A carregar...
Projeto de investigação
Sem título
Financiador
Autores
Publicações
Translation Alignment and Extraction Within a Lexica-Centered Iterative Workflow
Publication . Gomes, Luís Manuel dos Santos; Lopes, José
This thesis addresses two closely related problems. The first, translation alignment, consists of identifying bilingual document pairs that are translations of each other within
multilingual document collections (document alignment); identifying sentences, titles,
etc, that are translations of each other within bilingual document pairs (sentence alignment); and identifying corresponding word and phrase translations within bilingual
sentence pairs (phrase alignment). The second is extraction of bilingual pairs of equivalent word and multi-word expressions, which we call translation equivalents (TEs), from sentence- and phrase-aligned parallel corpora.
While these same problems have been investigated by other authors, their focus has
been on fully unsupervised methods based mostly or exclusively on parallel corpora.
Bilingual lexica, which are basically lists of TEs, have not been considered or given enough importance as resources in the treatment of these problems. Human validation of TEs, which consists of manually classifying TEs as correct or incorrect translations, has also not been considered in the context of alignment and extraction. Validation strengthens the importance of infrequent TEs (most of the entries of a validated lexicon) that otherwise would be statistically unimportant.
The main goal of this thesis is to revisit the alignment and extraction problems in the
context of a lexica-centered iterative workflow that includes human validation. Therefore, the methods proposed in this thesis were designed to take advantage of knowledge accumulated in human-validated bilingual lexica and translation tables obtained by unsupervised methods. Phrase-level alignment is a stepping stone for several applications, including the extraction of new TEs, the creation of statistical machine translation systems, and the creation of bilingual concordances. Therefore, for phrase-level alignment, the higher accuracy of human-validated bilingual lexica is crucial for achieving higher quality results in these downstream applications.
There are two main conceptual contributions. The first is the coverage maximization
approach to alignment, which makes direct use of the information contained in a lexicon, or in translation tables when this is small or does not exist. The second is the introduction of translation patterns which combine novel and old ideas and enables precise and productive extraction of TEs. As material contributions, the alignment and extraction methods proposed in this thesis have produced source materials for three lines of research, in the context of three PhD theses (two of them already defended), all sharing with me the supervision of my advisor. The topics of these lines of research are statistical machine translation, algorithms and data structures for indexing and querying phrase-aligned parallel corpora, and bilingual lexica classification and generation. Four publications have resulted directly from the work presented in this thesis and twelve from the collaborative lines of research.
Augmenting Translation Lexica by Learning Generalised Translation Patterns
Publication . Mahesh, Kavitha Karimbi; Lopes, José
Bilingual Lexicons do improve quality: of parallel corpora alignment, of newly extracted
translation pairs, of Machine Translation, of cross language information retrieval, among
other applications. In this regard, the first problem addressed in this thesis pertains to
the classification of automatically extracted translations from parallel corpora-collections
of sentence pairs that are translations of each other. The second problem is concerned
with machine learning of bilingual morphology with applications in the solution of first
problem and in the generation of Out-Of-Vocabulary translations.
With respect to the problem of translation classification, two separate classifiers for
handling multi-word and word-to-word translations are trained, using previously extracted
and manually classified translation pairs as correct or incorrect. Several insights
are useful for distinguishing the adequate multi-word candidates from those that are
inadequate such as, lack or presence of parallelism, spurious terms at translation ends
such as determiners, co-ordinated conjunctions, properties such as orthographic similarity
between translations, the occurrence and co-occurrence frequency of the translation
pairs. Morphological coverage reflecting stem and suffix agreements are explored as key
features in classifying word-to-word translations. Given that the evaluation of extracted
translation equivalents depends heavily on the human evaluator, incorporation of an
automated filter for appropriate and inappropriate translation pairs prior to human evaluation
contributes to tremendously reduce this work, thereby saving the time involved
and progressively improving alignment and extraction quality. It can also be applied
to filtering of translation tables used for training machine translation engines, and to
detect bad translation choices made by translation engines, thus enabling significative
productivity enhancements in the post-edition process of machine made translations.
An important attribute of the translation lexicon is the coverage it provides. Learning
suffixes and suffixation operations from the lexicon or corpus of a language is an extensively
researched task to tackle out-of-vocabulary terms. However, beyond mere words
or word forms are the translations and their variants, a powerful source of information
for automatic structural analysis, which is explored from the perspective of improving
word-to-word translation coverage and constitutes the second part of this thesis. In this
context, as a phase prior to the suggestion of out-of-vocabulary bilingual lexicon entries,
an approach to automatically induce segmentation and learn bilingual morph-like units by identifying and pairing word stems and suffixes is proposed, using the bilingual
corpus of translations automatically extracted from aligned parallel corpora, manually
validated or automatically classified. Minimally supervised technique is proposed to enable
bilingual morphology learning for language pairs whose bilingual lexicons are highly
defective in what concerns word-to-word translations representing inflection diversity.
Apart from the above mentioned applications in the classification of machine extracted
translations and in the generation of Out-Of-Vocabulary translations, learned bilingual
morph-units may also have a great impact on the establishment of correspondences of
sub-word constituents in the cases of word-to-multi-word and multi-word-to-multi-word
translations and in compression, full text indexing and retrieval applications.
Unidades organizacionais
Descrição
Palavras-chave
Contribuidores
Financiadores
Entidade financiadora
Fundação para a Ciência e a Tecnologia
Programa de financiamento
3599-PPCDT
Número da atribuição
PTDC/EIA-EIA/114521/2009