lu. 2019. Densely connected graph convolutional net- works for graph-to-sequence learning. Transactions of the Association of Computational Linguistics, 7:297-312.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "A convolutional neural network for modelling sentences", | |
| "authors": [ | |
| { | |
| "first": "Nal", | |
| "middle": [], | |
| "last": "Kalchbrenner", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Edward", | |
| "middle": [], | |
| "last": "Grefenstette", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Phil", | |
| "middle": [], | |
| "last": "Blunsom", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "655--665", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Nal Kalchbrenner, Edward Grefenstette, and Phil Blunsom. 2014. A convolutional neural network for modelling sentences. In Proceedings of ACL, pages 655-665.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "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. In Proceedings of ICLR.", | |
| "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": [ | |
| "M" | |
| ], | |
| "last": "Rush", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of ACL, System Demonstrations", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "67--72", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Guillaume Klein, Yoon Kim, Yuntian Deng, Jean Senellart, and Alexander M. Rush. 2017. Opennmt: Open-source toolkit for neural machine translation. In Proceedings of ACL, System Demonstrations, pages 67-72.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Neural AMR: Sequence-to-sequence models for parsing and generation", | |
| "authors": [ | |
| { | |
| "first": "Ioannis", | |
| "middle": [], | |
| "last": "Konstas", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Srinivasan", | |
| "middle": [], | |
| "last": "Iyer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Yatskar", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yejin", | |
| "middle": [], | |
| "last": "Choi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Luke", | |
| "middle": [], | |
| "last": "Zettlemoyer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "146--157", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ioannis Konstas, Srinivasan Iyer, Mark Yatskar, Yejin Choi, and Luke Zettlemoyer. 2017. Neural AMR: Sequence-to-sequence models for parsing and generation. In Proceedings of ACL, pages 146-157.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Modeling syntactic and semantic structures in hierarchical phrase-based translation", | |
| "authors": [ | |
| { | |
| "first": "Junhui", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Resnik", | |
| "middle": [], | |
| "last": "Philip", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proceedings of the 2013 conference of the north american chapter of the association for computational linguistics: Human language technologies", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "540--549", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Junhui Li, Resnik Philip, and Daum\u00e9 III Hal. 2013. Modeling syntactic and semantic structures in hierarchical phrase-based translation. In Proceedings of the 2013 conference of the north american chapter of the association for computational linguistics: Human language technologies, pages 540- 549.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Improving event detection with abstract meaning representation", | |
| "authors": [ | |
| { | |
| "first": "Xiang", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Thien", | |
| "middle": [], | |
| "last": "Huu Nguyen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kai", | |
| "middle": [], | |
| "last": "Cao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ralph", | |
| "middle": [], | |
| "last": "Grishman", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proceedings of the first workshop on computing news storylines", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "11--15", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xiang Li, Thien Huu Nguyen, Kai Cao, and Ralph Grishman. 2015. Improving event detection with abstract meaning representation. In Proceedings of the first workshop on computing news storylines, pages 11-15.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Modeling source syntax for neural machine translation", | |
| "authors": [ | |
| { | |
| "first": "Junhui", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Deyi", | |
| "middle": [], | |
| "last": "Xiong", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhaopeng", | |
| "middle": [], | |
| "last": "Tu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of ACL-2017", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "688--697", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Junhui Li, Deyi Xiong, and Zhaopeng Tu. 2017. Modeling source syntax for neural machine translation. In Proceedings of ACL-2017, pages 688-697.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "The stanford corenlp natural language processing tooklkit", | |
| "authors": [ | |
| { | |
| "first": "Christopher", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Manning", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mihai", | |
| "middle": [], | |
| "last": "Surdeanu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "John", | |
| "middle": [], | |
| "last": "Bauer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jenny", | |
| "middle": [], | |
| "last": "Finkel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Steven", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Bethard", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Mcclosky", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of ACL-2014", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "55--60", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Christopher D. Manning, Mihai Surdeanu, John Bauer, Jenny Finkel, Steven J. Bethard, and David McClosky. 2014. The stanford corenlp natural language processing tooklkit. In Proceedings of ACL-2014, pages 55-60.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Addressing a question answering challenge by combining statistical methods with inductive rule learning and reasoning", | |
| "authors": [ | |
| { | |
| "first": "Arindam", | |
| "middle": [], | |
| "last": "Mitra", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chitta", | |
| "middle": [], | |
| "last": "Baral", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Thirtieth AAAI Conference on Artificial Intelligence", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Arindam Mitra and Chitta Baral. 2016. Addressing a question answering challenge by combining statisti- cal methods with inductive rule learning and reasoning. In Thirtieth AAAI Conference on Artificial Intelligence.", | |
| "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": "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, Salim Roukos, Ward Todd, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of ACL, pages 311-318.", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "chrf++: words helping character n-grams", | |
| "authors": [ | |
| { | |
| "first": "Maja", | |
| "middle": [], | |
| "last": "Popovi\u0107", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of WMT", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "612--618", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Maja Popovi\u0107. 2017. chrf++: words helping character n-grams. In Proceedings of WMT, pages 612-618.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Generating english from abstract meaning representations", | |
| "authors": [ | |
| { | |
| "first": "Nima", | |
| "middle": [], | |
| "last": "Pourdamghani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kevin", | |
| "middle": [], | |
| "last": "Knight", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ulf", | |
| "middle": [], | |
| "last": "Hermjakob", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of the 9th International Natural Language Generation conference", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "21--25", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Nima Pourdamghani, Kevin Knight, and Ulf Hermjakob. 2016. Generating english from abstract meaning representations. In Proceedings of the 9th International Natural Language Generation conference, pages 21-25.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "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, Barry Haddow, and Alexandra Birch. 2016. Neural machine translation of rare words with subword units. In Proceedings of ACL, pages 1715-1725.", | |
| "links": null | |
| }, | |
| "BIBREF23": { | |
| "ref_id": "b23", | |
| "title": "Self-attention with relative position representations", | |
| "authors": [ | |
| { | |
| "first": "Peter", | |
| "middle": [], | |
| "last": "Shaw", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jakob", | |
| "middle": [], | |
| "last": "Uszkoreit", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ashish", | |
| "middle": [], | |
| "last": "Vaswani", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of NAACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "464--468", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Peter Shaw, Jakob Uszkoreit, and Ashish Vaswani. 2018. Self-attention with relative position represen- tations. In Proceedings of NAACL, pages 464--468.", | |
| "links": null | |
| }, | |
| "BIBREF24": { | |
| "ref_id": "b24", | |
| "title": "Amr-to-text generation as a traveling salesman problem", | |
| "authors": [ | |
| { | |
| "first": "Linfeng", | |
| "middle": [], | |
| "last": "Song", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yue", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaochang", | |
| "middle": [], | |
| "last": "Peng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhiguo", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Gildea", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proceedings of EMNLP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2084--2089", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Linfeng Song, Yue Zhang, Xiaochang Peng, Zhiguo Wang, and Daniel Gildea. 2016. Amr-to-text gener- ation as a traveling salesman problem. In Proceedings of EMNLP, pages 2084-2089.", | |
| "links": null | |
| }, | |
| "BIBREF25": { | |
| "ref_id": "b25", | |
| "title": "Amr-to-text generation with synchronous node replacement grammar", | |
| "authors": [ | |
| { | |
| "first": "Linfeng", | |
| "middle": [], | |
| "last": "Song", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiaochang", | |
| "middle": [], | |
| "last": "Peng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yue", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhiguo", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Gildea", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "7--13", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Linfeng Song, Xiaochang Peng, Yue Zhang, Zhiguo Wang, and Daniel Gildea. 2017. Amr-to-text gener- ation with synchronous node replacement grammar. In Proceedings of ACL, pages 7-13.", | |
| "links": null | |
| }, | |
| "BIBREF26": { | |
| "ref_id": "b26", | |
| "title": "A graph-to-sequence model for AMR-to-text generation", | |
| "authors": [ | |
| { | |
| "first": "Linfeng", | |
| "middle": [], | |
| "last": "Song", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yue", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Zhiguo", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Gildea", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1616--1626", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Linfeng Song, Yue Zhang, Zhiguo Wang, and Daniel Gildea. 2018. A graph-to-sequence model for AMR-to-text generation. In Proceedings of ACL, pages 1616-1626.", | |
| "links": null | |
| }, | |
| "BIBREF27": { | |
| "ref_id": "b27", | |
| "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. In Advances in neural information processing systems, pages 3104-3112.", | |
| "links": null | |
| }, | |
| "BIBREF28": { | |
| "ref_id": "b28", | |
| "title": "A discriminative model for semantics-to-string translation", | |
| "authors": [ | |
| { | |
| "first": "Ale\u0161", | |
| "middle": [], | |
| "last": "Tamchyna", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chris", | |
| "middle": [], | |
| "last": "Quirk", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Michel", | |
| "middle": [], | |
| "last": "Galley", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proceedings of the 1st Workshop on Semantics-Driven Statistical Machine Translation (S2MT 2015)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "30--36", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ale\u0161 Tamchyna, Chris Quirk, and Michel Galley. 2015. A discriminative model for semantics-to-string translation. In Proceedings of the 1st Workshop on Semantics-Driven Statistical Machine Translation (S2MT 2015), pages 30-36, Beijing, China, July. Association for Computational Linguistics.", | |
| "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, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N.Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Proceedings of NIPS, pages 5998-6008.", | |
| "links": null | |
| }, | |
| "BIBREF30": { | |
| "ref_id": "b30", | |
| "title": "Grammar as a foreign language", | |
| "authors": [ | |
| { | |
| "first": "Oriol", | |
| "middle": [], | |
| "last": "Vinyals", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Lukasz", | |
| "middle": [], | |
| "last": "Kaiser", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Terry", | |
| "middle": [], | |
| "last": "Koo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Slav", | |
| "middle": [], | |
| "last": "Petrov", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ilya", | |
| "middle": [], | |
| "last": "Sutskever", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Geoffrey", | |
| "middle": [], | |
| "last": "Hinton", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Advances in Neural Information Processing Systems", | |
| "volume": "28", | |
| "issue": "", | |
| "pages": "2773--2781", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Oriol Vinyals, Lukasz Kaiser, Terry Koo, Slav Petrov, Ilya Sutskever, and Geoffrey Hinton. 2015. Gram- mar as a foreign language. In Advances in Neural Information Processing Systems 28, pages 2773- 2781.", | |
| "links": null | |
| }, | |
| "BIBREF31": { | |
| "ref_id": "b31", | |
| "title": "Modeling graph structure in transformer for better amr-to-text generation", | |
| "authors": [ | |
| { | |
| "first": "Jie", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Junhui", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Muhua", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Longhua", | |
| "middle": [], | |
| "last": "Qian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Min", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Guodong", | |
| "middle": [], | |
| "last": "Zhou", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of EMNLP-2019", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "5458--5467", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jie Zhu, Junhui Li, Muhua Zhu, Longhua Qian, Min Zhang, and Guodong Zhou. 2019. Modeling graph structure in transformer for better amr-to-text generation. In Proceedings of EMNLP-2019, pages 5458-5467.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF0": { | |
| "text": "LDC2017T10\u3002\u4e24\u4efd\u8bed\u6599\u96c6\u5206\u522b\u5305\u542b\u4e86 16,833 \u548c 36,521 \u6761\u8bad\u7ec3\u6570\u636e\uff0c\u5e76\u4e14 \u5171\u4eab\u4e86 1,368 \u6761\u9a8c\u8bc1\u96c6\u548c 1,371 \u6761\u6d4b\u8bd5\u96c6\u3002\u8bad\u7ec3\u96c6\u548c\u9a8c\u8bc1\u96c6\u4f7f\u7528\u65af\u5766\u798f\u89e3\u6790\u5668 2 \u83b7\u53d6\u5230\u76ee\u6807\u7aef\u53e5 \u5b50\u6240\u5bf9\u5e94\u7684 Penn treebank-style \u98ce\u683c\u7684\u7ed3\u6784\u53e5\u6cd5\u6811\u3002 4.2 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c\u8bbe \u8bbe \u8bbe\u7f6e \u7f6e \u7f6e \u672c\u6587\u5206\u522b\u901a\u8fc7\u4f7f\u7528 10K \u548c 20K \u7684\u64cd\u4f5c\u6570\u6765\u5bf9 LDC2015E86 \u548c LDC2017T10 \u4e24\u4efd\u8bed\u6599\u8fdb", | |
| "num": null, | |
| "content": "<table><tr><td>\u8ba1\u7b97\u8bed\u8a00\u5b66</td><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"2\">\u76ee\u6807\u7aef\u53e5\u5b50\uff1a System (Konstas et al., 2017) *</td><td colspan=\"6\">LDC2015E86 BLEU Meteor chrF++ BLEU Meteor chrF++ LDC2017T10 The girl took apples from a bag 22.00 -----</td></tr><tr><td colspan=\"3\">\u77ed\u8bed\u7ed3\u6784\u53e5\u6cd5\u6811\uff1a (Cao and Clark, 2019) *</td><td>23.5</td><td>-</td><td>-</td><td>26.8</td><td>-</td><td>-</td></tr><tr><td colspan=\"2\">(Song et al., 2018) \u2020</td><td/><td>S 23.30</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td colspan=\"2\">(Beck et al., 2018) \u2020</td><td/><td>-</td><td>-</td><td>-</td><td>23.3</td><td>-</td><td>50.4</td></tr><tr><td colspan=\"5\">NP (Damonte and Cohen, 2019) \u2020 24.40 23.60 VP</td><td>-</td><td colspan=\"2\">24.54 24.07</td><td>-</td></tr><tr><td colspan=\"2\">(Guo et al., 2019) \u2020</td><td/><td>25.7</td><td>-</td><td>-</td><td>27.6</td><td>-</td><td>57.3</td></tr><tr><td colspan=\"5\">DT (Zhu et al., 2019)(CNN-based) \u2020 29.10 35.00 NN VBD NP</td><td colspan=\"3\">PP 62.10 31.82 36.38</td><td>64.05</td></tr><tr><td colspan=\"2\">(Song et al., 2016) \u2021</td><td/><td>22.44</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td>Baseline1</td><td>The</td><td>girl</td><td colspan=\"2\">took NNS 25.13 33.08</td><td colspan=\"3\">P 59.36 26.98 34.36 NP</td><td>61.05</td></tr><tr><td colspan=\"2\">+ Syntax Baseline2</td><td/><td colspan=\"5\">26.39 33.63 apples from DT NN 59.84 28.13 34.82 28.64 34.83 61.89 31.10 36.07</td><td>61.73 63.87</td></tr><tr><td colspan=\"2\">+ Syntax</td><td/><td colspan=\"5\">a bag 29.71 35.49 62.52 32.28 36.81 64.61</td></tr><tr><td colspan=\"2\">\u7ebf\u6027\u5316\u4e4b\u540e\u7684\u53e5\u6cd5\u6811\uff1a</td><td/><td/><td/><td/><td/></tr><tr><td colspan=\"8\">S NP DT the NN girl VP VBD took NP NNS apples PP P from NP DT a NN bag</td></tr><tr><td colspan=\"8\">\u56fe 3. \u4e00\u4e2a\u76ee\u6807\u7aef\u53e5\u5b50\u89e3\u6790\u7684\u77ed\u8bed\u7ed3\u6784\u53e5\u6cd5\u6811\u53ca\u5176\u7ebf\u6027\u5316\u793a\u4f8b</td></tr><tr><td colspan=\"8\">\u6cd5\u4fe1\u606f\u3002\u4e3a\u4e86\u4f7f\u6a21\u578b\u80fd\u591f\u5b66\u4e60\u5230\u76ee\u6807\u7aef\u53e5\u5b50\u7684\u53e5\u6cd5\u4fe1\u606f\u548c\u5185\u90e8\u7ed3\u6784\uff0c\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u663e\u5f0f\u7684\u65b9\u6cd5</td></tr><tr><td>\u6765\u878d\u5165\u76ee\u6807\u7aef\u7684\u53e5\u6cd5\u4fe1\u606f\u3002</td><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"8\">\u878d\u5408\u76ee\u6807\u7aef\u53e5\u6cd5\u4fe1\u606f\u7684\u57fa\u672c\u601d\u60f3\u662f\u5c06\u76ee\u6807\u7aef\u53e5\u5b50\u7ecf\u8fc7\u89e3\u6790\u5f97\u5230\u53e5\u6cd5\u6811\uff0c\u4e4b\u540e\u518d\u901a\u8fc7\u6df1\u5ea6\u4f18</td></tr><tr><td colspan=\"8\">\u5148\u904d\u5386\u5f97\u5230\u6700\u7ec8\u7684\u53e5\u6cd5\u6807\u7b7e\u5e8f\u5217\u3002\u53e5\u6cd5\u7684\u6807\u6ce8\u5f62\u5f0f\u5927\u81f4\u6709\u4e24\u79cd\uff0c\u77ed\u8bed\u7ed3\u6784\u53e5\u6cd5\u6811\u548c\u4f9d\u5b58\u53e5\u6cd5</td></tr><tr><td colspan=\"8\">\u6811\u3002\u672c\u6587\u5728\u8bad\u7ec3\u65f6\u9009\u62e9\u4f7f\u7528\u7ebf\u6027\u5316\u7684\u77ed\u8bed\u7ed3\u6784\u53e5\u6cd5\u6811(Vinyals et al., 2015)\u6765\u66ff\u6362\u76ee\u6807\u7aef\u7684\u53e5</td></tr><tr><td colspan=\"8\">\u5b50\u3002\u5982\u56fe3\u6240\u793a\uff0c\u7ed9\u51fa\u4e86\u4e00\u4e2a\u76ee\u6807\u7aef\u53e5\u5b50\u89e3\u6790\u7684\u77ed\u8bed\u7ed3\u6784\u53e5\u6cd5\u6811\u548c\u5176\u7ebf\u6027\u5316\u7ed3\u679c\u3002\u4e4b\u6240\u4ee5\u9009\u62e9\u77ed</td></tr><tr><td colspan=\"8\">\u8bed\u7ed3\u6784\u53e5\u6cd5\u6811\uff0c\u662f\u56e0\u4e3a\u4e0e\u4f9d\u5b58\u6811\u76f8\u6bd4\uff0c\u5b83\u5177\u6709\u826f\u597d\u7684\u7ebf\u6027\u5316\u987a\u5e8f\u7684\u4f18\u70b9\u3002\u6b64\u5916\uff0c\u77ed\u8bed\u7ed3\u6784\u53e5\u6cd5</td></tr><tr><td colspan=\"8\">\u6811\u4e5f\u66f4\u5bb9\u6613\u5b9e\u73b0\uff0c\u56e0\u4e3a\u5b83\u4eec\u6709\u6548\u7684\u5bf9\u5e94\u53e5\u5b50\u4e2d\u5355\u8bcd\u7684\u987a\u5e8f\u3002\u5728\u89e3\u7801\u9636\u6bb5\uff0c\u53ea\u9700\u8981\u5c06\u53e5\u6cd5\u6807\u7b7e\u53bb</td></tr><tr><td colspan=\"3\">\u9664\u4e4b\u540e\uff0c\u5c31\u662f\u6700\u7ec8\u9884\u6d4b\u751f\u6210\u7684\u53e5\u5b50\u3002</td><td/><td/><td/><td/></tr><tr><td colspan=\"8\">\u4e0d\u5e78\u7684\u662f\uff0cAMR\u6807\u6ce8\u6570\u636e\u5e76\u6ca1\u6709\u53d1\u653e\u53e5\u6cd5\u6807\u6ce8\u6570\u636e\u3002\u56e0\u6b64\uff0c\u672c\u6587\u4f7f\u7528\u65af\u5766\u798f\u89e3\u6790\u5668</td></tr><tr><td colspan=\"8\">(Stanford Parser)(Manning et al., 2014)\u89e3\u6790\u8bad\u7ec3\u96c6\u548c\u9a8c\u8bc1\u96c6\u8bed\u6599\uff0c\u4ece\u800c\u83b7\u5f97\u5bf9\u5e94\u7ed3\u6784\u8bed\u6cd5\u6811</td></tr><tr><td colspan=\"2\">\u7684\u94f6\u8bed\u6599(Silver-Standard)\u3002</td><td/><td/><td/><td/><td/></tr><tr><td>4 \u5b9e \u5b9e \u5b9e\u9a8c \u9a8c \u9a8c</td><td/><td/><td/><td/><td/><td/></tr><tr><td>4.1 \u6570 \u6570 \u6570\u636e \u636e \u636e\u96c6 \u96c6 \u96c6</td><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"8\">\u5b9e\u9a8c\u4e2d m \u7684\u5927\u5c0f\u5e38\u8bbe\u7f6e\u4e3a4\u3002 \u4e3a\u4e86\u8bc4\u4f30\u65b9\u6cd5\u7684\u6709\u6548\u6027\uff0c\u672c\u6587\u4f7f\u7528LDC\u53d1\u884c\u7684\u73b0\u5b58\u7684\u4e24\u4efd\u6807\u51c6\u82f1\u6587\u8bed\u6599\u96c6\u8fdb\u884c\u5b9e\u9a8c\uff0c\u5206\u522b</td></tr><tr><td colspan=\"8\">3.3 \u6570 \u6570 \u6570\u636e \u636e \u636e\u7a00 \u7a00 \u7a00\u758f \u758f \u758f\u6027 \u6027 \u6027 \u5728\u8bad\u7ec3AMR-to-Text\u6a21\u578b\u7684\u65f6\u5019\uff0c\u56e0\u4e3a\u8bed\u6599\u6570\u91cf\u7684\u9650\u5236\uff0c\u5e38\u5e38\u4f1a\u53d7\u5230\u6570\u636e\u7a00\u758f\u6027\u7684\u5f71\u54cd\u3002 \u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u524d\u4eba\u7684\u5de5\u4f5c\u6709\u91c7\u7528\u533f\u540d\u5316\u7684\u65b9\u6cd5\u6765\u5220\u9664\u547d\u540d\u5b9e\u4f53\u548c\u7f55\u89c1\u8bcd(Konstas et al., 2017)\uff0c\u6216\u8005\u4f7f\u7528\u590d\u5236\u673a\u5236(Gulcehre et al., 2016)\u6765\u5b66\u4e60\uff0c\u4f7f\u6a21\u578b\u53ef\u4ee5\u5b66\u4f1a\u4ece\u6e90\u7aef\u8f93\u5165\u590d\u5236\u672a \u767b\u5f55\u8bcd\u5230\u76ee\u6807\u7aef\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u63d0\u51fa\u4f7f\u7528\u5b57\u8282\u5bf9\u7f16\u7801(BPE)(Sennrich et al., 2016)\u5c06\u672a\u767b \u5f55\u8bcd\u62c6\u5206\u6210\u66f4\u7ec6\u7c92\u5ea6\uff0c\u66f4\u9ad8\u9891\u7684\u5355\u8bcd\u3002\u518d\u6839\u636e\u8be5\u4efb\u52a1\u7684\u7279\u6027\u8003\u8651\uff0c\u5171\u4eab\u4e86\u6e90\u7aef\u548c\u76ee\u6807\u7aef\u7684\u8bcd \u662f LDC2015E86 \u548c \u884cBPE\u64cd\u4f5c\u3002\u4eceBPE\u5904\u7406\u4e4b\u540e\u7684\u8bad\u7ec3\u96c6\u4e2d\u6839\u636e\u8bcd\u9891\u5efa\u7acb\u8bcd\u6c47\u8868\uff0c\u53c2\u8003Ge\u7b49(2019)\u7684\u5de5\u4f5c\uff0c\u5171\u4eab</td></tr><tr><td colspan=\"8\">\u8868\u3002Zhu\u7b49(2019)\u4e5f\u5728\u5b9e\u9a8c\u4e2d\u8bc1\u660e\u4e86\u8be5\u65b9\u6cd5\u7684\u6709\u6548\u6027\u3002 \u4e86\u6e90\u7aef\u548c\u76ee\u6807\u7aef\u7684\u8bcd\u6c47\u8868\u3002\u4e3a\u4e86\u516c\u5e73\u7684\u5bf9\u6bd4\uff0c\u6a21\u578b\u4e2d\u7684\u8bcd\u5411\u91cf\u4f7f\u7528\u968f\u673a\u521d\u59cb\u5316\u7684\u65b9\u5f0f\u3002</td></tr><tr><td colspan=\"8\">\u672c\u6587\u4f7f\u7528OpenNMT (Klein et al., 2017)\u7684\u6846\u67b6\u4f5c\u4e3aTransformer\u7684\u57fa\u51c6\u6a21\u578b 3 \u3002\u5728\u8d85\u53c2\u6570\u7684 3.4 \u878d \u878d \u878d\u5408 \u5408 \u5408\u53e5 \u53e5 \u53e5\u6cd5 \u6cd5 \u6cd5\u4fe1 \u4fe1 \u4fe1\u606f \u606f \u606f \u8bbe\u7f6e\u4e0a\uff0c\u6a21\u578b\u7684\u7f16\u7801\u5668\u548c\u89e3\u7801\u5668\u4e3a 6 \u5c42\u3002\u5728\u4f18\u5316\u5668\u65b9\u9762\uff0c\u672c\u6587\u4f7f\u7528beta1=0.1(Kingma and Ba,</td></tr><tr><td colspan=\"8\">\u524d\u4eba\u7684\u5de5\u4f5c\u90fd\u662f\u4f7f\u7528\u5e73\u884c\u8bed\u6599\u6765\u8fdb\u884c\u8bad\u7ec3\uff0c\u8f93\u5165\u6e90\u7aefAMR\u56fe\u53bb\u751f\u6210\u5bf9\u5e94\u7684\u53e5\u5b50\u3002\u4ed6\u4eec\u5927\u90fd 2015)\u7684Adam\u4f18\u5316\u7b97\u6cd5\u3002\u81ea\u6ce8\u610f\u5934\u7684\u6570\u91cf\u8bbe\u7f6e\u4e3a 8\u3002\u6b64\u5916\uff0c\u6a21\u578b\u4e2d\u5411\u91cf\u548c\u9690\u85cf\u72b6\u6001\u7684\u7ef4\u5ea6\u4f4d\u7f6e</td></tr><tr><td colspan=\"8\">\u662f\u5c06\u53e5\u5b50\u89c6\u4e3a\u5355\u8bcd\u5e8f\u5217\uff0c\u4f46\u662f\u5374\u5ffd\u7565\u4e86\u53e5\u5b50\u672c\u8eab\u7684\u4e00\u4e9b\u5916\u90e8\u77e5\u8bc6\uff0c\u6ca1\u6709\u8003\u8651\u5230\u53e5\u5b50\u4e2d\u6f5c\u85cf\u7684\u53e5 \u4e3a 512\uff0c\u6279\u5904\u7406\u5927\u5c0f(batch size)\u8bbe\u7f6e\u4e3a 4096\u3002\u4e3a\u4e86\u6a21\u578b\u8ba1\u7b97\u901f\u5ea6\u8003\u8651\uff0c\u9650\u5b9a\u8def\u5f84\u6807\u7b7e\u7684\u6700\u5927</td></tr><tr><td colspan=\"4\">0 \u5f53\u540c\u65f6\u5b58\u5728\u591a\u6761\u8def\u5f84\u7ec4\u5408\u65f6\uff0c\u9ed8\u8ba4\u9009\u62e9\u6700\u77ed\u7684\u90a3\u4e00\u6761\u3002</td><td/><td/><td/></tr><tr><td colspan=\"8\">1 (Zhu et al., 2019)\u4f7f\u7528\u4e86\u591a\u79cd\u65b9\u6cd5\u6765\u5b66\u4e60\u56fe\u7ed3\u6784\u8868\u793a\u65b9\u6cd5\uff0c\u672c\u6587\u9009\u62e9\u4e86CNN-based\u8fd9\u4e00\u65b9\u6cd5\u4f5c\u4e3a\u57fa\u7ebf\u6a21\u578b\u3002</td></tr></table>", | |
| "type_str": "table", | |
| "html": null | |
| } | |
| } | |
| } | |
| } |