Multiple System Combination for Transliteration

Abstract

We report the results of our experiments in the context of the NEWS 2015 Shared Task on Transliteration. We focus on methods of combining multiple base systems, and leveraging transliterations from multiple languages. We show error reductions over the best base system of up to 10% when using supplemental transliterations, and up to 20% when using system combination. We also discuss the quality of the shared task datasets.

Regulating Orthography-Phonology Relationship for English to Thai Transliteration

“…A 6-gram joint sourcechannel model was also used for this augmented system. Despite a simpler setup due to exploiting phonology knowledge, this system achieves comparable performance to that of systems reported in (Nicolai et al, 2015) and (Finch et al, 2015) for English to Thai transliteration task.…”

Comparison of Assorted Models for Transliteration

“…DIRECTL+ is a publicly available discriminative string transduction tool 1 , which was initially developed for grapheme-to-phoneme conversion (Jiampojamarn et al, 2008). Previous University of Alberta teams have successfully applied DI-RECTL+ to transliteration in the previous editions 1 https://code.google.com/archive/p/directl-p of the NEWS shared task (Jiampojamarn et al, 2009(Jiampojamarn et al, , 2010Bhargava et al, 2011;Kondrak et al, 2012;Nicolai et al, 2015). We apply M2M-aligner (Jiampojamarn et al, 2007) to align the sourcetarget pairs before training.…”

Leveraging Entity Linking and Related Language Projection to Improve Name Transliteration

“…Scores of DTL (DirecTL+), SEQ (Sequitur), and SMT (statistical machine translation) are reported in (Nicolai et al, 2015). Scores of various data representation methods, namely P (character), M (character+boundary marker), T (bigram), and M+T (bigram+boundary marker), are reported in (Kunchukuttan and Bhattacharyya, 2015) .…”