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.