| { |
| "paper_id": "Y06-1021", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T13:34:09.340956Z" |
| }, |
| "title": "Auto-extracting Paraphrases of Letter-word Phrases in Live Texts 1", |
| "authors": [ |
| { |
| "first": "Zezhi", |
| "middle": [], |
| "last": "Zheng", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Xiamen University", |
| "location": { |
| "postCode": "361005", |
| "settlement": "Xiamen" |
| } |
| }, |
| "email": "zhengzz@xmu.edu.cn" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "In this paper we will discuss the Auto-extraction of paraphrases of letter-word phrases in live Chinese texts. The paper discusses the modes of conventional dictionaries firstly, and then gives the principles of paraphrase of letter-word phrases; with an analysis of the examples of letter-word phrases paraphrases secondly, and then gives their formalized denotations and presents an auto-recognizing algorithm for bilingual synonymous letter-word phrases; lastly, based on the labeled result of our auto-labeling software of letter-word phrase, uses the vector space distance to extract the paraphrase of letter-word phrases in live Chinese texts.", |
| "pdf_parse": { |
| "paper_id": "Y06-1021", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "In this paper we will discuss the Auto-extraction of paraphrases of letter-word phrases in live Chinese texts. The paper discusses the modes of conventional dictionaries firstly, and then gives the principles of paraphrase of letter-word phrases; with an analysis of the examples of letter-word phrases paraphrases secondly, and then gives their formalized denotations and presents an auto-recognizing algorithm for bilingual synonymous letter-word phrases; lastly, based on the labeled result of our auto-labeling software of letter-word phrase, uses the vector space distance to extract the paraphrase of letter-word phrases in live Chinese texts.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "Traditional dictionaries have the veracious meaning expression, and rational, normative language, and better systematic semantic systems. The paraphrases in dictionaries are generalized from the collection of example sentences which are gathered from the fields that seemed most feasible. Therefore compiling a dictionary always cost a great deal manpower, material resources and time. The renewed period of traditional dictionaries has not adapted to the increasing and changing of new words and new meanings, so using computers to help the compile of dictionaries is brought forward.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "About the extraction for the definition, [zhang Yan, 2003 ] instructed a definition extraction means for Chinese terms, which based on Chinese syntactic parsing. They segmented and tagged the corpora of computer domain with part-of-speech firstly, then used two parsers to gain structures and phrases of sentences. They summarized the structures characteristics of term definitions and automatically extracted the patterns of definitions. Finally they gave an algorithm to define a new term according to their knowledge database. Their algorithm to define a term needs concept semantic knowledge and other knowledge, and segmented and tagged texts. In 2004, XU Yong et al presented an experiment of web based term definition retrieval system. For a given term, the system used the algorithm based on term definition patterns with indicated words to extract its definition. The two means of definition extraction are all for given terms, and have available to some extent.", |
| "cite_spans": [ |
| { |
| "start": 41, |
| "end": 57, |
| "text": "[zhang Yan, 2003", |
| "ref_id": "BIBREF3" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "We think that definitions are for terms, the paraphrases are suitable for letter-word phrases (hereinafter we will use \"LWP\" to denote \"letter-word phrase\") in live Chinese texts in general field. Therefore in this paper we'll discuss the principles of paraphrases of LWPs firstly, and then generalize their formalized patterns based on the collection of paraphrases of LWPs. Secondly, we'll analyse the especial paraphrases, bilingual synonymous LWPs, and present an algorithm to obtain this kind paraphrases. Thirdly, we'll label the LWPs using our auto-labeling software, and then put forward an algorithm to extract the paraphrases of LWPs.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "increasing quickly, we have to use machines to help extracting their paraphrases. But what is the extent? What are the principles?", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "With analyzing different requirements of dictionaries and investigating paraphrases of LWPs in live texts, we would conclude our interpretation criterion.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "From the book by FU Huiqing, and the paper by LI Xiyin, we can conclude that traditional dictionaries use about four forms to interpret a word. a) Genus name and differentiae addition. This pattern is used usually to define a term. The differentiae would be properties, reasons, degrees, functions, effects, or extents etc.. b) Synonymous words interpretation. Interpreting a word with its synonymous words. c) Translation with bilingual synonymous words. d) Depiction, explanation. Depicting or explaining a word with its phenomenon, characters, relations, parable etc. [ZHENG Shupu, 2005] Russian Scelba considered that cyclopedias and common dictionaries are different in defining a word. Common dictionaries provide an interpretation to help readers to understand the meaning of a word, not a complete interpretation of the word. And long before ,Russian term scholar Liefu'ermaciji presented that terms in textbooks or in news papers are secondhand, which are derived from scientific terms, their interpretation demands are different from scientific term system, for they just relate to some keywords, and don't relate to whole term system.", |
| "cite_spans": [ |
| { |
| "start": 571, |
| "end": 590, |
| "text": "[ZHENG Shupu, 2005]", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The LWPs in the paper were from general media, most of them were terms or proper nouns. According to the argumentation above, we think that paraphrases of LWPs are not the cyclopaedic definitions, they are generally common interpretations. In fact, for most readers it is better to give an interpretation like \"a standard of information communication.\", \"a sort of computer virus.\" than a scientific definition. Thereinafter, our paraphrases of LWPs are common dictionary interpretations.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "Our paraphrases extracting process is based on no segmented and tagged texts, we have to find out LWPs firstly, and then retrieval the paraphrase sentences.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "The paraphrase examples are a byproduct of our collating software, videlicet, during our collating we can enregister the encountered interpretations by the record function of the software. So we get hold of 400 paraphrases of LWPs. Some of these paraphrases are repeated, but they were from different texts, and with different contents, with different points of view. They are beneficial for us to find patterns of paraphrase.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Through investigation, we find that there were two types paraphrases of LWPs, one has steady forms, the other has no form. The former paraphrases with mark words. Such as:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "FAD2 \u662f\u4e00\u79cd\u5c06\u9971\u548c\u8102\u80aa\u9178\u8f6c\u5316\u6210\u4e0d\u9971\u548c\u8102\u80aa\u9178\u7684\u9176\uff0c\u5bf9\u54fa\u4e73\u52a8\u7269\u751f\u957f\u5177\u6709\u91cd\u8981\u4f5c\u7528\uff0c\u5728 \u83e0\u83dc\u7b49\u690d\u7269\u4e2d\u5b58\u5728\u3002\u4f46\u662f\u54fa\u4e73\u52a8\u7269\uff0c\u5305\u62ec\u4eba\u7c7b\u4f53\u5185\u4e0d\u542b\u8fd9\u79cd\u9176\uff0c\u5fc5\u987b\u4ece\u98df\u7269\u4e2d\u6444\u53d6\u3002 \u53ef\u6015\u7684\u7535\u78c1\u6b66\u5668\u5bb6\u65cf\u4e2d\u8fd8\u6709\"\u6740\u4eba\u96f7\u8fbe\" \uff0c\u7b80\u79f0 RF/MO \u6b66\u5668\u3002\u5b83\u662f\u529f\u7387\u7279\u5f3a\u7684\u5c04\u9891\u3001\u8d85\u9ad8 \u9891\u6216\u5fae\u6ce2\u6b66\u5668\uff0c\u5176\u53d1\u5c04\u529f\u7387\u4ece\u51e0\u4ebf\u74e6\u5230\u51e0\u5341\u4e07\u4ebf\u74e6\u4e0d\u7b49\u3002\u6839\u636e\u6240\u53d1\u5c04\u7535\u78c1\u6ce2\u4e0d\u540c\u7684\u9891\u7387\u3001 \u8c03\u5236\u65b9\u5f0f\u548c\u529f\u7387\u5f3a\u5ea6\uff0cRF/MO \u6b66\u5668\u53ef\u4ee5\u5728\u4eba\u4f53\u7684\u4e0d\u540c\u7ec4\u7ec7\u6216\u5668\u5b98\u4ea7\u751f\u4e0d\u540c\u7684\u6548\u5e94\u3002\u4eba\u4f53 \u6700\u6613\u906d\u53d7\u7535\u78c1\u6b66\u5668\u653b\u51fb\u7684\u7ec4\u7ec7\u548c\u5668\u5b98\u662f\u5927\u8111\u3001\u8116\u5b50\u3001\u80f8\u90e8\u548c\u751f\u6b96\u817a\u3002\u53d7\u5230\u653b\u51fb\u65f6\u7684\u75c7\u72b6 \u662f\u8eab\u5fc3\u75b2\u60eb\u3001\u8bb0\u5fc6\u7d0a\u4e71\u3001\u76ae\u80a4\u751f\u75c5\u3001\u773c\u775b\u51fa\u8840\u3001\u767d\u5185\u969c\u3001\u89d2\u819c\u548c\u89c6\u7f51\u819c\u635f\u4f24\uff0c\u751a\u81f3\u7f79\u60a3 \u764c\u75c7\u3002 The \"LWP\u662f\u4e00\u79cd\u2026\u2026\u7684\u2026\u2026\",\"LWP\uff0c\u5b83\u662f/\u662f\u2026\u2026\"are mark words. The latter interprets a LWP using its contexts without mark words, and readers may see the meaning from the contexts, such as:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "\u636e\u300a\u65e5\u520a\u5de5\u4e1a\u65b0\u95fb\u300b\u62a5\u9053\uff0c\u787c\u540c\u4f4d\u7d20\u6709 B10 \u548c B11\uff0c\u4e8c\u8005\u7684\u5b58\u5728\u6bd4\u4f8b\u4e3a 20\u223680\u3002B10 \u80fd \u5438\u6536\u4e2d\u5b50\uff0c\u7528\u4e2d\u5b50\u675f\u7167\u5c04\u6d53\u7f29\u787c\uff0c\u4f1a\u4ea7\u751f\u963f\u5c14\u6cd5\u5c04\u7ebf\u3002\u5b83\u7684\u80fd\u91cf\u8db3\u4ee5\u6740\u6b7b\u764c\u7ec6\u80de\uff0c\u4f46\u53c8 \u4e0d\u4f24\u5bb3\u6b63\u5e38\u7ec4\u7ec7\uff0c\u56e0\u800c\u4e0d\u4f1a\u4ea7\u751f\u526f\u4f5c\u7528\u3002 \u9664\u6765\u81ea\u82f1\u56fd\u3001\u52a0\u62ff\u5927\u3001\u5fb7\u56fd\u3001\u6cd5\u56fd\u3001\u610f\u5927\u5229\u3001\u65e5\u672c\u3001\u7f8e\u56fd\u548c\u4fc4\u7f57\u65af\u7684 8 \u56fd\u9886\u5bfc\u4eba\u5916\uff0c\u975e \u6d32\u7684\u5357\u975e\u3001\u5c3c\u65e5\u5229\u4e9a\u3001\u57c3\u53ca\u3001\u585e\u5185\u52a0\u5c14\u3001\u5b89\u54e5\u62c9 5 \u56fd\u7684\u9886\u5bfc\u4eba\u4e5f\u5c06\u53c2\u52a0\u4f1a\u8bae\u3002\u8fd9\u5728 G8 \u9996 \u8111\u4f1a\u8bae\u5386\u53f2\u4e0a\u8fd8\u662f\u7b2c\u4e00\u6b21\uff0c\u51f8\u663e\u672c\u6b21\u4f1a\u8bae\u7684\u975e\u6d32\u8bae\u9898\u3002\u8054\u5408\u56fd\u79d8\u4e66\u957f\u5b89\u5357\u4e5f\u5c06\u53c2\u52a0\u4f1a\u8bae\u3002", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "This sort LWP may have deep structure, but now we haven't do farther analysis. This paper we only study these LWP's paraphrases with steady forms.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "From our invesitigation, we find that just three types of common dictionarie paraphrases are shown in the paraphrases of LWP, such as genus name and differentiae addition, translation with bilingual synonymous words and depiction, explanation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "The paraphrases of LWP, which could be formalized, can be divided into six groups, such as: a) the form with \"\u662f\", like \"LWP \u662f\u4e00\u79cd\u2026\u2026\", \"LWP \u662f\u2026\u2026\u7684\u7f29\u5199|\u7b80\u79f0\uff0c\u662f\u4e00\u79cd\u2026\u2026\", etc. b) the form with \"\u5373\", like \"\u2026\u2026LWP \u5373|\uff0c\u5373\u2026\u2026\" c) the form with \"\u540d\u4e3a\", like \"\u7814\u5236\u51fa|\u63a8\u51fa\u7684|\u751f\u4ea7\u7684|\u53d1\u73b0\u7684\u540d\u4e3a LWP\u2026\u2026\u7684\u2026\u2026\" d) the form with \"\u79f0\u4e3a\", like \"\u88ab\u79f0\u4e3a|\u7b80\u79f0\u4e3a|\u79f0\u4e3a LWP \u2026\u2026\u7684\u2026\u2026\" e) translation with bilingual synonymous words, like \"\u4e2d\u56fd\u77f3\u6cb9\u96c6\u56e2\u7269\u63a2\u5c40(BGP),\u4e16\u754c\u8d38\u6613\u7ec4\u7ec7 (WTO)\" f) dash paraphrases, like \"\u2026\u2026--LWP\u2026\u2026\".", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "We don't differentiate between simple interpretations and complete definitions in extracting process. We obtain interpretation sentences from LWP whereabouts to next sentence, for most interpretations of LWP are covered to this extent.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "We find that the two types interpretation, translation with bilingual synonymous words (we call them bracket interpretations) and \"\u662f\" structure interpretation, take about 90 percent by our investigation, so this time we just extract the two types interpretation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The paraphrases patterns of letter-word phrases in live texts", |
| "sec_num": "3" |
| }, |
| { |
| "text": "In Chinese texts bilingual LWPs can be classified as follows:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Interpretations of translation with bilingual synonymous words", |
| "sec_num": "4" |
| }, |
| { |
| "text": "A\uff1acomplete About this type interpretation, we designed a module to process. In order to extract this type interpretation, we must consider three didymous punctuations, such as \"\u300a\u8865\u8d34\u4e0e\u53cd\u8865\u8d34\u63aa\u65bd\u534f\u5b9a\u300b(\"\u300aSCM \u534f\u5b9a\u300b\")\", it contains \u4e66\u540d\u53f7\uff0c marks and brackets. We design a code system firstly, for codes are simple and contain more information than texts, and then an algorithm was put forward based the code system.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Interpretations of translation with bilingual synonymous words", |
| "sec_num": "4" |
| }, |
| { |
| "text": "a) The code system this system memorizes the state about three didymous punctuations used in a bilingual LWP with six codes. Thereinto, the former twain codes denote brackets, the middle twain codes denote quotation marks, the latter twain codes denote \u4e66\u540d\u53f7. For every twain codes, the first one denotes whether one didymous punctuation appeared in a LWP, it has three meaning, such as \"1 \" means the left punctuation appeared, \"2\" means the right punctuation appeared, and \"3\" the left or the right punctuation don't appeared at first position of the LWP. The second one denotes position of one punctuation, it has three forms and four states, \"1\" the left punctuation appeared at first position of the LWP, \"2\" means the right punctuation appeared at first position of the LWP, \"0\" means one punctuation don't appeared or had appeared in couples. For example, \"113131\" can denotes a left bracket appeared at left first position, and a left quotation marks and a left \" \u300a\" appeared at other positions, such as \" (\" \u300aSCM\". \"103030\" mean three didymous punctuations appeared in couples, such as \"(\"GATT1994\")\".", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Interpretations of translation with bilingual synonymous words", |
| "sec_num": "4" |
| }, |
| { |
| "text": "The first step, we need to scan a letter string in one text, and transfer the function iskuohao()to encode the letter string with our code system. The second step, according to the codes, judge whether any punctuations had not appeared in couples, and if there are such punctuations, then transfer the function iskq()to scan two sides of the letter string to obtain the other one of these punctuations. The operation order is brackets, then quotation marks, and then \" \u300a,\u300b \". This step generally circulate two times. The third step, examine the string obtained from above step, if there are any quotation marks, then examine if there are any irregular punctuations within the quotation marks, if it is true, then take out the quotation marks part. If all punctuations in the string are in couples, then transfer the module, which obtains the collocation Chinese characters, and then record the bilingual LWP.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "b) The process flow", |
| "sec_num": null |
| }, |
| { |
| "text": "We have 712 items bilingual LWPs from the People's Daily in 2002.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "b) The process flow", |
| "sec_num": null |
| }, |
| { |
| "text": "Here, we will chose \"\u662f\" structure to illustrate our extracting experiment, the others we'll discuss in future work.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "b) The process flow", |
| "sec_num": null |
| }, |
| { |
| "text": "We conclude the paraphrases patterns from 400 examples of LWP paraphrases, such as: a) \"LWP \u662f\u2026\u2026\u7684\u2026\u2026\u7f29\u5199\" b) \"LWP \u662f\u2026\u2026\u7684\u2026\u2026\u7f29\u5199\uff0c\u2026\u2026\uff0c\u662f\u2026\u2026\" c) \"LWP \u662f\u2026\u2026\u7684\u2026\u2026\u7f29\u5199\uff0c\u2026\u2026\uff0c\u610f\u5373\u2026\u2026\" Such as: CMM\u662f\u8f6f\u4ef6\"\u80fd\u529b\u6210\u719f\u5ea6\u6a21\u578b\"\u7684\u7f29\u5199\uff0c\u662f\u4e00\u79cd\u7528\u4e8e\u8bc4\u4ef7\u8f6f\u4ef6\u627f\u5305\u80fd\u529b\u5e76\u5e2e\u52a9\u5176\u6539\u5584\u8f6f\u4ef6 \u8d28\u91cf\u7684\u65b9\u6cd5\uff0c\u662f\u8bc4\u4f30\u8f6f\u4ef6\u80fd\u529b\u4e0e\u6210\u719f\u5ea6\u7684\u4e00\u5957\u6807\u51c6\uff0c\u4fa7\u91cd\u4e8e\u8f6f\u4ef6\u5f00\u53d1\u8fc7\u7a0b\u7684\u7ba1\u7406\u53ca\u5de5\u7a0b \u80fd\u529b\u7684\u63d0\u9ad8\u4e0e\u8bc4\u4f30\u3002 d) LWP \u662f\u2026\u2026\u7684\u2026\u2026\u7b80\u79f0 e) LWP \u662f\u2026\u2026\u7684\u2026\u2026\u7b80\u79f0\uff0c\u2026\u2026\uff0c\u662f\u2026\u2026 f) LWP \u662f\u2026\u2026\u7684\u2026\u2026\u7b80\u79f0\uff0c\u2026\u2026\uff0c\u610f\u5373\u2026\u2026 Such as: QFII\u662fQualified Foreign Institutional Investors(\u5408\u683c\u7684\u5883\u5916\u673a\u6784\u6295\u8d44\u8005)\u7684\u7b80\u79f0\uff0c QFII\u673a\u5236\u662f\u6307\u5916\u56fd\u4e13\u4e1a\u6295\u8d44\u673a\u6784\u5230\u5883\u5185\u6295\u8d44\u7684\u8d44\u683c\u8ba4\u5b9a\u5236\u5ea6\u3002 \u4f5c\u4e3a\u4e00\u79cd\u8fc7\u6e21\u6027\u5236\u5ea6\u5b89\u6392\uff0c QFII\u5236\u5ea6\u662f\u5728\u8d44\u672c\u9879\u76ee\u5c1a\u672a\u5b8c\u5168\u5f00\u653e\u7684\u56fd\u5bb6\u548c\u5730\u533a\uff0c\u5b9e\u73b0\u6709\u5e8f\u3001\u7a33\u59a5\u5f00\u653e\u8bc1\u5238\u5e02\u573a\u7684\u7279", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the paraphrases patterns of \"\u662f\" structure", |
| "sec_num": "5" |
| }, |
| { |
| "text": "\u6b8a\u901a\u9053\u3002\u5305\u62ec\u97e9\u56fd\u3001\u53f0\u6e7e\u3001\u5370\u5ea6\u548c\u5df4\u897f\u7b49\u5e02\u573a\u7684\u7ecf\u9a8c\u8868\u660e\uff0c\u5728\u8d27\u5e01\u672a\u81ea\u7531\u5151\u6362\u65f6\uff0cQFII \u4e0d\u5931\u4e3a\u4e00\u79cd\u901a\u8fc7\u8d44\u672c\u5e02\u573a\u7a33\u5065\u5f15\u8fdb\u5916\u8d44\u7684\u65b9\u5f0f\u3002 g) LWP \u662f\u2026\u2026 h) LWP \u662f\u2026\u2026\u7684\u2026\u2026 Such as:\u80af\u5c3c\u2022G\u662f\u5f53\u4ee3\u5e7f\u53d7\u6b22\u8fce\u7684\u6f14\u594f\u5bb6\u548c\u97f3\u4e50\u5bb6\uff0c\u66fe\u83b7\u5f97\u683c\u83b1\u7f8e\u5956\u3001\u7f8e\u56fd\u97f3\u4e50\u5956\u7b49\u5956 \u9879\u3002\u4ed6\u7684\u300a\u56de\u5bb6\u300b\u65e9\u4e3a\u4e2d\u56fd\u542c\u4f17\u6240\u719f\u6089\u3002 i) LWP\uff0c\u662f\u2026\u2026\u7684\u2026\u2026 \u7b2c\u4e8c\u9636\u6bb5\u524a\u51cf\u6218\u7565\u6b66\u5668\u6761\u7ea6(START\u2161)\uff0c\u662f1993\u5e741\u6708\u7531\u53f6\u5229\u94a6\u548c\u8001\u5e03\u4ec0\u4e24\u4f4d\u603b\u7edf\u7b7e\u7f72 \u7684\u3002\u6761\u7ea6\u89c4\u5b9a10\u5e74\u5185\u5404\u81ea\u524a\u51cf2/3\u7684\u6838\u5f39\u5934\uff0c\u5e76\u5c06\u53ef\u643a\u5e26\u591a\u5f39\u5934\u7684\u9646\u57fa\u6d32\u9645\u5f39\u9053\u5bfc\u5f39\u5168\u90e8 \u9500\u6bc1\u3002 j) LWP \u662f\u6307\u2026\u2026 QFII\u673a\u5236\u662f\u6307\u5916\u56fd\u4e13\u4e1a\u6295\u8d44\u673a\u6784\u5230\u5883\u5185\u6295\u8d44\u7684\u8d44\u683c\u8ba4\u5b9a\u5236\u5ea6\u3002 k) LWP \u662f\u6307\u2026\u2026\u7684\u2026\u2026 GDP\u662f\u6307\u4e00\u4e2a\u56fd\u5bb6(\u6216\u5730\u533a)\u5728\u4e00\u5b9a\u65f6\u671f\u5185\u6240\u6709\u5e38\u4f4f\u5355\u4f4d\u751f\u4ea7\u7ecf\u8425\u6d3b\u52a8\u7684\u5168\u90e8\u6700\u7ec8\u6210\u679c\u3002 GDP\u662f\u6309\u56fd\u571f\u539f\u5219\u6838\u7b97\u7684\u751f\u4ea7\u7ecf\u8425\u7684\u6700\u7ec8\u6210\u679c\u3002\u6bd4\u65b9\u8bf4\uff0c\u5916\u8d44\u4f01\u4e1a\u5728\u4e2d\u56fd\u5883\u5185\u521b\u9020\u7684\u589e\u52a0 \u503c\u5c31\u5e94\u8be5\u8ba1\u7b97\u5728GDP\u4e2d\u3002 l) LWP\uff0c\u662f\u6307\u2026\u2026\u7684\u2026\u2026 m) LWP\uff0c\u6307\u7684\u662f\u2026\u2026 \u800c\u4eba\u4eec\u901a\u5e38\u6240\u8bf4\u76843G\uff0c\u6307\u7684\u662f\u4e0b\u4e00\u4ee3\u591a\u5a92\u4f53\u79fb\u52a8\u901a\u4fe1\u7cfb\u7edf\uff0c\u5b83\u5177\u6709\u66f4\u5bbd\u7684\u5e26\u5bbd\uff0c\u66f4\u9ad8\u7684 \u9891\u7387\u548c\u4f20\u8f93\u901f\u7387\uff0c\u53ef\u4ee5\u65b9\u4fbf\u5730\u8fdb\u884c\u56fe\u50cf\u751a\u81f3\u6d3b\u52a8\u753b\u9762\u7684\u4f20\u8f93\u3002 n) LWP \u662f\u4e00\u79cd\u2026\u2026 o) LWP\uff0c\u662f\u4e00\u79cd\u2026\u2026\u7684\u2026\u2026 FAD2\u662f\u4e00\u79cd\u5c06\u9971\u548c\u8102\u80aa\u9178\u8f6c\u5316\u6210\u4e0d\u9971\u548c\u8102\u80aa\u9178\u7684\u9176\uff0c\u5bf9\u54fa\u4e73\u52a8\u7269\u751f\u957f\u5177\u6709\u91cd\u8981\u4f5c\u7528\uff0c\u5728 \u83e0\u83dc\u7b49\u690d\u7269\u4e2d\u5b58\u5728\u3002 p) LWP\uff0c\u7b80\u79f0 LWP\uff0c\u662f\u2026\u2026 q) \u7b80\u79f0 LWP\uff0c\u662f\u4e00\u79cd\u2026\u2026 r) \u2026\u2026\u7b80\u79f0 LWP\uff0c\u5b83\u662f\u2026\u2026", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the paraphrases patterns of \"\u662f\" structure", |
| "sec_num": "5" |
| }, |
| { |
| "text": "\u5c0f\u7075\u901a\u53c8\u79f0\u65e0\u7ebf\u5e02\u8bdd(Personal Phone System)\uff0c\u7b80\u79f0PHS\uff0c\u662f\u4e00\u79cd\u4e2a\u4eba\u65e0\u7ebf\u63a5\u5165\u7cfb\u7edf\u3002 \u5b83\u91c7\u7528\u5fae\u8702\u7a9d\u6280\u672f\uff0c\u901a\u8fc7\u5fae\u8702\u7a9d\u57fa\u7ad9\u5b9e\u73b0\u65e0\u7ebf\u8986\u76d6\uff0c\u5c06\u7528\u6237\u7aef(\u5373\u65e0\u7ebf\u5e02\u8bdd\u624b\u673a)\u4ee5\u65e0 \u7ebf\u7684\u65b9\u5f0f\u63a5\u5165\u672c\u5730\u7535\u8bdd\u7f51\uff0c\u4f7f\u4f20\u7edf\u610f\u4e49\u4e0a\u7684\u56fa\u5b9a\u7535\u8bdd\u4e0d\u518d\u56fa\u5b9a\u5728\u67d0\u4e2a\u4f4d\u7f6e\uff0c\u53ef\u5728\u65e0\u7ebf\u7f51 \u7edc\u8986\u76d6\u8303\u56f4\u5185\u79fb\u52a8\u4f7f\u7528\uff0c\u968f\u65f6\u968f\u5730\u63a5\u542c\u3001\u62e8\u6253\u672c\u5730\u548c\u56fd\u5185\u3001\u56fd\u9645\u7535\u8bdd\u3002 s) \u2026\u2026\u662f LWP\u2026\u2026\u5c31\u662f\u2026\u2026 t) \u2026\u2026\u662f LWP\u2026\u2026 u) \u2026\u2026\u5c31\u662f\u2026\u2026LWP\u2026\u2026 \"\u2026\u2026\" are the other characters in paraphrases sentences excluding the label words, when we regard punctuations and the other characters as coordinate, the above patterns could be reduce to fifteen patterns. We find that \"\u662f\u6307\" and \"\u662f\u2026\u2026\"arranged in pairs or groups with \"\u7b80\u79f0\", \"\u7f29\u5199\" always a paraphrase sentence.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the paraphrases patterns of \"\u662f\" structure", |
| "sec_num": "5" |
| }, |
| { |
| "text": "When we could recognize a LWP, then we could extract a paraphrase sentence, so we make use of our auto-labeling software [see also ZHENG Zezhi, 2005 ] to tag the LWP firstly, and then extract LWP paraphrases. ", |
| "cite_spans": [ |
| { |
| "start": 131, |
| "end": 148, |
| "text": "ZHENG Zezhi, 2005", |
| "ref_id": "BIBREF0" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The auto-extracting algorithm", |
| "sec_num": "6" |
| }, |
| { |
| "text": "Our experiment is based on 500 text files, which are from the corpus of the \"People's Daily\" in 2002 and contains LWPs. From the 500 text files, we got 59 paraphrase pieces by manual work, in which 22 pieces are of \"\u662f\"structure paraphrases, 18 pieces are with mark words, others are without marker phrases and not of \"\u662f\"structure paraphrases.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "With the algorithm, we do an experiment with \"\u662f\"structure paraphrases. From the experiment result, we got 29 pieces are of \"\u662f\"structure paraphrases, in which 19 pieces are Useful results and 10 pieces are not., so the usefulness rate is 19/29=65.5%, and recall rate is 19/22=86.4%.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "The formula of usefulness rate is:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": ") d ( extdef ) d ( cordef p =", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "The cordef(w) means the number of useful paraphrase of extracted results. The extdef(w) means the number of extracted results.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "The formula of recall rate is:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": ") d ( orgdef ) d ( cordef r =", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "The cordef(w) means the number of useful paraphrase of extracted results. The orgdef(w) means the number of useful paraphrase of \"\u662f\"structure paraphrases from 500 text files.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "An example of useful results: <zm>P38 \u901a\u8def</zm>\u662f\u5c11\u6709\u7684\u51e0\u4e2a\u5728\u8fdb\u5316\u4e0a\u975e\u5e38\u4fdd\u5b88\u7684\u751f\u7269\u4fe1\u606f\u4f20\u5bfc\u673a\u5236\u3002\u6b64\u901a\u8def\u7684\u6fc0\u6d3b \u4e0e\u5931\u6d3b\u548c\u4f11\u514b\u3001\u5173\u8282\u708e\u3001\u52a8\u8109\u7ca5\u6837\u786c\u5316\u7b49\u6025\u6162\u6027\u514d\u75ab\u75be\u75c5\u6709\u91cd\u5927\u5173\u7cfb\u3002 An example of no-paraphrase results:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "<zm>CDMA</zm>\u662f\u4e2a\u597d\u6280\u672f\uff0c\u5173\u952e\u8fd8\u8981\u770b\u8054\u901a\u7ecf\u8425\u5f97\u600e\u4e48\u6837\uff0c\u8fd9\u662f\u4e0d\u5c11\u4e1a\u5185\u4eba\u58eb\u7684\u770b\u6cd5\u3002 In fact, no-paraphrase results are these paraphrases which can't provide enough information to interpret or to make readers understand the LWP's meaning.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "As a result, we find that these patterns always gave useful results, such as:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "ELWP\u662f\u2026\u2026\u7684\u2026\u2026\u7f29\u5199/\u7b80\u79f0\uff1b ELWP\u662f\u2026\u2026\u7684\u2026\u2026\u7f29\u5199/\u7b80\u79f0\uff0c\u2026\u2026\uff0c\u662f\u2026\u2026\uff1b ELWP\u662f\u2026\u2026\u7684\u2026\u2026\u7f29\u5199/\u7b80\u79f0\uff0c\u2026\u2026\uff0c\u610f\u5373\u2026\u2026\uff1b \u7b80\u79f0ELWP\uff0c\u662f\u4e00\u79cd\u2026\u2026 And the no-paraphrase results were from these patterns, such as: ELWP\u662f\u2026\u2026 ELWP\u662f\u2026\u2026\u7684\u2026\u2026 If we provide category names, like \"\u7cfb\u7edf|\u6807\u51c6|\u7ec4\u7ec7|\u4f01\u4e1a|\u516c\u53f8\u2026\u2026\", i.e. add category names in the \"\u662f\" structure patterns, the number of useful results will be increased, for this measure would get rid of some no-paraphrase results. Also if we add no-paraphrase definitive, like \"\u826f\u673a|\u673a\u9047\u2026\u2026\", some no-paraphrase results would be eliminated.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "the experiment result", |
| "sec_num": "7" |
| }, |
| { |
| "text": "By now, our training set only had correct instances, so our farther work is to tidy up the no-paraphrase results from our experiment to form our exceptional rules, and collect category names and no-paraphrase definitive, in order to increase the precision rate of our algorithm.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Farther work", |
| "sec_num": "8" |
| }, |
| { |
| "text": "We have finished extracting \"\u662f\" structure paraphrases, but is not all-sided, we must try other no-\"\u662f\" paraphrases extracting. And we should study the relation between our patterns and the normative patterns of dictionaries, ascertain the power coefficient, and then rank the same LWP's paraphrases.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Farther work", |
| "sec_num": "8" |
| }, |
| { |
| "text": "Sustentation fund: science fund (code: 60475022), and XIAMEN University start-up fund.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "The Research on Lettered-word Extraction in Chinese Texts, Chinese Information Transaction", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Zheng Zezhi", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Zhangpu", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "", |
| "volume": "19", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "ZHENG Zezhi, ZHANGPu, et al, The Research on Lettered-word Extraction in Chinese Texts, Chinese Information Transaction, 2005.1, vol. 19.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Ten Relation Groups in Dictionaries, Dictionary Study, 2005.1, SHANGHAI dictionary publishing company", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Li Xiyin", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "LI Xiyin, Ten Relation Groups in Dictionaries, Dictionary Study, 2005.1, SHANGHAI dictionary publishing company.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Analysis and depiction of acceptations, BEIJING, Chinese publishing company", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Fu Huaiqing", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "FU Huaiqing, Analysis and depiction of acceptations, BEIJING, Chinese publishing company, 1996.1", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Structure Analysis and Extraction for the Definitions of Chinese terms, Chinese Information Transaction", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Zhang Yan", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Chengqing", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "", |
| "volume": "6", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "ZHANG Yan, ZONG Chengqing, et al, Structure Analysis and Extraction for the Definitions of Chinese terms, Chinese Information Transaction, 2003. 6,vol. 17.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "A Web Term Definition Extracting System, Chinese Information Transaction", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Xu Yong", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Endong", |
| "suffix": "" |
| } |
| ], |
| "year": 2004, |
| "venue": "", |
| "volume": "4", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "XU Yong, XUN Endong, et al, A Web Term Definition Extracting System, Chinese Information Transaction, 2004. 4, vol. 18.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "the Compendium of Russian Terminological Thesaurus theory, Dictionary Study, 2005.1, SHANGHAI dictionary publishing company", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Zheng", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Shupu", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "ZHENG Shupu, the Compendium of Russian Terminological Thesaurus theory, Dictionary Study, 2005.1, SHANGHAI dictionary publishing company.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "text": "Tag the texts choused with auto-labeling software; 2) Express the sentences and patterns with vectors; 3) Extract LWPs and their paraphrases; 4) Estimate the results from auto-extracting manually and conclude erratum rules; 5) Go back 2), continue training.The auto-extracting algorithm of LWP paraphrases 1) Express the paraphrase patterns with the mark vector patterns, i T r =(w, b 1 , b 2 , b 3 , b 4 ),(i=1, 2, 3,\u2026,15), thereinto, w is a LWP, b j (j=1, 2, 3 , 4) are mark words, b 1 can't be a null, b 2 , b 3 , b 4 could be a null. Found a vector space with the fifteen mark vector patterns; 2) Extract the sentences with a LWP or more;3) Express these sentences extracted at 2) step with mark vector patterns, S r Calculate the distance between the two vectors: then the current sentence is paraphrase sentence.5)Take out the paraphrase of current LWP. Scan back forward to the head of current sentence, and scan along to next sentence end, and then take the two sentences as the paraphrase.", |
| "uris": null, |
| "type_str": "figure" |
| } |
| } |
| } |
| } |
