| 摘要 |
Methods, systems, and apparatus, including computer program products, for ranking the output suggestions of a transliteration process. In one aspect, a method includes receiving a plurality of transliteration training pairs comprising a source word in a first writing system and a target word in a second writing system. A plurality of 1-to-N transliteration rules is received. Blocks are generated for each source word character comprising the source word character and a string characters preceding and following the source word character. Generated blocks are associated with the right-hand-side of respective 1-to-N transliteration rules to create a plurality of block rules. Block rules are used to generate a possibly-partial transliteration of a source word. A score is assigned based on an edit distance between the source word and each of one or more respective transliteration output suggestions. |
