In addition to the natural language understanding, the ingested catalog content is used to help train automated speech - recognition (ASR) models to translate spoken word into the correct text,
particularly for more challenging linguistic translations such as principal and principle, which are
homophones.