Machine Translation (MT), as many other natural language processing (NLP) areas, lacks a new look to the perspectives of the technology, as well as to the barriers preventing further MT commercialization. In this study we express and give prove to support our point of view about the future of MT (mainly emphasizing on Statistical MT (SMT)), ( Read more... )

SAMT (Syntax Augmented Machine Translation) is a hierarchical phrase-based translation model which annotates and generalizes existing phrase translations with syntactic categories derived from parsing the target side of a parallel corpus. This system roots from the "Chiang´s style" rules-system which deals with purely statistical categories. ( Read more... )

The presentation that I gave in the Center for Language Technology (Macquarie University)
slides

The slides and tracnsription of my talk given on the Eurolan 2007 Doctoral Consortium.

Slides (pdf)
Talk (pdf)

Due to scientific community's high interest in the SMT field, there are several shared tasks (evaluation campaigns), proposed to the research grouped all over the world to evaluate translation quality of their translation systems on an unseen test set. Evaluation is normally based on the automatic or/and human criteria, robusting evaluation results ( Read more... )

Research in the MT implies many difficulties, among which performance evaluation is one of the most challenging. So far, scientific community has not accepted unified criteria of MT evaluation. It would be ideal to have a generally accepted criteria of a SMT system automatic evaluation, like it is in the ASR field (word error rate - WER).

Nowadays ( Read more... )

Word reordering is one of the most challenging and important problems in SMT. Without reordering capabilities, sentences can be translated correctly only in case when both languages implied in translation have a similiar word order. When translating is between language pairs with high disparity in word order, word reordering is extremely desirable ( Read more... )

In parallel to the phrase-based approach to the SMT, the n-gram based approach has appeared, operating with bilingual units extracted from the aligned bilingual corpus, referred to as tuples [degispert02]. It regards translation as a stochastic process maximizing the joint probability p(f,e), leading to a decomposition based on bilingual Ngrams, ( Read more... )

The majority of the state-of-the-art systems is based on the approach proposed in the [brown93], it consists in using refined statistical models in the translation process. The idea is to estimate a translation model with the word
alignment as a hidden variable, as shown below:


( Read more... )

The idea of SMT lies in the translation of a source sentence f (traditionally referred to French) into a sentence in the target language e (English).
The problem is formulated in terms of source and target languages and is defined as arg max operation, as described by the following equation :


( Read more... )