This is idea 4 of 4 concerning premiere captioning and transcription.
Idea:
access to per-word timecode data. The fact that clicking a word in the transcription jumps the playhead to that point in the timeline perfectly tells you that per-word timecode data is indeed captured by Premiere! But you can't actually access it as an end user. "Export to CSV" gives you timecoded BLOCKS of text. For scripting and automation purposes, having access to frame-accurate, per-word transcription timecode would be very powerful.
API exposure to this data would also be fantastic.