A link (probably other elements too) followed by "'s" causes the token generation to fail. E.g. the utterance:
<p><a href="/w/index.php?title=Bob_Johnston&action=edit&redlink=1" class="new" title="Bob Johnston (page does not exist)">Bob Johnston</a>'s</p>
only creates token for "Bob".
Description
Description
Related Objects
Related Objects
Event Timeline
Comment Actions
The more general problem was handling a token that is broken up by a cleaned tag. E.g. from the original HTML:
tok<hr />en
Comment Actions
I have a solution that should fix this, but it will need some work after T148622: Highlight recited sentence, since it still uses positions and not paths.
Comment Actions
I can't reproduce this with the current revision (aca4abacc4a41145d3f30ba5434416aed4018896) on the test wiki. The reason for this is that the tokens are just added without looking in the original text nodes.