I said, “Why not add some simple captions to Mallard?” Philip said “Why not embed an existing format, like TTML?” Philip was so right.
<media type="video" mime="application/ogg" src="figures/gnotravex-video.ogv">
<p>Simple demonstration of a game</p>
<tt:tt xmlns:tt="http://www.w3.org/ns/ttml">
<tt:body>
<tt:div begin="1s" end="7s">
<tt:p>Drag pieces from the right to the left, making sure that
adjacent edges have the same number<em><tt:span begin="1s"> and
color</tt:span></em>.</tt:p>
</tt:div>
<tt:div begin="6.5s" end="14s">
<tt:p>Press <keyseq><key>Ctrl</key>arrow keys</keyseq>
to move all the placed pieces at once.</tt:p>
</tt:div>
</tt:body>
</tt:tt>
</media>
A nice bonus feature is the tt:span element. One second after the first caption is displayed, some extra text appears inside it.
We should host a TTML profile for Mallard+TTML on projectmallard.org. Authors concerned with interchange would then be able to specify that profile using ttp:profile to prevent non-Mallard TTML processors from choking up on the inline Mallard syntax.

What Phil* wants, Phil* gets.