See also http://www.microsoft.com/typography/developers/opentype/detail.htm which explains how ligatures can be turned off on a language-dependent basis.
Laurentiu Peter Kirk asked: > In Turkish and Azeri the sequences f - i and f - dotless i both occur, > and are fairly frequent. So it is inappropriate in these languages to > use fi ligatures in which the dot on the i is lost or invisible, at > least where the second character is a dotted i. Has any thought been > given to this issue? Is it possible to block such ligation on a > language-dependent basis?