Turn on the unicode properties (*UCP) so \b becomes unicode-aware. It's seeing those characters as non-word boundaries of some sort, otherwise.
Code:
(*UCP)\s(?=([st]|re|ve|ll)\b)
I used this text as a test case:
Code:
<p>a séance töten don t</p>
<p>don tyou see sheriff s</p>
<p>we ll I'll be a mönkey s uncle</p>