The timestamped chunks make citation much more useful, and I would keep the caption track identity with them as well. Creator captions, automatic captions and translated captions can give materially different text for the same video and time range.
Store the selected language, track type, video ID and extraction version alongside start and end, then expose the language in a citation when it differs from the spoken audio. A successful fetch in the 81/82 check establishes availability, but a small multilingual audit is still needed to establish that the selected track preserves names, numbers and negation accurately.