Lexical Word Similarity for Re-ranking in Vietnamese-English Named Entity Back Transliteration

Transliteration is the transformation of word in original language to another language based on its pronunciation. Back transliteration is the transformation of already transliterated word in another language back to its original form. This backward process is in nature more challenging than the forward direction because of more information lost. In many cases, the back transliteration can return almost exact result, which has a minor difference in spelling compared with the original word form. We propose in this work a lexical word similarity for dictionary matching in order to re-rank the candidates and enhance the performance of a grapheme-based location name back transliteration. This method is experimented on Vietnamese-English language pair and showed improvement.