This talk, using real-world examples, examines the common pitfalls of translating application prompts. Strategies for successful translation into multiple languages and cultures are presented, such as creation and maintenance of the integrity of source text, consideration of cultural elements of both source and target language, selecting the best-qualified translator, why a single translator is preferred, how and when to use reverse translations, a glossary of company/industry terms for use by the translator, and concatenation of sentence fragments.

The localization of speech recognition for large-scale spoken dialogue systems can be a tremendous exercise. Usually, grammars are translated by a language expert, and new data is collected, transcribed, and annotated for statistical utterance classifiers — resulting in a time-consuming and expensive undertaking. Typically, though, a vast number of transcribed and annotated utterances exist for the source language. This talk proposes to translate such data via machine translation and use them along with their associated (original) annotations to train statistical grammars.