о뷢չ
JOURNAL OF COMPUTER RESEARCH AND DEVELOPMENT
1999Vol.36No.7P.892-896



ͬƵʵĺ緽
Ź
ժҪעϢȻԴеһҪ.˹бעһʮַĹҺѰձ׼.ˣԶעоԵΪУҪʵԶעһҪ.ڶԡִʵ䡷¼ơʵ䡷עУͨͳڴϴĳƵʹһͬƵʾ.һͬƵʾ󼯳ϵĲνṹؼϡ.ڴ˻ϣԡʵ䡷еĶʽ磬Ϊ.
ؼʣͬƵʾ 봮
ţTP391
A CHINESE SENSE DISAMBIGUATION METHOD BASED ON
SENSE CO-OCCURRENCE FREQUENCY
ZHANG Yong-Kui 
(Department of Computer Science, Shanxi University, Taiyuan 030006)
ZHANG Guo-Qing
(Department of Computer Science, Shanxi University, Taiyuan 030006)
AbstractSense tagging is very important for the information retrieval and natural language processing. Tagging the sense code by hand is a time-consuming work and it is hard to make the sence codes consistent with each other. Therefore the research on automatic sense tagging is a very important work. To do this, the disambiguation of polysemous words must be solved first. In the research of sense tagging in Modern Chinese Dictionary(MCD) , a sense co-occurrence frequency matrix (SCFM) set is constructed by making statistics on the frequencies of sense code combination (SCC) of adjacent words that occur in a sample set . This SCFM makes full use of the hierarchy structure of the semantic system, and reduces data sparsness and redundancy. The disambiguation of those polysemous words are done and the words in MCD are tagged by employing a method based on the SCFM set. And the result is satisfying.
Key wordsco-occurrence frequency matrix, sense code combination(SCC)
1
ڶȻԵоУʵ䷽Ŀǰõһַ.ַоͨԶԴʵıķΪе.Ȼںһԣδı޷ֱʹã뾭ִʡԻעȴԺ󣬲ΪǵоṩõϢ.עĳɹ񣬹ϵоЧԼ׼ȷʣмҪ.ڱעеһؼ⣬ζԶʽ紦.ӢĶԹ· ˹ܷ ڴʵķ ϿͳƵķ.У˫Ͽķ1ݺͻڷķ2ݽΪɹ.廪ѧ1990꿪ʼԺעо.ԡִʵ䡷3жֵĽΪעʵı.ݱϵͳıעȷȶ80%ϣ45.ڱУһֻıڴϢ㷨ʵ֤һ㷨ǿе.
2ͬϢĻȡ
ͬϢǽжĻĳɹֱӹϵĳɰ.½򵥽ͬƵϢĻȡ.
ȣӡִʵ䡷ơʵ䡷ȡԼ6000壨Լռʵ䡷10ƥ䷨ıԶִ.Ȼ󣬶ԷִʺȫıעеʵıעɻԶУʵıעͨ˹ķ.עáͬʴ֡6ݣơ֡ķϵ.֡һʵ䣬մȫз࣬12࣬94࣬1428С.
ıע̣յõһѱע.ǶôͳƣôУռ5743Ϊ.һǶԡʵ䡷ȫıôͳƽͬһݴĳ˵õǻɿ.
֮ǴѱעгȡʵϴƵͳƣõйصͬƵϢͬƵʾʽʾЩϢ.ǵͬƵʾССʽ׼ȷȻһİ.һ棬ǶԪ֣ͬһ14281428ľݸѴ򼶣ʵѺѣԪԪֵͬͳƸǲ.ˣССʽһϡ⣬ʹͬƵʾ޷Ӧе.
ڻͬƵʾע㷨УȻͬƵʾԪԽ࣬Աע׼ȷԽ.Ԫϡ輰ִԵì.
ϼ㣬ڶͳʱǵϵĲԣûвССʽ.ĿٵĴС࣬öֻϵģʽӶһ̶Ͽ˷ϡ輰.ںܶ£Ѵṩ㹻Ϣ.ǲöģʽͳƣӶõһͬƵʾ.ǷһеͬƵʾ.
(1) ͬϴĳ
ϴΪn(nͨȡ234ͬƵʾ.һͬƵʾԪͨͳnʵϵ.һͬƵʾģʽǴࡢࡢСɵģ磺һԪͬƵʾСһԪͬƵʾ󣬵ȵ.
(2) ʵͷ
ȣǽܼ.
W뼯Ϊ{N1M1V1,N2M2V2NkMkVk} (NiΪ, MiΪ, Vi ΪС.(i=1,2,,k)),¶壺
1.  ( ij)NiNjWΪ.
2.  ( ij and Ni=Nj)MiMjWΪ.
3.  ( ij and Ni=Nj and Mi=Mj)ViVjWΪС.
ʵĲֿͬͣɰͬ־Ϊͬ־󣨹9ͬ־󣨹3Сͬ־󣨹9.3ͬ־ƵĹɣǴʸͷֱȡࡢࡢС룬ʣǰĴʣȡ룬ϴȷΪԪԪԪ3.Ϊʣôȡ룻Ϊʣôȡ룻ΪСʣôȡС. 
Դͬ־Ϊ˵.
ͬ־Ǵȡ࣬ǰȡ࣬һͬ־9ǽֱΪE1, E2,,E9. 1г9ͬ־ģʽ䲿Ԫ.
1

ͬ־ģʽԪؾ
E1
 |  |  | (AaEdHjK,0.000012)(EdHjKdK,0.000098)
(HjKdKcE,0.000010)(KdKcEaC,0.000021)
E2
 |  |  | (JbKaKHj,0.000011)(KaKdHDn,0.000013)
(KdHjDHj,0.000087)(DnDnDKb,0.000067)
E3
 |  |  | (KaIIhDn,0.000032)(IeIDnEb,0.000022)
(DkFEbIb,0.000013)(FaEIbKd,0.000032)
E4
 |  |  | (KGbDeEd,0.000054)(GDeEdKa,0.000038)
(DEdKaEe,0.000062)(KHiIfKa,0.000039)
E5
 |  | (DnKcD,0.000045)(KcDnD,0.000027)
(DnDnC,0.000037)(DnCbJ,0.000048)
E6
 |  | (IgCKd,0.000028)(CaKHh,0.000074)
(KdHHh,0.000084)(KbEBp,0.000073)
E7
 |  |  (CHiDd,0.000073)(HDdEd,0.000092)
(FDlKc,0.000078)(DKcDl,0.000082)
E8
 | (KaE,0.00028)(EdD,0.00018)
(DbK,0.00010)(KcI,0.00021)
E9
 | (EEa,0.00039)(EAa,0.00003)
(AEd,0.00008)(EIg,0.00029)


22ͬ־⣬ǻͳСԪͬ־E222ʾ
2

ͬ־ģʽԪؾ
E22
СС(Dj03Ca02,0.000064)(Ca02Ca02,0.000021)
(Ca02Kb05,0.000043)(Kb05Bq01,0.000084) 


3ͬƵʵ㷨 
ںһԣܶڲͬ໷вͬĺ壬ԣǽôڲͬ໷µ.ͨѱעͳƵóͬƵʾӳ˲ͬϴıеĳƵʣͬʱҲӳͬϻ£ͬϳֵƵ.ԣǵĻͬƵʵ㷨Ǻ.
4. 峤ȣWiΪSеĶʣʼѾĶʣѱ궨ʣܸΪWiSеķ峤ȣL(Wi)ʾ.
ɶ4Կ峤ȡһҪĸȡڶʵΧͬʱֱӰ쵽ʵѳ̶.ǿ԰ĳΧѱ궨ʣʼʣΪöʵĿɿ໷峤ԽöʱõĿɿ໷ԽõģʽҲԽӶ˸öʵ׼ȷ.ǽжڸʶǸΧķ峤ȣʹõͬ־.
棬ԴΪ˵ͬ־ķ.
һWiôK룬G(Wi )Ti1Ti2TikNi1Mi1Vi1Ni2Mi2Vi2NikMikVik ȻNi1Ni2Nik඼ͬ.Ҳ˵WiKK.ڴWi⣬ͿתΪһ⣺Wi໷£ѰKNi1Ni2NikеĳƵһҲѰڸ³ֿһ.
Wi໷ΪR
P(Nij|R)max(P(Ni1|R)P(Ni2|R)P(Nik|R))NijΪĴ࣬Ҳ˵TijΪWi .
Ƿֲͬ໷ۣ
1峤L(Wi)3ʱǰ׺ȵԭȡWiΧ3ѱ궨ʵ.磺WiǰֻWi-2Wi-1Ϊѱ궨ʣʱȡģʽΪNi-2Mi-2Ni-1Mi-1XNi+1Mi+1XΪNi1Ni2Nikеһ.ʱͳƵõͬƵʾE2вkϴֵƵʣPNi-2Mi-2Ni-1Mi-1Ni1Ni+1Mi+1PNi-2Mi-2Ni-1Mi-1Ni2Ni+1Mi+1PNi-2Mi-2Ni-1Mi-1NikNi+1Mi+1 ȡֵPNi-2Mi-2Ni-1Mi-1NijNi+1Mi+1 NijΪWiڵǰµĴ룬TijΪWi.
2峤L(Wi)=2ʱǰ׺ȵԭȡWiΧ2ѱ궨ʵ.
3峤L(Wi)=1ʱǰ׺ȵԭȡWiΧ1ѱ궨ʵ.
4Զע㷨Խ
һS=W1W2WnWiӦΪ{Ti1Ti2Tik}={Ni1Mi1Vi1,Ni2Mi2Vi2,,NikMikVik},(kΪ.WiΪʣk=1WiΪʣk>1.)
S=W1W2WnԶע£
1Sвڶʣע.
2ÿӦķ峤LWiз峤ĶWj
LWj=max{LWi| WiΪSеĶ}
LWj=0ת3ת4
3WjΪSз峤ĴʣLWj=0˵SеĴȫΪ.ʱȡڴWiWi+1пܵϴWiWi+1ӦΪTi1Ti2TimT(i+1)1T(i+1)2T(i+1)nпܵϴΪTi1T(i+1)1Ti1T(i+1)2Ti1T(i+1)nTi2T(i+1)1Ti2T(i+1)2Ti2T(i+1)nTimT(i+1)1TimT(i+1)2TimT(i+1)nȻǴͬƵʾE22вÿϴĳƵʣǰĽ֪ͬƵʾE22ģʽΪСС࣬ԣڴʵһϴɴE22ҵһӦƵֵȻǴеСԪϴȡһƵֵϴеСֱӦڴʣʱSоѱ궨ʣת2.
4WjΪSз峤ֵһʣʱΪWjĿɿ໷󣬹ʶȶWj磬Wj֮Ȼͬʱʵķ峤ȣҲ˵ʵĿɿ໷ӶΪʵ.
ڶWjǰȶWjĸзӶ϶Wjһ.ѡӦͬƵʾ󣬲һڽܵ㷨.
ֻͬƵʾ㷨Ľ˳õֻһͬƵʾķ֮ͬƵʾ.ⲻĸӶȣ׼ȷ.
˼룬Ƕԡʵ䡷ı˻Զעʵ飬1000ıעֹעĽ˱Ƚϣ֤㷨ɽ׼ȷߵ91.82.֤һ㷨ǿе.
ĿõȻѧĿ69575011ɽʡѧԱʡ.
߼飺У19456ڣҪоΪ˹ܺϢ.
Ź壬Ů197410˶ʿоִȻⷽо.
ߵλɽѧѧϵ̫ԭ030006
Ź壨ɽѧѧϵ̫ԭ030006
οף
1Gale W A, Church K W, Yarowsky D. Using bilingual material to develop word sense disambiguation methods. In: Proc of the Conference on Theoretical and Methodological Issues in Machine Translation, Montreal, 1992. 101112
2Yarowsky D. Word sense disambiguation using statistical models of rogetiscategories trained on large corpora. In: Proc of the 14th Intl Conf on Computational Linguistics, Nantes, 1992. 454460
3йѧԺо. ִʵ. ӡݣ1985
(Lexicon Editorial Office in Language Research Institute of the Chinese Academy of Social Sciences. Modern Chinese Dictionary (in Chinese). Beijing: Commercial Affairs Publication House, 1985)
4Ji Donghong, Huang Changning. Sense tagging of Tong Yi Ci Ci Lin using itself. The Latest Technological Advancement & Application. Singapore1996. 159164
5ݻƲͯ. ʵıԶע. Ӧã1993(4): 1825
(Huang Changning, Tong Xiang. The automatic sense tagging of real Chinese texts. Language and Word Application (in Chinese), 1993(4): 1825)
6÷ҾԵ. ͬʴ. Ϻ: Ϻ磬1989
(Mei Jiaju et al. Tong Yi Ci Ci Lin (in Chinese). Shanghai: The Shanghai Dictionary Publication House, 1983)
7ݻƲ. ڴģʵı̸. Ӧã1993(2): 110
(Huang Changning. A conversation about large scale real text processing.Language and Word Application(in Chinese),1993(2):110)
ոڣ1998-09-21
޸ڣ1999-01-20
