Citation Context

...ch yields better and hence only those results are reported in the paper. 4 Experimental Configuration We trained 15 systems over three domains - tourism, health and general. We use the MOSES toolkit (=-=Koehn et al., 2007-=-) for our PB-SMT system. We use the grow-diag-final-and heuristic for extracting phrases and the msd-bidirectionalfe model for lexicalized reordering. We tuned the trained models using Minimum Error R...

Citation Context

... were trained using the Leipzig Corpus 5 and the monolingual data provided for the competition. We used a 5-gram language model built using SRILM (Stolcke and others, 2002) with Kneser-Ney smoothing (=-=Kneser and Ney, 1995-=-) with 1.5 million sentences from WMT monolingual corpus. The evaluation was done using the BLEU (Papineni et al., 2002) and METEOR (Banerjee and Lavie, 2005) metrics. For METEOR, the synonyms were ob...

Citation Context

...ering in a small window is sufficient. We use a rule based approach which applies reordering transformations to the English sentence parse tree to generate the HoE sentences (Ramanathan et al., 2008; =-=Patel et al., 2013-=-). Although these rules were primarily developed for Hindi, they work for translation fromEnglish to any Indian languages since the the structural divergences are similar (Kunchukuttan et al., 2014). ...

Citation Context

...l., 2008; Patel et al., 2013). Although these rules were primarily developed for Hindi, they work for translation fromEnglish to any Indian languages since the the structural divergences are similar (=-=Kunchukuttan et al., 2014-=-). 3.2 Source-Side Word-Segmentation Morphologically-rich/agglutinative languages have low token-to-type ratio and can generate a potentially infinite number of word types by combining morphemes. Due ...