But when I try with a Turkish dictionary, it does work. Then the
dotless-i is indeed part of the word. Probably because the tr.dat has
"charset iso8859-9", while en.dat has "charset iso8859-1". I didn't look
in the dat-files before. But this is a bit silly; I am using UTF-8!
Is this artifacts of old non-utf8 Aspell; it still needs to tie a
(narrow) character set to the dictionary it is spelling with?

Could all this really be expected behaviour?
(Moreover, I am not using any user dictionary.)