other,5-1-P97-1017,ak </term> . It is challenging to translate <term> names </term> and <term> technical </term> terms across
other,7-1-P97-1017,ak to translate <term> names </term> and <term> technical </term> terms across <term> languages </term>
other,10-1-P97-1017,ak <term> technical </term> terms across <term> languages </term> with different <term> alphabets </term>
other,13-1-P97-1017,ak <term> languages </term> with different <term> alphabets </term> and <term> sound inventories </term>
other,15-1-P97-1017,ak different <term> alphabets </term> and <term> sound inventories </term> . These <term> items </term> are commonly
other,1-2-P97-1017,ak <term> sound inventories </term> . These <term> items </term> are commonly transliterated , i.e.
other,11-2-P97-1017,ak i.e. , replaced with approximate <term> phonetic equivalents </term> . For example , computer in <term>
other,5-3-P97-1017,ak </term> . For example , computer in <term> English </term> comes out as ~ i/l : : : ' = --
other,23-3-P97-1017,ak : : ' = -- ~ -- ( konpyuutaa ) in <term> Japanese </term> . Translating such <term> items </term>
other,2-4-P97-1017,ak Japanese </term> . Translating such <term> items </term> from <term> Japanese </term> back to
other,4-4-P97-1017,ak Translating such <term> items </term> from <term> Japanese </term> back to <term> English </term> is even
other,7-4-P97-1017,ak </term> from <term> Japanese </term> back to <term> English </term> is even more challenging , and of
other,20-4-P97-1017,ak practical interest , as transliterated <term> items </term> make up the bulk of <term> text phrases
other,26-4-P97-1017,ak <term> items </term> make up the bulk of <term> text phrases </term> not found in <term> bilingual dictionaries
lr,31-4-P97-1017,ak <term> text phrases </term> not found in <term> bilingual dictionaries </term> . We describe and evaluate a method
tech,8-5-P97-1017,ak evaluate a method for performing <term> backwards transliterations </term> by machine . This method uses a <term>
model,4-6-P97-1017,ak </term> by machine . This method uses a <term> generative model </term> , incorporating several distinct
tech,13-6-P97-1017,ak incorporating several distinct stages in the <term> transliteration process </term> . We present an <term> algorithm </term>
hide detail