Proceedings Template - WORD - Columbia CS - Columbia University

12 downloads 0 Views 57KB Size Report
Information from Glossaries. The extractions are used to build automatically a lexical knowledge base from on-line domain specific glossary sources.
Extracting Taxonomic Relationships from On-Line Definitional Sources Using LEXING Judith Klavans

Brian Whitman

Columbia University th 540 W. 120 Street NY, NY, 10027 212-939-7117

Columbia University th 540 W. 120 Street NY, NY, 10027 212-939-7108

[email protected]

[email protected]

$%675$&7

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µOLQNLQJ SUHSRVLWLRQV¶IRU XVH LQ VNLSSLQJ SDVWHPSW\KHDGJHQXVSKUDVHV7KLVV\VWHPLVSDUWRIDSURMHFWWR H[WUDFW DQG VWUXFWXUH RQWRORJLFDO LQIRUPDWLRQ IRU HQHUJ\UHODWHG JORVVDU\LQIRUPDWLRQ

&DWHJRULHVDQG6XEMHFW'HVFULSWRUV

>5HSUHVHQWDWLRQ,VVXHV@2QWRORJLHVDQGWD[RQRPLHV

.H\ZRUGV

2QWRORJLHV JORVVDULHV GHILQLWLRQV OH[LFDO NQRZOHGJH EDVHV /.% LQIRUPDWLRQUHWULHYDOQDWXUDOODQJXDJHSURFHVVLQJ

 86,1*'(),1,7,21672%8,/' 7$;2120,(6)25',*,7$/ &2//(&7,216

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` R[\JHQDWHEOHQGLQJ 5%2% EXWH[FOXGHR[\JHQDWHV DOFRKROV HWKHUV EXWDQHDQGSHQWDQHVSOXV1RWH 2[\JHQDWHVDUHUHSRUWHG DVLQGLYLGXDOFRPSRQHQWVDQGDUHLQFOXGHGLQWKHWRWDOIRURWKHU K\GURFDUERQVK\GURJHQVDQGR[\JHQDWHV )LJXUH  6DPSOH*ORVVDU\'HILQLWLRQ

&RPSDUH WKLV UHODWLYHO\ XQVWUXFWXUHG HQWU\ ZLWK D W\SLFDO GLFWLRQDU\ GHILQLWLRQ ZKLFK KDV ULFK LQWHUQDO VWUXFWXUH HJ SURQXQFLDWLRQ HW\PRORJ\ VHQVH QXPEHUV HWF ,Q WKH GLFWLRQDU\ DQDO\VLVSURMHFWVLQ>@RYHUILIW\RIVXFKILHOGVDUHLGHQWLILHG7R DFKLHYHRXUJRDOVIRUWKH('&SURMHFWZHPXVW   

,GHQWLI\ WKH GHILQLWLRQDO PDWHULDO IURP D ZHE SDJH RU RWKHURQOLQHVRXUFH LGHQWLILFDWLRQ VHH>@ 3DUVH WKH GHILQLWLRQ IRU LWV PRVW VDOLHQW SURSHUWLHV DQG IHDWXUHV /.%JHQHUDWLRQJHQXVILQGLQJ

,QFRUSRUDWH WKH VWUXFWXUHG LQIRUPDWLRQ LQWR D ODUJHU RQWRORJ\ LQFOXGLQJ OLQNLQJ DQG PHUJLQJ GHILQLWLRQV IURPGLIIHUHQWDJHQFLHVDQGVRXUFHV

 *(1867(50$1'3+5$6(),1',1*

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¶V UHVXOWVDJDLQVWDPDQXDOO\ WDJJHG VHW RIJORVVDU\GHILQLWLRQVIURPGLIIHUHQWVRXUFHV 'HILQLWLRQÆ +HDG7HUP  'HILQLWLRQ7H[W

'HILQLWLRQ7H[WÆ *HQXV3KUDVH  5HPDLQGHU 5HPDLQGHUÆ 7H[W

*HQXV3KUDVHÆ 13 RI " *HQXV3KUDVH *HQXV7HUPÆ ODVWQRXQRIILUVW*313

)LJXUH  *HQXV3KUDVHDQG7HUP*UDPPDU

6LQFH HDFK GRPDLQ FRXOG XVH GRPDLQ VSHFLILF VHPDQWLF VHSDUDWLRQVZHDOVRLQWURGXFHWKHQRWLRQRIDXWRPDWLFDOO\GHULYHG

VHPDQWLF DWWULEXWHV WKDW DUH LQIHUUHG VLPSO\ IURP WKHLU IUHTXHQF\ LQ WKH WH[W  7KH /(;,1* V\VWHP LGHQWLILHV VHSDUDWRUV VXFK DV KDYLQJDXVHGIRURUFRQWDLQLQJDDUH³FXHSKUDVHV´WKDWLGHQWLI\ WKHQH[WFODXVHDVVXLWDEOHIRUVHPDQWLFFKXQNLQJ$VDQH[DPSOH WKH SKUDVH KDYLQJ D ZDV QRW LQ RXU RULJLQDO OLVW RI PDQXDOO\ GHULYHG VHSDUDWRUV EXW DIWHU UXQQLQJ D ELJUDP DQDO\VLV ZH GLVFRYHUHGLWVIUHTXHQF\DQGLPSRUWDQFH

 5(68/762)/(;,1*(9$/8$7,21 :HKDYHSHUIRUPHGWZRHYDOXDWLRQVRQ/(;,1*RXWSXW  

'HILQLWLRQ &RQWHQW $QDO\VLV ZH UDQ RXU V\VWHP RQ YDULRXV GHILQLWLRQDOVRXUFHVWRGHWHUPLQHLIRXULGHDVRIFRQWHQWZHUH FRUUHFWDQGZKLFKILHOGVDUHIUHTXHQW

/(;,1* $FFXUDF\ ZH HYDOXDWHG WKH JHQXV WHUP LGHQWLILFDWLRQDOJRULWKPDJDLQVWDJROGVWDQGDUG

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

(,$ (GLWHG (,$ :HE (3$ :HE 0HGLFDO $XWR

7HUPV

*HQXV 3KUDVHV

3URS HUWLHV

4XDQW LILHUV

,QFOXGHV H[FOXGH



       

       

       

       



 

7KH UHVXOWV LQ 7DEOH  VKRZ WKDW GHILQLWLRQDO FRQWHQW KLQJHV ODUJHO\ RQ WKH JHQXV SKUDVH DQG ZRUG 6HPDQWLF SURSHUWLHV DUH IRXQGLQZHOOHGLWHGJORVVDULHV VXFKDV(,$HGLWHG 1RWHWKDWWKH UHVXOWVDERYHGRQRWLQGLFDWH/(;,1*¶VSHUIRUPDQFH5DWKHUWKLV HYDOXDWLRQ LQGLFDWHV WKH SURILOH RI VRXUFH GHILQLWLRQV LQ WHUPV RI FRPSOH[LW\ )RUWKHVHFRQGHYDOXDWLRQWRGHWHUPLQHWKHDFFXUDF\RI/(;,1* ZH PDQXDOO\ WDJJHG  GHILQLWLRQV  HDFK IURP  GRPDLQV &LYLO (QJLQHHULQJ &RPSXWHU 7HUPV %LRPHGLFDO ,QIRUPDWLRQ *HQHUDO0HGLFDO,QIRUPDWLRQDQG(QHUJ\,QIRUPDWLRQ LQRUGHUWR HVWDEOLVKDPHDVXUHPHQWVWDQGDUGRU³JROGVWDQGDUG´7RFRPSXWH VFRULQJIRUJHQXVWHUPLGHQWLILFDWLRQDPDWFKZDVGHILQHGDVERWK WKH KXPDQ WDJJHU DQG WKH FRPSXWHU FKRRVLQJ WKH VDPH JHQXV WHUP:H WKHQ FRPSXWHGWKH DFFXUDF\RYHUHDFK GHILQLWLRQDOVHW 7R FRPSXWH VFRULQJ IRU JHQXV WHUP LGHQWLILFDWLRQ D PDWFK ZDV GHILQHGDVERWKWKHKXPDQWDJJHUDQGWKH FRPSXWHUFKRRVLQJWKH VDPHJHQXVWHUP 7DEOH*HQXV7HUP)LQGLQJ5HVXOWV

'RPDLQ *HQXV 7HUP

&LYLO (QJ

&RPS

%LRPHGLFDO

0HGLFDO

(QHUJ\



 















7KURXJK WKH ILUVW H[SHULPHQW RXWOLQHG LQ WKH DERYH VHFWLRQ ZH FRQWLQXDOO\ LVRODWHG DQG IL[HG HUURUV UHODWHG WR WKH VHPDQWLF DWWULEXWHV DQG VHSDUDWRUV $OWKRXJK VRPH RI RXU GHILQLWLRQDO VHWV GLG QRW FRQWDLQ DV PXFK H[WUDFWDEOH FRQWHQW DV ZH ZRXOG KDYH OLNHG WKH UHVXOWV RQ WKH HGLWHG VHWV ZHUH PRUH IUXLWIXO 7KURXJK WKLVIHHGEDFNPHFKDQLVP UXQZLWK NQRZQVHWFKHFNUHVXOWVDQG IL[  ZH DOVR LPSURYHG RXU JHQXV WHUP H[WUDFWLRQ WR DFKLHYH WKH VWURQJUHVXOWVVKRZQLQWKHVHFRQGH[SHULPHQW

 &21&/86,21

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

 5()(5(1&(6

>@ %\UG5-%.%RJXUDHY-/.ODYDQVDQG061HII

)URP6WUXFWXUDO$QDO\VLVRI/H[LFDO5HVRXUFHVWR6HPDQWLFV LQD/H[LFDO.QRZOHGJH %DVH8=HUQLN HGV 3URFHHGLQJV RIWKH)LUVW,QWHUQDWLRQDO:RUNVKRSRQ/H[LFDO$FTXLVLWLRQ 'HWURLW0LFKLJLDQ

>@ :KLWPDQ%ULDQ$XWRPDWLFDOO\'HWHUPLQLQJ'HILQLWLRQDO &RQWHQWIURP:HE'RFXPHQWV7HFKQLFDO5HSRUWLQ SURJUHVV&ROXPELD8QLYHUVLW\ 

>@ 1HII0DU\DQG%UDQ%RJXUDHY'LFWLRQDULHVGLFWLRQDU\

JUDPPDUVDQGGLFWLRQDU\HQWU\SDUVLQJ3URFHHGLQJVRIWKH 7ZHQW\VHYHQWK$QQXDO0HHWLQJRIWKH$VVRFLDWLRQIRU &RPSXWDWLRQDO/LQJXLVWLFV9DQFRXYHU&DQDGD

>@ .ODYDQV-&KRGRURZ0 :DFKROGHU1)URP

GLFWLRQDU\WRNQRZOHGJHEDVHYLDWD[RQRP\3URFHHGLQJVRI WKHWK $QQXDO&RQIHUHQFHRIWKH8:&HQWUHIRUWKH1HZ 2[IRUG(QJOLVK'LFWLRQDU\:DWHUORR2QWDULRSS 

>@ (YDQV'.ODYDQV-DQG:DFKROGHU1  'RFXPHQW 3URFHVVLQJZLWK/LQN,75HFKHUFKHG ,QIRUPDWLRQV$VVLVWpH SDU2UGLQDWHXU &RQWHQW%DVHG0XOWLPHGLD,QIRUPDWLRQ $FFHVV 3DULV)UDQFHSS

>@ .ODYDQV-/0XUHVDQ6'(),1'(55XOH%DVHG0HWKRGV IRUWKH([WUDFWLRQRI0HGLFDO7HUPLQRORJ\DQGWKHLU $VVRFLDWHG'HILQLWLRQVIURP2Q/LQH7H[W3URFHHGLQJVRI $0,$6\PSRVLXPS(YDQV'

.ODYDQV-DQG:DFKROGHU1