Learning to select a good translation

Tidhar, Dan; Küssner, Uwe

doi:10.3115/992730.992768

Cited by 7 publications

(6 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Multi‐engine machine translation (MEMT), which combines the best results from a variety of MT systems working simultaneously on the same text to improve the overall quality, has been a very active area in machine translation research (Nirenburg and Frederking, 1994). Different approaches have been proposed and experiments conducted to combine results from multiple systems (Nirenburg and Frederking, 1994; Tidhar and Küssner, 2000; Akiba et al , 2002; Callison‐Burch and Flournoy, 2001; Nomoto, 2004; Jayaraman and Lavie, 2005; Matusov et al , 2006; Rosti et al , 2007; Chen et al , 2007). MEMT has the potential to achieve significantly better performance than any single MT system (Callison‐Burch et al , 2008).…”

Section: Discussionmentioning

confidence: 99%

A preliminary evaluation of metadata records machine translation

Chen

Ding²,

Jiang³

et al. 2012

The Electronic Library

View full text Add to dashboard Cite

Purpose of this paperThe purpose of this study is to evaluate freely available machine translation (MT) services' performance in translating metadata records. Design/methodology/approachRandomly selected metadata records were translated from English into Chinese using Google, Bing, and SYSTRAN Machine Translation (MT) systems. These translations were then evaluated using a five point scale for both Fluency and Adequacy. Missing Count (words not translated) and Incorrect Count (words incorrectly translated) were also recorded. FindingsConcerning both Fluency and Adequacy, Google and Bing's translations of more than 70% of test data received scores equal to or greater than three, representative of 'non-native Chinese' and 'much coverage,' respectively. SYSTRAN scored lowest in both measures. However, these differences were not statistically significant. A Pearson correlation analysis demonstrated a strong relationship (r= .86) between Fluency and Adequacy. Missing Count and Incorrect Count strongly correlated with Fluency and Adequacy. Research limitations/implicationsThis study was conducted in a specific domain with a small sample size. It is necessary to conduct the evaluation with a larger, more representative test dataset. Also, other language pairs should be evaluated applying similar technologies. Originality/valueMost existing digital collections can be accessed in English alone. Few digital collections in the United States support multilingual information access (MLIA) that enables users of differing languages to search, browse, recognize and use information in the collections. Human translation is one solution, but it is neither time nor cost effective for most libraries. This study serves as a first step to understand the 2 performance of current MT systems and to design effective and efficient MLIA services for digital collections.

show abstract

Section: Discussionmentioning

confidence: 99%

A preliminary evaluation of metadata records machine translation

Chen

Ding²,

Jiang³

et al. 2012

The Electronic Library

View full text Add to dashboard Cite

show abstract

“…Combinations of MT systems into multi-engine architectures have a long tradition, starting perhaps with (Frederking and Nirenburg, 1994). Multiengine systems can be roughly divided into simple Figure 1: Architecture for multi-engine MT driven by a SMT decoder architectures that try to select the best output from a number of systems, but leave the individual hypotheses as is (Tidhar and Küssner, 2000;Akiba et al, 2001;Callison-Burch and Flournoy, 2001;Akiba et al, 2002;Nomoto, 2004;Eisele, 2005) and more sophisticated setups that try to recombine the best parts from multiple hypotheses into a new utterance that can be better than the best of the given candidates, as described in (Rayner and Carter, 1997;Hogan and Frederking, 1998;Bangalore et al, 2001;Jayaraman and Lavie, 2005;Matusov et al, 2006;Rosti et al, 2007).…”

Section: Architectures For Multi-engine Mtmentioning

confidence: 99%

Multi-engine machine translation with an open-source decoder for statistical machine translation

Chen

Eisele

Federmann

et al. 2007

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

View full text Add to dashboard Cite

We describe an architecture that allows to combine statistical machine translation (SMT) with rule-based machine translation (RBMT) in a multi-engine setup. We use a variant of standard SMT technology to align translations from one or more RBMT systems with the source text. We incorporate phrases extracted from these alignments into the phrase table of the SMT system and use the open-source decoder Moses to find good combinations of phrases from SMT training data with the phrases derived from RBMT. First experiments based on this hybrid architecture achieve promising results.

show abstract

“…The increasing availability of MT engines and the need for better quality has motivated considerable efforts to combine multiple engines into one "super-engine" that is hopefully better than any of its ingredients, an idea pionieered in (Frederking and Nirenburg, 1994). So far, the larger group of related publications has focused on the task of selecting, from a set of translation candidates obtained from different engines, one translation that looks most promising (Tidhar and Küssner, 2000;Akiba et al, 2001;Callison-Burch and Flournoy, 2001;Akiba et al, 2002;Nomoto, 2004). But also the more challenging problem of decomposing the candidates and re-assembling from the pieces a new sentence, hopefully better than any of the given inputs, has recently gained considerable attention (Rayner and Carter, 1997;Hogan and Frederking, 1998;Bangalore et al, 2001;Jayaraman and Lavie, 2005).…”

Section: Motivation and Long-term Perspectivementioning

confidence: 99%

First steps towards multi-engine machine translation

Eisele

2005

Proceedings of the ACL Workshop on Building and Using Parallel Texts - ParaText '05

View full text Add to dashboard Cite

We motivate our contribution to the shared MT task as a first step towards an integrated architecture that combines advantages of statistical and knowledge-based approaches. Translations were generated using the Pharaoh decoder with tables derived from the provided alignments for all four languages, and for three of them using web-based and locally installed commercial systems. We then applied statistical and heuristic algorithms to select the most promising translation out of each set of candidates obtained from a source sentence. Results and possible refinements are discussed.

show abstract

Learning to select a good translation

Cited by 7 publications

References 11 publications

A preliminary evaluation of metadata records machine translation

A preliminary evaluation of metadata records machine translation

Multi-engine machine translation with an open-source decoder for statistical machine translation

First steps towards multi-engine machine translation

Contact Info

Product

Resources

About