Using Hunspell 1.2.11 (but 1.2.14 seems to show the same problem)
When run in pipe mode (-a option) hunspell seems to count UTF-8 chars as a sum of unibyte chars instead as of a single multibyte char. That is causing problems under Emacs.
With attached file,
$ cat test-utf8-shift.txt | hunspell -d en_US -i utf-8 -a | grep ^\&
& Feedbooks 8 24: Feed books, Feed-books, Feedbacks, Feedback, Feedbags, Studbooks, Feedbag, Letterbox
when it should be (showing aspell behavior for comparison)
$ cat test-utf8-shift.txt | aspell --encoding=utf-8 -d en_US -a | grep ^\&
& Feedbooks 6 22: Feed books, Feed-books, Feedback's, Feedbags, Feedbag's, Feedback
Note that the conflicting UTF-8 apostrophe is three byte multi-byte char, thus causing the two characters shift.