Back in February Matt Weiner wrote a nice little paper with three examples designed to show that not all implicatures are cancellable. Unfortunately I didn’t know it was a nice little paper at the time, since I didn’t actually get around to reading it. (This is particularly embarassing since it is only six pages.)
However, it has now been accepted to Analysis, and this time I did read the paper. I liked it so much that instead of writing up my commentary for next weekends WCPA (sorry Kevan, I’ll do it tomorrow) I spent the afternoon writing up this little comment on what I think is going on here (link excised).
Edit: updated the comment a little, link similarly updated.
Edit: removed link as paper is heavily under revision now.