@article{,
title = {Language-independent classifier-based modelling of source-side context information in Statistical Machine Translation (Data set)},
journal = {},
author = {Maarten van Gompel and Antal van den Bosch},
year = {},
url = {},
license = {},
abstract = {  We present a series of experiments focusing on the modelling of
  source-side context to improve Phrase-based Statistical Machine
  Translation. Statistical Machine Translation systems typically
  consist of a translation model and a language model. The former maps
  phrases in the source language to the target language, without
  regard for the context in which the source phrases occur. The latter
  models just the target language, and acts as a target-side model of
  context information after translation. We attempt to independently
  reproduce a line of existing research and test whether considering
  context information directly in the translation model has a positive
  effect on translation quality.  We furthermore investigate various
  ways discriminative classifier-based models can be integrated into
  Statical Machine Translation.  We will use proven techniques from
  Word Sense Disambiguation, effectively integrating these techniques in
  Statistical Machine Translation. Our approach is
  language-independent and knowledge-poor: we do not employ any
  explicit linguistic features computed by part-of-speech taggers,
  word sense disambiguation systems, supertaggers, or parsers, as used
  by previous work. We find only limited improvement of translation quality for
  certain formulaic corpora and conclude that explicit modelling of source-side
  context information does not add much to the data already implicitly
  available in the decode process.}
}