Esta dissertação relata o primeiro trabalho de pesquisa em alinhamento automático de textos paralelos envolvendo o português brasileiro (PB). Neste trabalho foram implementados cinco métodos de alinhamento sentencial automático bastante referenciados na literatura, incluindo métodos empíricos, lingüísticos e híbridos, avaliados com textos paralelos PB-inglês. Os resultados mostraram-se compatíveis com os relatados para outros pares de línguas, sendo que as maiores precisões (acima de 94%) foram obtidas em corpora sem ruídos (sem erros gramaticais e de tradução), conforme era esperado. Além disso, os resultados apontam muita semelhança no desempenho de todos os métodos, o que impossibilita a eleição de um deles como o melhor. Além da implementação dos métodos de alinhamento sentencial e dos corpora paralelos construídos para avaliá-los, outros recursos lingüísticos e computacionais de grande valor para as pesquisas em PLN foram gerados durante este trabalho. / As the first attempt at automatic parallel text alignment involving Brazilian Portuguese, in this research we implemented five well-known automatic sentence alignment methods, including empirical, linguistic and hybrid techniques, and evaluated them as applied to Brazilian Portuguese-English parallel texts. The results are in accordance with those reported for other pairs of languages, even in that highest precisions (above 94%) were obtained for corpora without noise (i.e. grammatical or translation errors), as expected. Furthermore, the results point to a virtual tie between the methods, it being impossible to elect one as the best. In addition to the implementations of those methods and the parallel corpora built to evaluate them, other linguistic and computational resources were built during this work which are of great value to PLN research.

In this masters’ thesis report we present a graph theoretical method used for automatic bilingual lexicon acquisition with parallel texts. We analyze the concept of graph similarity and give an interpretation, of the parallel texts, connected to the vector space model. We represent the parallel texts by a directed, tripartite graph and from here use the corresponding adjacency matrix, A, to compute the similarity of the graph. By solving the eigenvalue problem ρS = ASAT + ATSA we obtain the self-similarity matrix S and the Perron root ρ. A rank k approximation of the self-similarity matrix is computed by implementations of the singular value decomposition and the non-negative matrix factorization algorithm GD-CLS. We construct an algorithm in order to extract the bilingual lexicon from the self-similarity matrix and apply a statistical model to estimate the precision, the correctness, of the translations in the bilingual lexicon. The best result is achieved with an application of the vector space model with a precision of about 80 %. This is a good result and can be compared with the precision of about 60 % found in the literature.

<p>In this masters’ thesis report we present a graph theoretical method used for automatic bilingual lexicon acquisition with parallel texts. We analyze the concept of graph similarity and give an interpretation, of the parallel texts, connected to the vector space model. We represent the parallel texts by a directed, tripartite graph and from here use the corresponding adjacency matrix, A, to compute the similarity of the graph. By solving the eigenvalue problem ρS = ASAT + ATSA we obtain the self-similarity matrix S and the Perron root ρ. A rank k approximation of the self-similarity matrix is computed by implementations of the singular value decomposition and the non-negative matrix factorization algorithm GD-CLS. We construct an algorithm in order to extract the bilingual lexicon from the self-similarity matrix and apply a statistical model to estimate the precision, the correctness, of the translations in the bilingual lexicon. The best result is achieved with an application of the vector space model with a precision of about 80 %. This is a good result and can be compared with the precision of about 60 % found in the literature.</p>

Esta dissertação relata o primeiro trabalho de pesquisa em alinhamento automático de textos paralelos envolvendo o português brasileiro (PB). Neste trabalho foram implementados cinco métodos de alinhamento sentencial automático bastante referenciados na literatura, incluindo métodos empíricos, lingüísticos e híbridos, avaliados com textos paralelos PB-inglês. Os resultados mostraram-se compatíveis com os relatados para outros pares de línguas, sendo que as maiores precisões (acima de 94%) foram obtidas em corpora sem ruídos (sem erros gramaticais e de tradução), conforme era esperado. Além disso, os resultados apontam muita semelhança no desempenho de todos os métodos, o que impossibilita a eleição de um deles como o melhor. Além da implementação dos métodos de alinhamento sentencial e dos corpora paralelos construídos para avaliá-los, outros recursos lingüísticos e computacionais de grande valor para as pesquisas em PLN foram gerados durante este trabalho. / As the first attempt at automatic parallel text alignment involving Brazilian Portuguese, in this research we implemented five well-known automatic sentence alignment methods, including empirical, linguistic and hybrid techniques, and evaluated them as applied to Brazilian Portuguese-English parallel texts. The results are in accordance with those reported for other pairs of languages, even in that highest precisions (above 94%) were obtained for corpora without noise (i.e. grammatical or translation errors), as expected. Furthermore, the results point to a virtual tie between the methods, it being impossible to elect one as the best. In addition to the implementations of those methods and the parallel corpora built to evaluate them, other linguistic and computational resources were built during this work which are of great value to PLN research.

The majority of the world's languages are poorly represented in informational media like radio, television, newspapers, and the Internet. Translation into and out of these languages may offer a way for speakers of these languages to interact with the wider world, but current statistical machine translation models are only effective with a large corpus of parallel texts - texts in two languages that are translations of one another - which most languages lack. This thesis describes the Babylon project which attempts to alleviate this shortage by supplementing existing parallel texts with texts gathered automatically from the Web -- specifically targeting pages that contain text in a pair of languages. Results indicate that parallel texts gathered from the Web can be effectively used as a source of training data for machine translation and can significantly improve the translation quality for text in a similar domain. However, the small quantity of high-quality low-density language parallel texts on the Web remains a significant obstacle.

The aim of this paper is to analyse what strategies may be the most suitable ones when dealing with problems in the translation of a text on human trafficking. The focus of the study is how to translate terminology related to human trafficking, but also on how to deal with references to culture-specific phenomena. In order to find possible solutions for translation problems in the text, different dictionaries and term banks were used, as well as parallel texts and different translation strategies.           The analysis is based on the translation strategies laid down by Vinay and Darbelnet, Ingo and Newmark. The findings of the study show that a number of different translation strategies had to be applied in order to solve the different translation problems. Regarding the translation of terminology, modulation, equivalence, addition and omission were the most useful strategies, as well as using parallel texts. The strategies found most useful in translating culture-specific phenomena were modulation, equivalence and addition, but componential analysis, transference and established translations were also used. Factors that decided what were the most appropriate strategies to apply in the different instances were the aim of the text, the perspective of the target reader and parallel texts.

The purpose of this essay is to analyze the translation of the specific technical language of railway transport from Swedish into German. The analysis is based on a self-conducted translation of selected passages of rules and regulations issued by the Swedish government agency Trafikverket (The Swedish Transport Administration). The following questions serve as a starting point: Which factors influence the final choice of a translation equivalent when several options are available? Which translation strategies can be used to close a lexical gap in the target language? And, furthermore, how can ”false friends” be identified so that terminological errors can be avoided in the target text? The essay is divided into two parts. The first part introduces major concepts such as equivalence, denotative equivalence and false friends. The second part consists of an analysis which discusses a selection of representative examples from the translation. The first part of the analysis focuses on cases where the target language provides several translation equivalents. By analyzing the semantics and comparing the related meanings of these equivalents, it was possible to pin down the most suitable term in the target language. In most cases the final term choice was confirmed by parallel texts. The second part discusses cases where the target language lacks an equivalent technical term. In most cases the lexical gaps could be closed by loaning the source-language term and adding a short explanation in the target text. The final section is devoted to false friends. By consulting parallel texts, several false friends in the target language were identified and also removed from an earlier version of the target text.

By combining quantitative and qualitative methods, the present study aims to investigate what strategies may be used in the translation of a text on football from English to Swedish, with focus on terminology and the metaphor FOOTBALL IS WAR. Special attention is paid to the trait of football language as a special language. Dictionaries and parallel texts, as well as different sources on football language and translation theory, such as Bergh &amp; Ohlander (2012), Levin (2008) and Newmark (1988) were consulted in the translation and the subsequent analysis. As expected, the results show that borrowing is a very common strategy in the translation of football terminology. The loanwords were categorised as either direct (letter-for-letter or adjusted) or indirect (with the same or reversed order of inherent elements), and the most common kind was indirect loans (calques) with retained order of elements. Moreover, some terminology had to be translated by means of Vinay &amp; Darbelnet’s (1958) equivalence. With regard to terms based on the FOOTBALL IS WAR metaphor, many could be reproduced in the target text, while others were changed to another image, and yet others reduced to literal language, after careful consideration of their frequency and appropriateness in Swedish football language. One of the major conclusions drawn from the present study is that some knowledge about the topic of the source text, as well as extensive and proper use of parallel texts is necessary, considering the distinctiveness of football language as opposed to general language.

Aim of this thesis is to implement system for translation words from source language into the target language with pair input texts. There are descriptions  of terms and methods used in machine translation and machine build dictionary. The thesis also contains a concept and specification of each part created system including final evaluation. There is analysed options which make extension of existing dictionatry.

This study investigates expressions akin to ‘again’, which inhabit the semantic domain of repetition and restitution, from a cross-linguistic perspective. Using massively parallel corpora as the primary source of data the aim of this study is to investigate whether the encoding of repetitive and restitutive meaning is a cross-linguistically valid difference and if there are any patterns in the language specific variation of the repetitive and restitutive domain. By using Multi-Dimensional Scaling and Partitioning Around Medoids to investigate how the expressions ‘third time’, ‘second time’, ‘again’, ‘back’ and ‘return’ make up the semantic space of the domain, it was determined that the domain in question forms a continuum of meanings. This scale, named the TURN-hierarchy, is comprised of repetitive expressions like ‘third time’ to the far left, ambiguous expressions like ‘again’ in the intermediate section and restitutive expressions such as ‘return, back’ to the far right. Furthermore, the results show that repetitive and restitutive meaning is encoded differently in a majority of the sample languages, and that there is asymmetry in the encoding of repetition and restitution where repetitive meaning is privileged. Thus,it is proposed that all languages have at least one exclusively repetitive expression.