Statistical post-editing for a statistical MT system

2012

Statistical-based methods are the prevalent approaches for implementing machine translation systems today. However the resulted translations are usually flawed to some degree. We assume that a statistical baseline system can be re-used to automatically learn how to (partially) correct translation errors, i.e. to turn a "broken" target translation into a better one. By training and testing on initial bilingual data, we constructed a system S1 which was used to translate the source language part of the training corpus. The new translated corpus and its reference translation are used to train and test another similar system S2. Without any additional data, the chain S1+S2 shows a sensible quality increase against S1 in terms of BLEU scores, for both translation directions (English to Romanian and Romanian to English).

Proceedings of the Second Workshop on Statistical Machine Translation - StatMT '07, 2007

This article describes the combination of a SYSTRAN system with a "statistical postediting" (SPE) system. We document qualitative analysis on two experiments performed in the shared task of the ACL 2007 Workshop on Statistical Machine Translation. Comparative results and more integrated "hybrid" techniques are discussed.

2007

We propose to use a statistical phrasebased machine translation system in a post-editing task: the system takes as input raw machine translation output (from a commercial rule-based MT system), and produces post-edited target-language text. We report on experiments that were performed on data collected in precisely such a setting: pairs of raw MT output and their manually post-edited versions. In our evaluation, the output of our automatic post-editing (APE) system is not only better quality than the rule-based MT (both in terms of the BLEU and TER metrics), it is also better than the output of a stateof-the-art phrase-based MT system used in standalone translation mode. These results indicate that automatic post-editing constitutes a simple and efficient way of combining rule-based and statistical MT technologies.

2010

Abstract In this paper, we present the insights gained from a detailed study of coupling a highly modular English-Hindi RBMT system with a standard phrase-based SMT system. Coupling the RBMT and SMT systems at various stages in the RBMT pipeline, we observe the effects of the source transformations at each stage on the performance of the coupled MT system. We propose an architecture that systematically exploits the structural transfer and robust generation capabilities of the RBMT system.

Conference of the Association for Machine Translation in the Americas, 2008

We propose a two-stage system for spoken language machine translation. In the first stage, the source sentence is parsed and paraphrased into an intermediate language which retains the words in the source language but follows the word order of the target language as much as feasible. This stage is mostly linguistic. In the second stage, a statistical MT is performed to translate the intermediate language into the target language. For the task of English-to-Mandarin translation, we achieved a 2.5 increase in BLEU score and a 45% decrease in GIZA-Alignment Crossover, on IWSLT-06 data. In a human evaluation of the sentences that differed, the two-stage system was preferred three times as often as the baseline.

2010

Although the scoring features of state-of-theart Phrase-Based Statistical Machine Translation (PB-SMT) models are weighted so as to optimise an objective function measuring translation quality, the estimation of the features themselves does not have any relation to such quality metrics. In this paper, we introduce a translation quality-based feature to PB-SMT in a bid to improve the translation quality of the system. Our feature is estimated by averaging the edit-distance between phrase pairs involved in the translation of oracle sentences, chosen by automatic evaluation metrics from the N -best outputs of a baseline system, and phrase pairs occurring in the N -best list. Using our method, we report a statistically significant 2.11% relative improvement in BLEU score for the WMT 2009 Spanish-to-English translation task. We also report that using our method we can achieve statistically significant improvements over the baseline using many other MT evaluation metrics, and a substantial increase in speed and reduction in memory use (due to a reduction in phrase-table size of 87%) while maintaining significant gains in translation quality.

2007

It is generally acknowledged that the performance of rulebased machine translation (RMBT) systems can be greatly improved through domain-specific system adaptation. To that end, RBMT users often choose to invest significant resources into the development of ad hoc MT dictionaries. In this paper, we demonstrate that comparable customization effects can be achieved automatically. One effective way to do that is to post-edit the translations produced by a vanilla RBMT system using a specially-trained statistical machine translation (SMT) system. Our experiments indicate that this method is just as effective as manual customization of system dictionaries in reducing the need for manual postediting.

Communications in Computer and Information Science, 2014

Post-editing has been successfully applied to correct the output of MT systems to generate better translation, but as a downstream task its positive feedback to MT has not been well studied. In this paper, we present a novel rule refinement method which uses Simulated Post-Editing (SiPE) to capture the errors made by the MT systems and generates refined translation rules. Our method is system-independent and doesn't entail any additional resources. Experimental results on largescale data show a significant improvement over both phrase-based and syntax-based baselines.

2009 Seventh Brazilian Symposium in Information and Human Language Technology, 2009

In this paper we describe some experiments carried out to test the impact of automatic casing and punctuation changes when training and testing statistical translation models. The experiments described here concern the translation from/to English and Brazilian Portuguese texts but since the superficial changes investigated are language independent, we believe that the conclusions can be applied to many other pairs of languages. These experiments were designed aiming at setting a baseline scenario for future training and testing of more complex statistical translation models such as the factored ones. From the experiments presented here it is possible to see that case and punctuation changes have a significant impact on automatic translation results.