<p>One of the notable developments in current natural language processing is the practical efficacy of probabilistic word representations, where words are embedded in high-dimensional continuous vector spaces that are optimized to reflect their distributional relationships. For sequences of words, such as phrases and sentences, distributional representations can be estimated by combining word embeddings using arithmetic operations like vector averaging or by estimating composition parameters from data using various objective functions. The quality of these compositional representations is typically estimated by their performance as features in extrinsic supervised classification benchmarks. Word and compositional embeddings for a single language can be induced without supervision using a large training corpus of raw text. To handle multiple languages and dialects, bilingual dictionaries and parallel corpora are often used for learning cross-lingual embeddings directly or to align pre-trained monolingual embeddings. In this work, we explore and develop various cross-lingual alignment techniques, compare the performance of the resulting cross-lingual embeddings, and study their characteristics. We pay particular attention to the bilingual data requirements of each approach since lower requirements facilitate wider language expansion. To begin with, we analyze various monolingual general-purpose sentence embedding models to better understand their qualities. By comparing their performance on extrinsic evaluation benchmarks and unsupervised clustering, we infer the characteristics of the most dominant features in their respective vector spaces. We then look into various cross-lingual alignment frameworks with different degrees of supervision. We begin with unsupervised word alignment, for which we propose an approach for inducing cross-lingual word mappings with no prior bilingual resources. We rely on assumptions about the consistency and structural similarities between the monolingual vector spaces of different languages. Using comparable monolingual news corpora, our approach resulted in highly accurate word mappings for two language pairs: French to English, and Arabic to English. With various refinement heuristics, the performance of the unsupervised alignment methods approached the performance of supervised dictionary mapping. Finally, we develop and evaluate different alignment approaches based on parallel text. We show that incorporating context in the alignment process often leads to significant improvements in performance. At the word level, we explore the alignment of contextualized word embeddings that are dynamically generated for each sentence. At the sentence level, we develop and investigate three alignment frameworks: joint modeling, representation transfer, and sentence mapping, applied to different sentence embedding models. We experiment with a matrix factorization model based on word-sentence co-occurrence statistics, and two general-purpose neural sentence embedding models. We report the performance of the various cross-lingual models with different sizes of parallel corpora to assess the minimal degree of supervision required by each alignment framework.

Thesis (Ph.D.)--Victoria University (Melbourne, Vic.), 2010.

Orientador: Eliane Martins / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-24T16:35:59Z (GMT). No. of bitstreams: 1 Letizio_CarolineCastello_M.pdf: 1670478 bytes, checksum: c2f330cd004addd705aef75384effcd2 (MD5) Previous issue date: 2013 / Resumo: Teste baseado em modelo é uma técnica na qual um sistema é modelado e geram-se testes a partir do modelo. Esta técnica apresenta como uma das maiores dificuldades a modelagem do sistema. Alguns dos empecilhos encontrados ao modelar são o que considerar do sistema e como representar essas considerações no modelo, incluindo em aplicações Web. Sendo assim, este trabalho propõe um método para auxiliar na modelagem de aplicações Web. Esta escolha é motivada pelo fato que esse tipo de sistema tem sido cada vez mais utilizado nos últimos anos e, com a sua evolução, sua complexidade tem aumentado consideravelmente, devido à dinamicidade e interatividade que esse tipo de aplicação oferece. A abordagem proposta é baseada em trabalhos correlatos, para se saber o que modelar da aplicação Web e como criar o modelo, descrevendo-se o passo-a-passo para se gerar os testes a partir dos modelos desenvolvidos e aplicá-los no sistema. A abordagem deste trabalho sugere tanto formas manuais como automáticas para os testes baseados em modelos. O método foi proposto para testar uma aplicação real: o ambiente de ensino à distância TelEduc. Os resultados, obtidos a partir da aplicação da proposta deste trabalho por um analista de teste e pela própria equipe do TelEduc, são analisados, a fim de validar a proposta e auxiliar no processo de desenvolvimento e aplicação dos testes de equipes de qualidade de software, principalmente, daquelas que não possuem nenhuma forma de organização de testes / Abstract: Model based testing is a technique in which the system is modeled and, from this model, tests can be generated. The major difficulty about this technique is to generate the model, since it is problematic to define what should be considered from the system and insert these characteristics into the model. This technique is also used in Web applications which presents the same challenges. Therefore, the present work proposes a method to assist in the modeling of Web applications. This sort of system has been increasingly used in recent years, where the system¿s complexity has considerably grown due to the dynamic and interactivity that this kind of application offers. The proposed method is based on related works in order to know what to model from the applications and how to create this model. The method contains the steps to generate tests cases from the created model, and how to execute them in the system. This work also suggests manual and automatic ways to execute the method. The proposal has been applied to test a real application: the e-learning Web environment TelEduc. The results obtained from the method¿s application by a test analyst and by the TelEduc team, are here analyzed to validate the method and to assist in the process development and tests execution by quality teams, especially those teams where there are no means of tests organization. / Mestrado / Ciência da Computação / Mestra em Ciência da Computação