Cross-Lingual Phonetic Distance Learning via Transliteration Pairs

In this study, a method has been developed to measure phonetic distances between text pairs from languages that use different writing systems. The Wagner-Fischer algorithm has been applied based on the distance cost of cross-lingual letter pairs in order to extract the letter alignment that minimizes the distance between two texts. The cost matrix, whose initial values are based on letter co-occurrences, is learned iteratively with expectation maximization through optimal alignment of transliteration pairs. The effectiveness of the proposed distance measure was demonstrated by evaluating its accuracy in a transliteration extraction task for test queries composed of Latin letters, in combination with various phonetic encoding techniques.