Using paraphrases for parameter tuning in statistical machine translation

Abstract

Most state-of-the-art statistical machine translation systems use log-linear models, which are defined in terms of hypothesis features and weights for those features. It is standard to tune the feature weights in order to maximize a translation quality metric, using held-out test sentences and their corresponding reference translations. However, obtaining reference translations is expensive. In this paper, we introduce a new full-sentence paraphrase technique, based on English-to-English decoding with an MT system, and we demonstrate that the resulting paraphrases can be used to drastically reduce the number of human reference translations needed for parameter tuning, without a significant decrease in translation quality.

Paraphrasing Training Data for Statistical Machine Translation

“…Compared to Callison-Burch et al (2006), Madnani et al (2007), or Nakov (2008) we are very conservative in our paraphrasing, and this is probably why we get a slightly lower improvement in quality. We could do more extravagant paraphrasing, but would have to retrain our HPSG parser's generation model to effectively rank the new lexical paraphrases.…”

“…Madnani et al (2007) obtained English language paraphrases by identifying paraphrases using a pivot language as in Callison-Burch et al (2006) and produced sentence-level English paraphrases by training an English→English hierarchical SMT system (Chiang 2005). Experiments showed that paraphrasing the tuning data used for MERT in a Chinese→English hierarchical SMT system performed competitively with increasing human references.…”

Paraphrasing Revisited with Neural Machine Translation

“…Extensions include representing paraphrases via rules obtained from a synchronous context free grammar (Ganitkevitch et al, 2011;Madnani et al, 2007) as well as labeling paraphrases with linguistic annotations such as CCG categories (Callison-Burch, 2008) and partof-speech tags (Zhao et al, 2008).…”

Improving Statistical Machine Translation with a Multilingual Paraphrase Database

“…Sentence level paraphrasing has been used for generating alternative reference translations (Madnani et al, 2007;Kauchak and Barzilay, 2006), or augmenting the training data with sentential para-phrases (Bond et al, 2008;Nakov, 2008;Mirkin et al, 2009). Phrase level paraphrasing was done using crowdsourcing or by using paraphrases in lattice decoding (Onishi et al, 2010;Du et al, 2010).…”