Comparing Parallel Corpora and Evaluating their Quality

1996

We report on methods of improving multilingual text alignments that have been produced in a simple dynamic-programming scheme, by automated detec- tion of possible misalignments. Details of methods involving cognates, specially- identified words, and propositional contents of sentences are given, together with notable features of their performance on parallel corpora in a number of different types of European languages. 1.

2004

This paper presents a simple way of producing symmetric, phrase-based alignments, combining two single-word based alignments. Our algorithm exploits the asymmetries in the superposition of the two word alignments to detect the phrases that must be aligned as a whole. It was run with baseline word alignments produced by the Giza++ software and improved these alignments. The ability to treat some groups of words as a whole is essential in applications like machine translation. The paper also addresses the difficulty of the alignment evaluation task.

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop

Parallel corpora are key to developing good machine translation systems. However, abundant parallel data are hard to come by, especially for languages with a low number of speakers. When rich morphology exacerbates the data sparsity problem, it is imperative to have accurate alignment and filtering methods that can help make the most of what is available by maximising the number of correctly translated segments in a corpus and minimising noise by removing incorrect translations and segments containing extraneous data. This paper sets out a research plan for improving alignment and filtering methods for parallel texts in low-resource settings. We propose an effective unsupervised alignment method to tackle the alignment problem. Moreover, we propose a strategy to supplement state-of-theart models with automatically extracted information using basic NLP tools to effectively handle rich morphology.

This paper profiles the Europarl part of an English-Swedish parallel corpus and compares it with three other subcorpora of the same parallel corpus. We first describe our method for comparison which is based on alignments, both at the token level and the structural level. Although two of the other subcorpora contains fiction, it is found that the Europarl part is the one having the highest proportion of many types of restructurings, including additions, deletions and long distance reorderings. We explain this by the fact that the majority of Europarl segments are parallel translations.

In machine translation, document alignment refers to finding correspondences between documents which are exact translations of each other. We define pseudo-alignment as the task of finding topical-as opposed to exact-correspondences between documents in different languages. We apply semisupervised methods to pseudo-align multilingual corpora. Specifically, we construct a topicbased graph for each language. Then, given exact correspondences between a subset of documents, we project the unaligned documents into a shared lower-dimensional space. We demonstrate that close documents in this lower-dimensional space tend to share the same topic. This has applications in machine translation and cross-lingual information analysis. Experimental results show that pseudo-alignment of multilingual corpora is feasible and that the document alignments produced are qualitatively sound. Our technique requires no linguistic knowledge of the corpus. On average when 10% of the corpus consists of exact correspondences, an on-topic correspondence occurs within the top 5 foreign neighbors in the lowerdimensional space while the exact correspondence occurs within the top 10 foreign neighbors in this this space. We also show how to substantially improve these results with a novel method for incorporating language-independent information.

2011

Resumen: Este artículo presenta un algoritmo independiente de lengua para la alineación de corpus paralelo a nivel de documento, de oración y de vocabulario, tomando comoúnica fuente de información el mismo corpus a alinear. La entrada es un conjunto de documentos escritos en dos lenguas desconocidas A y B, donde cada documento en la lengua A tiene su correspondiente traducción a la lengua B. El problema consiste en: 1) dividir el conjunto de documentos en las dos lenguas; 2) alinear a nivel de documento: determinar qué documento en la lengua A es el original (o la traducción) de cada documento en la lengua B; 3) alinear a nivel de oración: determinar qué oración en el original corresponde a qué oración en la traducción y 4) alinear a nivel del vocabulario: determinar qué palabra en una lengua es equivalente a cada palabra en la traducción. El algoritmo es iterativo, ya que utiliza el vocabulario bilingüe resultante para realinear el corpus. La evaluación en inglés, castellano y francés muestra resultados competitivos en todos los niveles. Palabras clave: Alineación de corpus paralelo, Extracción de información, traducción automática Abstract: This paper presents a language independent algorithm for the alignment of parallel corpora at the document, sentence and vocabulary levels using the to-be aligned corpus itself as the only source of information. The input is a set of documents written in two unknown languages A and B, where every document in language A has its corresponding translation into language B. The problem thus consists of: 1) dividing the set of documents in the two languages; 2) aligning at the document level to determine which document in language A is the original (or translation) of each document in language B; 3) aligning at the sentence level to determine which sentence in the original corresponds to each sentence in the translation and 4) aligning at the vocabulary level to determine which word in one language is equivalent to each word in the translation. The algorithm is iterative, using the resulting bilingual vocabulary to re-align the corpus. Evaluation figures in English, Spanish and French show competitive results at all levels of the alignment.

Recent Advances in …, 2005

The choice of natural language technology appropriate for a given language is greatly impacted by density (availability of digitally stored material). More than half of the world speaks medium density languages, yet many of the methods appropriate for high or low density languages yield suboptimal results when applied to the medium density case. In this paper we describe a general methodology for rapidly collecting, building, and aligning parallel corpora for medium density languages, illustrating our main points on the case of Hungarian, Romanian, and Slovenian. We also describe and evaluate the hybrid sentence alignment method we are using.

2005

This paper presents the task definition, resources, participating systems, and comparative results for the shared task on word alignment, which was organized as part of the ACL 2005 Workshop on Building and Using Parallel Texts. The shared task included English-Inuktitut, Romanian-English, and English-Hindi sub-tasks, and drew the participation of ten teams from around the world with a total of 50 systems.

2012

Abstract In this paper, we describe an accurate, robust and language-independent algorithm to align paragraphs with their translations in a parallel bilingual corpus. The paragraph alignment is tested on 998 anchors (combination of 7 books) of English-Hindi language pair of Gyan-Nidhi corpus and achieved a precision of 86.86% and a recall of 82.03%.

2007

Multilingual technologies, which to a large extent are language independent, provide a powerful support for easier building of annotated linguistic resources for languages where such resources are scarce or missing. All these technologies require parallel corpora in order to achieve their ends. Parallel texts encode extremely valuable linguistic knowledge because the linguistic decisions made by the human translators in order to faithfully convey the meaning of the source text can be traced and used as evidence on linguistic facts which, in a monolingual context, might be unavailable to or overlooked by a computer program. In this paper we will briefly present some underlying multilingual technologies and methodologies we developed for exploiting parallel corpora and we will discuss their relevance for cross-linguistic annotation transfer and applications.