An unmerger of two formerly merged phonemes is a sound change that
restores the former distinction. The distinction between two sounds,
to recapitulate the discussion of page ff,
involves not just two pronunciations, or sounds, but also two sets of
words, where the first sound occurs in words of the first set, and the
second occurs in the second set. An unmerger, in a dialect which
merged two sounds, would therefore require the reintroduction of, on
the one hand, the distinction between the two pronunciations, and on
the other, the distinction between the two word classes. That is, if
X and Y merge into one sound, then to unmerge them, the dialect must
not merely acquire the two phonetic targets x and y: The two phonetic
forms must also be associated with the original sets of words which
originally contained them. That is, an unmerger requires that the
descendant reflexes of merged phonemes A and B are sorted out again
into the original word classes, so that the new pair of sounds is
pronounced in the correct set of words.
Unmerger is impossible. That is, it does not occur. Counterexamples to the principle that mergers are irreversible are false. Besides near-mergers, which are discussed below, the only well-documented counterexample is discussed in Haeri (1991), from which the following discussion is taken.