Identifying semantic relations in natural language text is an important component of many knowledge extraction systems. This thesis studies the task of hypernym discovery, i.e discovering terms that are related by the hypernymy (is-a) relation. Specifically, this thesis explores how state-of-the-art methods for hypernym discovery perform when applied in specific language domains. In recent times, state-of-the-art methods for hypernym discovery are mostly made up by supervised machine learning models that leverage distributional word representations such as word embeddings. These models require labeled training data in the form of term pairs that are known to be related by hypernymy. Such labeled training data is often not available when working with a specific language domain. This thesis presents experiments with an automatic training data collection algorithm. The algorithm leverages a pre-defined domain-specific vocabulary, and the lexical resource WordNet, to extract training pairs automatically. This thesis contributes by presenting experimental results when attempting to leverage such automatically collected domain-specific training data for the purpose of domain adaptation. Experiments are conducted in two different domains: One domain where there is a large amount of text data, and another domain where there is a much smaller amount of text data. Results show that the automatically collected training data has a positive impact on performance in both domains. The performance boost is most significant in the domain with a large amount of text data, with mean average precision increasing by up to 8 points.

Submitted by Maike Costa (maiksebas@gmail.com) on 2016-07-22T11:51:31Z No. of bitstreams: 1 arquivo total.pdf: 1246811 bytes, checksum: 4cada73ad2abaf6671beba97954e7973 (MD5) / Made available in DSpace on 2016-07-22T11:51:32Z (GMT). No. of bitstreams: 1 arquivo total.pdf: 1246811 bytes, checksum: 4cada73ad2abaf6671beba97954e7973 (MD5) Previous issue date: 2012-06-04 / The coreference processing is a phenomenon that has been studied internationally. In order to contribute to these national studies, we investigated the coreferential processing from the phenomenon of the Charge Repeated Name Penalty and Informational Charge Hypothesis in patients with aphasia. The overall objective is to analyze the processing of coreference in patients with Broca's aphasia, investigating, specifically if pronouns are processed more readily than names repeated in Brazilian Portuguese in subjects with and without the pathology; and if Superordinate Nominal Phrases are processed more quickly than hyponyms. Two experiments were conducted; and the experimental technique used was automonitoring reading in both experimentos, the independent variable is the type of recovery (pronoun and name repeated the in experiment 01, and Hyponym and hyperonym in the experiment 02) and as dependent variables the reading time of resumption and the accuracy rate to end-of-sentence questions. In experiment 01, the results of the control group were in the direction of what has been found in the literature, about to aphasic patients, the results showed insensitivity to the type of recovery, these data are discussed from a possible deficit in lexicon-semantic integration In experiment 02, once again, the control group follows the direction of what has been found in the literature, and the experimental group was observed that there is not a recovery-type preference. At the end of each phrase a question about it was performed. The subjects were able to answer such questions, indicating the establishment of coreference. We understand, from these data that patients with Broca's aphasia can access the lexical information and establish coreference, but need more time to do it. / O processamento da correferência é um fenômeno que vem sendo estudado internacionalmente. Buscando contribuir com essas estudos no âmbito nacional, investigamos o processamento correferencial a partir do fenômeno da Penalidade do Nome Repetido e Hipótese da Carga Informacional em pacientes com afasia. O objetivo geral do trabalho é analisar o processamento da correferência em pacientes com afasia de Broca, investigando, especificamente, se pronomes são processados mais prontamente do que nomes repetidos em português brasileiro tanto em sujeitos sem patologia, quanto em pacientes com afasia de Broca; e se Sintagmas Nominais (SN) superordenados são processados mais rapidamente do que hipônimos. Foram realizados dois experimentos; em ambos a técnica experimental utilizada foi a leitura automonitorada, sendo a variável independente o tipo de retomada (Pronome e Nome repetido no experimento 01, e Hipônimo e Hiperônimo no experimento 02) e como variáveis dependentes o tempo de leitura da retomada e o índice de acerto às perguntas de final de frase. No experimento 01, os resultados do grupo controle foram na direção do que já foi encontrado na literatura, já em relação aos pacientes afásicos, os resultados mostraram insensibilidade ao tipo de retomada, esses dados são discutidos a partir de um possível déficit na integração léxico-semântica. No experimento 02, mais uma vez, o grupo controle segue a direção do que já foi encontrado na literatura, e para o grupo experimental foi observado que não há uma preferência de tipo de retomada. Ao final de cada frase era realizada uma pergunta sobre a mesma. Os sujeitos conseguiram responder tais questões, indicando o estabelecimento da correferência. Entendemos, a partir desses dados que pacientes com afasia de Broca conseguem acessar a informação lexical e estabelecer a correferência, mas que necessitam de um maior tempo para fazê-lo.

Hiponimia ni uhusiano wa kifahiwa unaodhihirika baina ya leksimu ya jumla (hipanimu) na mahususi (hiponimu). Kama vile hipanimu mzazi hujumuisha hiponimu baba na mama. Uhusiano huu wa kihiponimia ulidhukuriwa na wanaisimu wa awali kuwa unahusisha leksimu nomino pekee. Hata hivyo, tafiti za hivi punde zinadhihirisha kuwa hiponimia huweza kudhihirika pia miongoni mwa kategoria za vivumishi, vielezi na vitenzi. Ingawa kuna midhihiriko ya hiponimia za vitenzi vya Kiswahili, wataalamu kadha wameelekea kushughulikia hiponimia za nomino na kutotilia maanani vitenzi. Hivyo basi, makala hii imechanganua uhusiano wa kihiponimia unaodhihirika miongoni mwa vitenzi teule vya Kiswahili. Katika kushughulikia suala hili, hipanimu vitenzi 24 kutoka kamusi za Kiswahili zimeteuli¬wa kimakusudi na hiponimu husika kutolewa. Nadharia ya Uchanganuzi Vijenzi kwa mujibu wa Katz na Fodor imezingatiwa katika uchanganuzi wa hiponimia hizi. Katika nadharia hii, sifa bainifu za hiponimu husika huonyeshwa kwa kutumia alama maalum za [+, -]. Data kuhusu hiponimia za vitenzi ilipekuliwa kutoka kamusi za Kiswahili kwa kuzingatia mwelekeo wa kiishara au kisintaksia kwa mujibu wa Hearst, na Snow na wenzake ili kuweza kutambua hipanimu na hiponimu husika katika sentensi. Hiponimia hizi zimechanganuliwa na kuwasilishwa kwa mtindo wa nadharia ya seti. / Hyponymy is a sense relation existing between general lexemes (hypernym) and the specific ones (hyponym). For instance, a hypernym parent includes hyponyms like father and mother. The hyponymy relation was regarded by earlier linguists that it could exist only amongst nouns. However, recent studies indicate that hyponymy can also be manifested in other categories such as adjectives, adverbs and verbs. Though there is hyponymy relation existing amongst Kiswahili verbs, various scho¬lars have tended to focus on nominal hyponymy and disregard verbal hyponymy. Therefore, this article has analyzed the hyponymy relation existing amongst selected Kiswahili verbs. In this regard, 24 verbal hypernyms from Kiswahili dictionaries were purposively sampled and their specific hyponyms indicated. Componential Analysis theory by Katz and Fodor has been used in the analysis of these hyponyms. The theory focuses on distinctive features of specific hyponyms that are normally shown by the use of specific signs [+, -]. The data on verbal hyponymy was observed from the Kiswahili dictionaries by the application of symbolic or syntactic approach propounded by Hearst, and Snow et al in order to identify the hypernyms and specific hyponyms in a sentence. These hyponymy relations are analyzed and presented using the set theory style.

Hiponimia ni uhusiano wa kifahiwa unaodhihirika baina ya leksimu ya jumla (hipanimu) na mahususi (hiponimu). Kama vile hipanimu mzazi hujumuisha hiponimu baba na mama. Uhusiano huu wa kihiponimia ulidhukuriwa na wanaisimu wa awali kuwa unahusisha leksimu nomino pekee. Hata hivyo, tafiti za hivi punde zinadhihirisha kuwa hiponimia huweza kudhihirika pia miongoni mwa kategoria za vivumishi, vielezi na vitenzi. Ingawa kuna midhihiriko ya hiponimia za vitenzi vya Kiswahili, wataalamu kadha wameelekea kushughulikia hiponimia za nomino na kutotilia maanani vitenzi. Hivyo basi, makala hii imechanganua uhusiano wa kihiponimia unaodhihirika miongoni mwa vitenzi teule vya Kiswahili. Katika kushughulikia suala hili, hipanimu vitenzi 24 kutoka kamusi za Kiswahili zimeteuli¬wa kimakusudi na hiponimu husika kutolewa. Nadharia ya Uchanganuzi Vijenzi kwa mujibu wa Katz na Fodor imezingatiwa katika uchanganuzi wa hiponimia hizi. Katika nadharia hii, sifa bainifu za hiponimu husika huonyeshwa kwa kutumia alama maalum za [+, -]. Data kuhusu hiponimia za vitenzi ilipekuliwa kutoka kamusi za Kiswahili kwa kuzingatia mwelekeo wa kiishara au kisintaksia kwa mujibu wa Hearst, na Snow na wenzake ili kuweza kutambua hipanimu na hiponimu husika katika sentensi. Hiponimia hizi zimechanganuliwa na kuwasilishwa kwa mtindo wa nadharia ya seti. / Hyponymy is a sense relation existing between general lexemes (hypernym) and the specific ones (hyponym). For instance, a hypernym parent includes hyponyms like father and mother. The hyponymy relation was regarded by earlier linguists that it could exist only amongst nouns. However, recent studies indicate that hyponymy can also be manifested in other categories such as adjectives, adverbs and verbs. Though there is hyponymy relation existing amongst Kiswahili verbs, various scho¬lars have tended to focus on nominal hyponymy and disregard verbal hyponymy. Therefore, this article has analyzed the hyponymy relation existing amongst selected Kiswahili verbs. In this regard, 24 verbal hypernyms from Kiswahili dictionaries were purposively sampled and their specific hyponyms indicated. Componential Analysis theory by Katz and Fodor has been used in the analysis of these hyponyms. The theory focuses on distinctive features of specific hyponyms that are normally shown by the use of specific signs [+, -]. The data on verbal hyponymy was observed from the Kiswahili dictionaries by the application of symbolic or syntactic approach propounded by Hearst, and Snow et al in order to identify the hypernyms and specific hyponyms in a sentence. These hyponymy relations are analyzed and presented using the set theory style.