Measuring semantic similarity is a task needed in many Natural Language Processing (NLP) applications. For example, in Machine Translation evaluation, semantic similarity is used to assess the quality of the machine translation output by measuring the degree of equivalence between a reference translation and the machine translation output. The problem of semantic similarity (Corley and Mihalcea, 2005) is de ned as measuring and recognising semantic relations between two texts. Semantic similarity covers di erent types of semantic relations, mainly bidirectional and directional. This thesis proposes new methods to address the limitations of existing work on both types of semantic relations. Recognising Textual Entailment (RTE) is a directional relation where a text T entails the hypothesis H (entailment pair) if the meaning of H can be inferred from the meaning of T (Dagan and Glickman, 2005; Dagan et al., 2013). Most of the RTE methods rely on machine learning algorithms. de Marne e et al. (2006) propose a multi-stage architecture where a rst stage determines an alignment between the T-H pairs to be followed by an entailment decision stage. A limitation of such approaches is that instead of recognising a non-entailment, an alignment that ts an optimisation criterion will be returned, but the alignment by itself is a poor predictor for iii non-entailment. We propose an RTE method following a multi-stage architecture, where both stages are based on semantic representations. Furthermore, instead of using simple similarity metrics to predict the entailment decision, we use a Markov Logic Network (MLN). The MLN is based on rich relational features extracted from the output of the predicate-argument alignment structures between T-H pairs. This MLN learns to reward pairs with similar predicates and similar arguments, and penalise pairs otherwise. The proposed methods show promising results. A source of errors was found to be the alignment step, which has low coverage. However, we show that when an alignment is found, the relational features improve the nal entailment decision. The task of Semantic Textual Similarity (STS) (Agirre et al., 2012) is de- ned as measuring the degree of bidirectional semantic equivalence between a pair of texts. The STS evaluation campaigns use datasets that consist of pairs of texts from NLP tasks such as Paraphrasing and Machine Translation evaluation. Methods for STS are commonly based on computing similarity metrics between the pair of sentences, where the similarity scores are used as features to train regression algorithms. Existing methods for STS achieve high performances over certain tasks, but poor results over others, particularly on unknown (surprise) tasks. Our solution to alleviate this unbalanced performances is to model STS in the context of Multi-task Learning using Gaussian Processes (MTL-GP) ( Alvarez et al., 2012) and state-of-the-art iv STS features ( Sari c et al., 2012). We show that the MTL-GP outperforms previous work on the same datasets.

Submitted by Fabricia Fialho Reginato (fabriciar) on 2015-07-01T23:00:34Z No. of bitstreams: 1 EvandroFlores.pdf: 1289007 bytes, checksum: 44450c63dc59c23ca86b3e4fdbdcea30 (MD5) / Made available in DSpace on 2015-07-01T23:00:34Z (GMT). No. of bitstreams: 1 EvandroFlores.pdf: 1289007 bytes, checksum: 44450c63dc59c23ca86b3e4fdbdcea30 (MD5) Previous issue date: 2014 / CNPQ – Conselho Nacional de Desenvolvimento Científico e Tecnológico / GVDASA / A rápida evolução das tecnologias de informação e comunicação tem possibilitado o desenvolvimento de modalidades de ensino e educação, tais como a Educação a Distância, capazes de alcançar pessoas anteriormente impossibilitadas de frequentar o ensino superior. Um aspecto importante destas modalidades é o amplo uso de recursos de mediação digital, sendo que estes podem gerar um grande volume de dados o qual, por vezes, não é viável para utilização proveitosa de forma manual pelos professores envolvidos nesta interação. Este contexto gera a necessidade e oportunidade de definição de ferramentas que possam atuar para automatizar parte deste trabalho. Uma destas possibilidades é a verificação de correção de respostas textuais, onde o objetivo é identificar vinculações entre amostras textuais que podem ser, por exemplo, diferentes respostas textuais a uma pergunta. Embora sejam utilizadas com bons resultados, as técnicas atualmente aplicadas a este problema apresentam deficiências ou características que diminuem sua precisão ou adequação em diversos contextos. Poucos trabalhos são capazes de realizar a vinculação textual caso seja alterada a flexão verbal, outros não são capazes de identificar informações importantes ou em que posição na frase as informações se encontram. Além disso, poucos trabalhos são adaptados para a língua portuguesa. Este trabalho propõe um modelo de reconhecimento de vinculação textual baseado em regras linguísticas e informações morfossintáticas voltado para ambientes virtuais de ensino e aprendizagem, que busca contornar estes problemas apresentando uma nova abordagem através do uso combinado da análise sintática, morfológica, regras linguísticas, detecção da flexão de voz, tratamento de negação e do uso de sinônimos. O trabalho também apresenta um protótipo desenvolvido para avaliar o modelo proposto. Ao final são apresentados os resultados obtidos, que até o momento são promissores, permitindo a identificação da vinculação textual de diferentes amostras textuais com precisão e flexibilidade relevantes. / The fast evolution of information and communication technologies has enabled the development of modalities of teaching and learning, such as distance education, that allow to reach people previously unable to attend higher education. An important aspect of these modalities is the extensive use of digital mediation resources. These resources can generate a large volume of data that sometimes is not feasible for beneficial manual use by the teachers involved in this interaction. In this context there is a necessity and opportunity for defining tools and approaches that can act to automate part of this work. One of these possibilities is the verification of textual responses correctness, where the goal is to identify linkages between textual samples, which can be, for example, different textual answer to a question. Although presenting good results, techniques currently applied to this problem have deficiencies or characteristics that decrease their accuracy or suitability in several contexts. Few studies are able to perform textual entailment in case the verbal inflection was changed; others are not able to identify important information or position in the sentence where the information is found. Moreover, few works are adapted to Portuguese. This work proposes a model to recognition of textual entailment based on linguistic rules, which seeks to overcome these problems by presenting a new approach through the combined use of syntactic analysis, morphology, linguistic rules, detection of the bending voice, treatment of denial and the use of synonyms. This work also presents a prototype developed to evaluate the model proposed herein. The end results, which are promising, allow the identification of textual linking of different textual samples accurately and with flexibility.