Purest ever example-based machine translation: Detailed presentation and assessment

Abstract

Abstract. We have designed, implemented and assessed an EBMT system that can be dubbed the 'purest ever built': it strictly does not make any use of variables, templates or patterns, does not have any explicit transfer component, and does not require any preprocessing or training of the aligned examples. It only uses a specific operation, proportional analogy, that implicitly neutralises divergences between languages and captures lexical and syntactical variations along the paradigmatic and syntagmatic axes without explicitly decomposing sentences into fragments. Exactly the same genuine implementation of such a core engine was evaluated on different tasks and language pairs. To begin with, we compared our system on two tasks of a previous MT evaluation campaign to rank it among other current state-of-the-art systems. Then, we illustrated the 'universality' of our system by participating in a recent MT evaluation campaign, with exactly the same core engine, for a wide variety of language pairs. Finally, we studied the influence of extra data like dictionaries and paraphrases on the system performance.Keywords: Example-based machine translation, proportional analogies, divergences across languages.Abbreviations: MT -machine translation; EBMT -example-based machine translation; mWER -multiple word error rate * The research reported here was supported in part by a contract with the Japanese National Institute of Information and Communications Technology entitled "A study of speech dialogue translation technology based on a large corpus". We are particularly indebted to Prof. C. Boitet for his many comments on an earlier version of the draft that considerably helped to improve clarity. Thanks also to the reviewers who pointed out some errors in the draft. Both authors are currently with the Japanese National Institute of Information and Communications Technology (NiCT).

Entry Generation by Analogy – Encoding New Words for Morphological Lexicons

“…More power is probably required only in higher-level NLP tasks, e.g. syntactic analysis and translation as introduced by Lepage and Denoual (2005).…”

“…In what can be characterized as a surface approach to compositional semantics, Lepage and Denoual (2005) present a machine translation system between English, Japanese, Chinese and Korean and demonstrate that the only thing needed by machine translation between these languages is analogy and a sufficiently large bilingual training corpus. They rely on the aligned sentences of the training corpus to be semantically equivalent.…”

Inflating a Small Parallel Corpus into a Large Quasi-parallel Corpus Using Monolingual Data for Chinese-Japanese Machine Translation

“…In Ref. [22], they use proportional analogies to translate sentences in an examplebased machine translation. Translation of unknown words by analogy has also been proposed in Refs.…”

Empirical lower bounds on alignment error rates in syntax-based machine translation

“…English-Chinese (Wellington et al, 2006) and English-Spanish (Lepage and Denoual, 2005). While our main focus is on configurations that involve discontinuous translation units, the frequencies of inside-out alignments in our parallel corpora are also reported.…”