I think I figured this out for myself, but for anyone who is interested you can initialize a Reco context with the Dictation Grammar and then call EmulateRecognize() and pass the word you want to query for. It returns a RecognitionResult object when will be NULL if the word is unknown...
If you get back a RecognitionResult object then it will have the pronunciation available in it...