Experiments in discriminating phrase-based translations on the basis of syntactic coupling features

Abstract

We describe experiments on discriminating English to French phrase-based translations through the use of syntactic "coupling" features. Using a robust rule-based dependency parser, we parse both the English source and the French translation candidates from the nbest list returned by our phrase-based system; we compute for each candidate a number of coupling features, that is, values that depend on the amount of alignment between edges in the source and target structures, and discriminatively train the weights of these coupling features. We compare different feature combinations. Although the improvements in terms of automatic measures such as Bleu and Nist are inconclusive, an initial human assessment of the results appears to show certain qualitative improvements.

Domain Adaptation of Statistical Machine Translation Models with Monolingual Data for Cross Lingual Information Retrieval

“…We propose integrating the IR metric (MAP) into the translation model optimisation step via the reranking framework. Previous attempts to apply the reranking approach to SMT did not show significant improvements in terms of MT evaluation metrics Nikoulina and Dymetman, 2008). One of the reasons being the poor diversity of the Nbest list of the translations.…”

“…• features relying on dependency structures: called therein coupling features (proposed by (Nikoulina and Dymetman, 2008));…”

“…In this work we experiment with two different coupling variants proposed in (Nikoulina and Dymetman, 2008), namely, Lexicalised and Label-specific coupling features.…”