| { | |
| "paper_id": "O09-2001", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T08:10:48.726988Z" | |
| }, | |
| "title": "Automatic Term Pair Extraction from Bilingual Patent Corpus", | |
| "authors": [ | |
| { | |
| "first": "Yuen-Hsien", | |
| "middle": [], | |
| "last": "Tseng", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "samtseng@ntnu.edu.tw" | |
| }, | |
| { | |
| "first": "Chao-Lin", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "This paper proposes two approaches to extract translation term pairs from Chinese-English bilingual corpus with more than 500,000 patents. One approach is precision-oriented, in which we compare six term alignment methods. Based on our experiments, we find that the EM (Expectation Maximization) method is the best. However, it is time-consuming and hard to extract many-to-many translations for the same concept. While the MI (mutual information) method performs worst, the term pairs extracted may be totally different from those by EM. This may inspire subsequent researches to study the possibility of hybrid term alignment methods. The other approach is recall-oriented, in which a simple idea was proposed. With an efficient implementation, 20% more term pairs were extracted based on an existing lingual lexicon which already has more than one million term pairs merged from several sources.", | |
| "pdf_parse": { | |
| "paper_id": "O09-2001", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "This paper proposes two approaches to extract translation term pairs from Chinese-English bilingual corpus with more than 500,000 patents. One approach is precision-oriented, in which we compare six term alignment methods. Based on our experiments, we find that the EM (Expectation Maximization) method is the best. However, it is time-consuming and hard to extract many-to-many translations for the same concept. While the MI (mutual information) method performs worst, the term pairs extracted may be totally different from those by EM. This may inspire subsequent researches to study the possibility of hybrid term alignment methods. The other approach is recall-oriented, in which a simple idea was proposed. With an efficient implementation, 20% more term pairs were extracted based on an existing lingual lexicon which already has more than one million term pairs merged from several sources.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Foundations of Statistical Natural Language Processing", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Christopher", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hinrich", | |
| "middle": [], | |
| "last": "Manning", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Schutze", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2001, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Christopher D. Manning and Hinrich Schutze, Foundations of Statistical Natural Language Processing, The MIT Press, 2001.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Multilingual Keyword Extraction for Term Suggestion", | |
| "authors": [ | |
| { | |
| "first": "Y.-H", | |
| "middle": [], | |
| "last": "Tseng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "21st International ACM SIGIR Conference on Research and Development in Information Retrieval -SIGIR '98", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tseng, Y.-H. Multilingual Keyword Extraction for Term Suggestion. in 21st International ACM SIGIR Conference on Research and Development in Information Retrieval - SIGIR '98. 1998. Australia.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Automatic Thesaurus Generation for Chinese Documents", | |
| "authors": [ | |
| { | |
| "first": "Y.-H", | |
| "middle": [], | |
| "last": "Tseng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Journal of the American Society for Information Science and Technology", | |
| "volume": "53", | |
| "issue": "13", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tseng, Y.-H., Automatic Thesaurus Generation for Chinese Documents. Journal of the American Society for Information Science and Technology, 2002. 53(13): p. 9.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Text Mining Techniques for Patent Analysis. Information Processing and Management", | |
| "authors": [ | |
| { | |
| "first": "Y.-H", | |
| "middle": [], | |
| "last": "Tseng", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C.-J", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y.-I", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "", | |
| "volume": "43", | |
| "issue": "", | |
| "pages": "1216--1247", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tseng, Y.-H., C.-J. Lin, and Y.-I. Lin, Text Mining Techniques for Patent Analysis. Information Processing and Management, 2007. 43(5): p. 1216-1247.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Automatic Cataloguing and Searching for Retrospective Data by Use of OCR Text", | |
| "authors": [ | |
| { | |
| "first": "Y.-H", | |
| "middle": [], | |
| "last": "Tseng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2001, | |
| "venue": "Journal of the American Society for Information Science and Technology", | |
| "volume": "52", | |
| "issue": "5", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tseng, Y.-H., Automatic Cataloguing and Searching for Retrospective Data by Use of OCR Text. Journal of the American Society for Information Science and Technology, 2001. 52(5): p. 12.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Content-Based Retrieval for Music Collections", | |
| "authors": [ | |
| { | |
| "first": "Y.-H", | |
| "middle": [], | |
| "last": "Tseng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "Proceedings of the 22nd International ACM SIGIR Conference on Research and Development in Information Retrieval -SIGIR '99", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "176--182", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tseng, Y.-H., \"Content-Based Retrieval for Music Collections,\" Proceedings of the 22nd International ACM SIGIR Conference on Research and Development in Information Retrieval -SIGIR '99, Aug. 15-19, Berkeley, U.S.A., 1999, pp.176-182.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Search Engine: Information Retrieval in Practice", | |
| "authors": [ | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bruce", | |
| "middle": [], | |
| "last": "Croft", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Donald", | |
| "middle": [], | |
| "last": "Emtzler", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Trevor", | |
| "middle": [], | |
| "last": "Strohman", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "W. Bruce Croft, Donald Emtzler, Trevor Strohman, Search Engine: Information Retrieval in Practice, Addison-Wesley, 2009.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF0": { | |
| "num": null, | |
| "content": "<table><tr><td>4. Dice</td><td colspan=\"2\">p(c,e) DC, Dice coefficient 506510</td><td>c</td><td>EM</td><td>OCR e</td><td>p(x) liquid crystal display device n=50 25963 n</td></tr><tr><td colspan=\"7\">IPO, Intellectual Property Office WIPO EPO UPSTO TIPO Method for producing rutin Chinese medicine composition PO, Patent Office JPO [8] MIDI [9] maximum symmetric SIPO x likelihood estimation KPO MI MI(c,e)=MI(e,c) MI [4] e4= driving circuit 4 4 4/8 2 0.7273 2.9090 0.8649 3.4595 0.9343 display device 88.2 1 0 0.1347 6.39 0.23 -1 118 163 1589 [10 e3=lamp driving circuit 1 1 1/8 0.125 0.0455 0.0455 0.0135 0.0135 0.0036 0 MI semiconductor device 3 0.99 0 0.0017 4.75 0.02 -1 3 6 3429 MI Loop 1 Loop 2 Loop 3 Loop 4 liquid crystal display liquid crystal 50 c e FC p(e|c) p(c|e) DC MI CC LR f 11 Fc Fe 160 Dice DC(c,e) =2f 11 /(Fc+Fe)=2f 11 /(2f 11 +f 12 +f 21 ) c= f 11 s(c,e) p(e|c) s(c,e) p(e|c) s(c,e) p(e|c) s(c,e) p(e|c) e1= display devices 2 2 2/8 0.5 0.1818 0.3636 0.1081 0.2162 0.0584 e2= electroluminescent lamp 1 1 1/8 0.125 0.0455 0.0455 0.0135 0.0135 0.0036 p(e|c) 1 0 EM {EM>LR>FC} > {DC=CC} >> MI EM LR FC DC CC semiconductor assembly 1.5 0 0.56 0.0014 5.06 0.02 5.66 2 2764 4 semiconductor equipment 0.5 0 170 O(NM 2 ) 0.0 0.0007 3.06 0.01 1.29 1 2764 8 semiconductor arrangement 0.5 0 0.0 0.0007 3.06 0.01 1.29 1 2764 8 1445000 50 170 170</td></tr><tr><td colspan=\"7\">[1] Manning Schutze aligned term pairs translation model 87 33 3 language model 30 506510 123280 method, producing, rutin, Chinese medicine, composition combinational 40% 490 rutin, Chinese medicine, composition, producing, method Run 2: Co( , rutin) Co( , Chinese medicine) Co( , Chinese medicine) Co( , rutin) , rutin) , Chinese medicine) [2] , composition) , rutin) , composition) , Chinese medicine) , Chinese medicine) explosion [3] [4] Run 1: Co( Co( Co( Co( Co( Co( Co( Co( , rutin) Co( , composition) Co( , Chinese medicine) Co( , composition) Co( , rutin) Co( , composition) Co( , Chinese medicine) Co( , rutin) Co( , composition) Co( , rutin) Co( , Chinese medicine) Run 3: Co( , rutin) greedy 3 123280 method for making yarns and constitution structure thereof , , , , , making yarns constitution structure :4, :3, :2, :2, :2 making yarns:2, winding:4, yarn:3, weaving:2, constitution structure:2 , MI [10] 2. CC, correlation coefficient Chi-square [4] * FeFe * FcFc ) f f -f f ( ) , ( 2 21 12 22 11 2 u u e c F f 11 f 21 f 12 f 22 c e c e c e Fc Fc* Fe Fe* N c e c c e f 11 f 12 Fe 5. FC, fractional count LR log(0) -1 LR 10 display device 0.5 0.5 0 0.0013 6.86 0.03 -1 1 1 1589 n FC EM DC MI CC LR display device 0.5 1 0 0.0013 6.86 0.03 -1 1 1 1589 liquid crystal display device 772 0.98 1 0.6409 6.21 0.66 -1 1078 2072 1292 1. Set i = 1 to M M ( ) inter-sentence intra-sentence 1/6 rutin 2/6 FC ) ( ) , ( ) , ( i sp e c p e c FC 506510 LR 10 c e FC p(e|c) p(c|e) DC MI CC LR f 11 Fc Fe 50 50 50 50 50 50 3 0 6 39 6 1 lcd device 21.9 0 1 0.0265 5.82 0.09 97.1 28 2072 44 lcd apparatus 9.3 0 2. Set j = 1 to M 2971 1 0.0105 5.85 0.06 38.4 11 2072 17 lcd equipment 0.5 0 0.5 0.001 6.47 0.02 4.49 1 2072 1 3. Set c ij = c i c j c i i 25920 463MB Pentium 4, 3.20GHz, 2GB RAM 5945 epoxy resin composition 44.62 0.97 1.00 0.87 10.47 0.87 757.59 99 114 113 illumination system 67.64 1.00 1.00 0.82 10.32 0.83 731.76 98 133 106 record carrier n=50 electric connector 31.1 0 1 0.0879 6.95 0.18 212 49 1045 70 cable connector 2 0 4. Set e ij = e i e j e i i 1560 0 0.0037 3.18 0.01 2.67 2 1045 39 5. Set k = 1 to N N 50.93 1.00 1.00 0.87 10.58 0.87 703.34 91 105 104 photosensitive resin composition 65.58 1.00 1.00 0.72 9.86 0.72 697.28 102 137 148 data processing system 52.29 1.00 1.00 0.87 10.65 0.87 675.94 87 100 100 FC EM DC MI CC LR FC electrical connector 81.9 1 0 0.1803 7.11 0.28 -1 105 136 1029 electrical connector 2 0.97 0 0.0077 7.16 0.06 -1 4 6. c ij CP k e ij EP k (c ij , e ij ) 5 1029 electrical connector 1 1 0 0.0019 7.48 0.03 -1 1 1 1029 CP k EP k k \u00a6 ) ( ) , .( . , _ _ i sp e c t s i all for sp(i) i c e p sp(i) (c,e) i (c,e) FC 6. EM Expectation-Maximization analysis penalization of Number of bilingual texts processed (463MB) heat exchanger 60.58 1.00 1.00 0.81 10.40 0.81 662.40 89 119 102 base station 38.96 1.00 1.00 0.80 10.42 0.80 638.43 86 111 104 electrical connector 0.3 0.5 0 0.0019 7.48 0.03 -1 1 1 1029 EM 9 (18%) 506510 Number of empty Chinese Title information storage medium 48.62 1.00 1.00 0.83 10.59 0.83 633.72 83 103 96 DC 0 0 0 Number of empty Chinese Abstracts 281 Number of empty English Title semiconductor chip 55.06 1.00 1.00 0.59 9.33 0.59 630.94 101 186 155 silicon wafer 51.56 1.00 1.00 0.80 10.55 0.81 602.65 80 106 93 MI 0 0 0 CC 0 0 semiconductor 1. I assembly 45 (90%) 0 0 Number of empty English Abstracts 166532 N LR 2 (4%) 10 (20%) 0 0 0 2. A N Z39 50 is protocol for 3. Set i = 1 to M M making yarns, constitution structure e f 21 f 22 Fe* Fc Fc* N Chi-square test for dependence 0 1 implausible alignment Run 2 Run 3 EM Number of Doc. with both abstracts empty 2 Number of bilingual texts yielding empty term pair 322309 Number of term pairs generated 1180281 Number of unique term pairs after merging 455696 N EM 1 1 8 5 94 f 11 DC CC 4. (c i , e i ) I (c i , e i ) S = {s 1 ,\u2026,s P } {EM>LR>FC} > 160 455696 5. Set j = s 1 to s P {DC=CC} >> MI FC 7050 6. push i to A[j] A[j] j i LR EM 25963 25963 7. Set p = 1 to N N LR FC MI</td></tr><tr><td>8.</td><td>Set x =</td><td>A[p]</td><td/><td/><td/></tr><tr><td colspan=\"7\">bootstrap term pairs [ ] \u00d9 yellow wind precision-oriented recall-oriented 160 \u00d9 Drying heat ( ) \u00d9nematic + \u00d9 dragonet + \u00d9 cymenes rutin maximal implausible 1. MI mutual information MI [4] log ) , ( e c MI FcFc 77 Tseng \u00d9liquid crystal; thermotropic 170 \u00d9liquid crystal; display; LCD \u00d9LCD (=liquid crystal display) translation \u00d9liquid crystal display (LCD) \u00d9liquid-crystal display {= LCD} 110 [5-7] ) ( ) ( ) , ( 2 e p c p e c p * FeFe * ) f f -f f ( ) , ( 21 12 22 11 u u e c CC -1 +1 CC co-occurrence Chi-square CC 3. LR, likelihood ratios Chi-square CC 20 [4] LR [4] )) ( *, , ( log )) ( , , ( log *)) | ( *, , ( log )) | ( , , ( log ) , ( 12 11 12 11 e p Fc f L e p Fc f L c e p Fc f L c e p Fc f L e c LR L(k,n,x)=x k (1-x) n-k , p(e)=Fe/N, p(e|c)=f 11 /Fc, p(e|c*)=f 12 /Fc* LR E 455696 1340 Intel 9. Set y = A[p] M E-step c e p(e|c) 11 ) ( ) , .( . , _ _ * ) | ( ) | ( ) , ( f c e p c e p e c s i sp e c t s i all for \u00a6 f 11 p(e|c) 1 FC(c,e) M-step \u00a6 v v c s e c s c e p ) , ( ) , ( ) | ( v s(c,v) 0 c= f 11 p(c|e) 1 EM CPU T2500 2.0 GHz, 2 GB RAM 15 EM p(c|e) p(e|c) 1296 29 6 (p(c|e)+p(e|c))/2=1.0 18322 (p(c|e)+p(e|c))/2=1.0 and f 11 >=5 94 c e 10. Set c xy = c x c y c x x FC p(e|c) p(c|e) DC MI CC LR f 11 Fc Fe semiconductor device 2078.3 1 11. Set e xy = e x e y e x x 1 0.79 5.58 0.79 -1 2455 2764 3429 FC 10 FC DC=1.0 156353 DC=1.0 and f 11 >=5 51 MI=17.49= maximum semiconductor devices 28.583 0 1 0.03 4.93 0.09 130 48 2764 105 display device 867.83 1 1 0.70 6.22 0.70 -1 1225 1909 1589 display devices 8.0595 0 1 0.02 6.20 0.08 61.7 16 1909 21 c xy CP p e xy EP p (c xy , e xy ) CP p Title: Backlight modules with connector EP p p modules and conductive blocks thereof 152907 10 c e FC p(e|c) p(c|e) DC MI CC LR f 11 Fc Fe semiconductor device 2078.30 1.00 1.00 0.79 5.58 0.79 -1 2455 2764 3429 display device 867.83 1.00 1.00 0.70 6.22 0.70 -1 1225 1909 1589 liquid crystal display device 772.08 0.98 1.00 0.64 6.21 0.65 -1 1078 2072 1292 electrical connector 750.25 1.00 1.00 0.80 7.15 0.79 -1 830 1045 1029 electronic device 326.45 1.00 1.00 0.57 6.90 0.59 -1 541 1180 706 backlight module 323.25 1.00 1.00 0.74 7.70 0.74 -1 498 807 546 liquid crystal display 292.15 0.01 1.00 0.27 5.03 0.27 -1 451 2072 1227 semiconductor memory device 237.50 1.00 0.98 0.57 7.74 0.59 -1 302 394 661 * liquid crystal 218.15 0.01 1.00 0.27 5.10 0.27 -1 431 2072 1119 thin film transistor 197.08 1.00 1.00 0.71 8.43 0.72 -1 280 467 320 MI=17.49 and f 11 >=5 electrical connector 750.25 1 1 0.80 7.15 0.80 -1 830 1045 1029 Abstract: Backlight modules for large size flat 33 CC=1.0 156353 CC=1.0 and f 11 >=5 51 LR>186.13 1 249 LR>186.13 and f 11 >=5 249 FC>311.01 2 6 FC>311.01 and f 11 >=5 electrical connector 4 1 0 0.01 7.00 0.06 -1 5 7 1029 electrical connector 1 1 0 0.00 7.48 0.03 -1 1 panel displays are provided. A backlight 1 1029 electronic device 326.5 1 module comprises a plate, a plurality of first 1 0.57 6.90 0.59 -1 541 1180 706 electron device 2.667 0 1 0.01 6.80 0.05 21.1 5 1180 7 thin film transistor 197 1 1 0.71 8.43 0.72 -1 280 467 320 thin-film transistor 28 0 1 0.15 8.39 0.27 215 39 467 46 recording medium 162.4 1 1 0.65 7.98 0.66 -1 315 556 414 \u2026 and second lamps, a first connector and an optical assembly. \u2026 4 200 50 4 [1] [2] 96 . 2008; Available: 6 1 3 =26.07+3*53.35 2 3 =28.20+3*94.27 recording media 15.87 0 1 0.09 7.88 0.19 135 27 556 38 record medium 7.667 0 1 0.04 8.13 0.13 58.4 11 556 13 record media 0.143 0 0.25 0.00 8.37 0.04 5.80 1 556 1 recording medium 5.333 1 0 0.04 8.10 0.11 -1 8 13 414 406184 20% =40/(170+40) http:/[3] and . . 2007; Available: c e4 * computer system 159.9 1 1 0.85 8.74 0.85 -1 313 374 361 http:/</td></tr></table>", | |
| "html": null, | |
| "type_str": "table", | |
| "text": "Patents By Country, State, and Year -Utility Patents (December 2008). 2009; Available: http://www.uspto.gov/go/taf/cst_utl.htm. [Accessed: July 10, 2009]. /www.tipo.gov.tw/ch/MultiMedia_FileDownload.ashx?guid=e24d3489-c729-4159-ab2a-eab9e6788d49.pdf. [Accessed: July 10, 2009]. /kousyoukai.japan.org.tw/tokusennchu.pdf." | |
| } | |
| } | |
| } | |
| } |