Machine Transliteration deals with the conversion of text strings from one orthography to another, while preserving the phonetics of the strings in the two languages. Transliteration is an important problem in machine translation or crosslingual information retrieval, as most proper names and generic iconic terms are out-of-vocabulary words, and therefore need to be transliterated. There are numerous methods explored in the literature for machine transliteration, ranging from rule-based techniques to statistical learning techniques. Here we focus our attention on language-independent techniques that potentially can scale well with a large number of languages. In this paper, we present a modular, statistical learning framework that lends itself for easy experimentation with transliteration tasks between a variety of different languages, in a language-independent manner. The workbench includes a variety of components – algorithms, data-sets and transliterations scripts – for a quick assembly of an effective transliteration system across langauges. We believe that such workbenches would be important in an increasingly multilingual world, for building systems that span a number of languages, quickly and effectively.
[1]
Yaser Al-Onaizan,et al.
Machine Transliteration of Names in Arabic Texts
,
2002,
SEMITIC@ACL.
[2]
Sanjeev Khudanpur,et al.
Transliteration of Proper Names in Cross-Lingual Information Retrieval
,
2003,
NER@ACL.
[3]
Leah S. Larkey,et al.
Statistical transliteration for english-arabic cross language information retrieval
,
2003,
CIKM '03.
[4]
Robert L. Mercer,et al.
The Mathematics of Statistical Machine Translation: Parameter Estimation
,
1993,
CL.
[5]
Jian Su,et al.
A Joint Source-Channel Model for Machine Transliteration
,
2004,
ACL.