Short answers, pulled from the story.
The first version of Microsoft's machine translation system was developed between 1999 and 2000 within Microsoft Research. It was based on logical forms derived from the grammar correction feature built for Microsoft Word.
Microsoft Translator supports 181 languages and language varieties. Of those, 76 are supported by the text-to-speech tool.
The BLEU score, short for Bilingual Evaluation Understudy, is the method Microsoft Translator uses to evaluate the quality of its machine translation output. It measures how closely the machine's translation matches that of a human, and was among the first automated metrics to correlate strongly with human quality judgments.
Neural machine translation became the default translation method for the Microsoft Translator API in May 2018. Deep neural networks had already been introduced for nine of the highest-traffic languages, including all speech languages and Japanese, in November 2016.
The Live feature is a real-time conversation translation tool that supports up to 500 people simultaneously across multiple devices, multiple languages, and in-person settings. It is free and available through the Microsoft Translator apps on Android, iOS, and Windows.
The Microsoft Translator Hub has been used by communities to build translation systems for minority and endangered languages, including Hmong, Mayan, Nepali, and Welsh. Partners such as the Welsh Parliament, the Government of Nunavut for Inuktitut, and two New Zealand universities for Maori have all contributed to the service.