LINGUIST List 3.551

Fri 03 Jul 1992

Disc: Phonology

Editor for this issue: <>


  1. David Stampe, 3.546 Dissimilation Summary
  2. David Stampe,

Message 1: 3.546 Dissimilation Summary

Date: Fri, 3 Jul 92 04:54:09 HST3.546 Dissimilation Summary
From: David Stampe <stampeuhunix.uhcc.Hawaii.Edu>
Subject: 3.546 Dissimilation Summary

Thanks to Jose Ignacio Hualde <> for the
summary of responses to Jennifer Cole and Chuck Kisseberth's earlier
request for examples of dissimilations. It would be nice if the full
detail of the responses could be put on the LINGUIST server.

But I have some questions about Hualde's conclusions. He says

 we have not received any examples of true long-distance
 dissimilations of the type that /milit-alis/-> [militaris] would
 exemplify were this a regular synchronic rule; that is, between
 segments which are not in adjacent syllables.

Changes in non-adjacent syllables as in militaris, lunaris, etc., may
be sporadic, but they are obviously reflections of the same process
that applies in adjacent syllables as in polaris, similaris, vulgaris.
The former are surely long-distance dissimilations, regular or not!

Hurch is quoted as saying that "productivity gradually decreases the
more distant the [dissimilator] is from the [dissimilatee]". Good.
But if adjacency favors dissimilation, why doesn't ll dissimilate in
mellis (not *melris)? Because of the Geminate Integrity Principle?
(But that's not an answer, it's just a fancy name for the question.)
The point of examples like "kisse[s] > kisse[z]" and Grassmann's or
Dahl's Law is that the dissimilator and the dissimilatee are close
together but MAY NOT BE adjacent. So how can we accept the conclusion
that "these cases can be reformulated as respecting adjacency with
some relaxation of this notion to include, e.g., syllable-adjacency"?
If the sounds are actually adjacent, the dissimilations don't apply!

Another puzzle: Dissimilations, like other fortitions, apply in
narrower prosodic domains only if they apply in wider domains. For
example, if a dissimilation like uw > iw applies at syllable peaks (as
in Old French) it also applies at the peak of feet (that is when
accented); but not the reverse -- early English uw > ow applied in
accented but not in unaccented position.

This is the opposite of assimilations, which, like all lenitions,
apply in wider prosodic domains only if they apply in narrower ones.
For example, if iw mutually assimilates to y(y) at the head of a foot
(i.e. iff it's accented), it will do so also at the head of a syllable
(i.e. regardless of whether it's accented). [Here I'm using head in
the dependency sense: the sonority peak of a prosodic domain.]

These apparent paradoxes were resolved in Donegan's 1978 thesis, On
the Natural Phonology of Vowels. She pointed out that if fortitions
precede lenitions, and if sounds are more likely to assimilate the
more similar they are (cf. Zwicky, Hutcheson, Hankamer & Aissen), then
clearly dissimilations can block assimilations. If dissimilation
blocks assimilation, then it's more likely in sounds that are more
adjacent, since more adjacent sounds are more likely to assimilate.
On the other hand, since dissimilation per se is limited to wider
domains, such as phrases or feet, then it's more likely in sounds that
are separated by major prosodic boundaries, like the rise and fall
(nucleus and offglide) of a foot. Thus in early modern English
accented ey as in _white_ (< iy < i:) dissimilated to ay /hwayt/, and
thereby escaped the fate of unaccented ey in profundity (<
profundite), which assimilated to iy (> i). Actually some early
dialects optionally dissimilated unaccented ey as well, so that lady
was sometimes rhymed with eye.

In nonadjacent dissimilation, there is this difference: the sound that
is threatened by assimilation is BETWEEN the similar sounds. As e.g.
O. Jesperson noted, a sound may assimilate to like flanking sounds and
yet not assimilate to the same sounds if they only precede or follow:
prob[schwa]bly -> prob[b]bly, but not cred[schwa]ble -/-> *cred[b]ble,
and not ib[schwa]dem -/-> *ib[b]dem. In the example I cited of the
English -(e)s affixes becoming -(e)z, the unaccented vowel in words
like kisses was in danger of assimilating to the flanking voiceless
sounds, thus becoming completely inaudible (kissss), and this result
was blocked by dissimilating voicelessness in the final s. The same
dissimilation occurred in missu[z] (Mrs.) beside mistress, and in a
common pronunciation of Missi[z]sippi, which otherwise often loses the
s-flanked vowel in quick speech: Miss'sippi.

The dissimilation in Bantu of voiceless consonants separated by a
(short?) vowel, called Dahl's Law, may have had the same function of
blocking the devoicing of the vowel. And Grassmann's Law, ChVCh >
CVCh, blocked aspiration of the vowel in Greek and Sanskrit. And in
many languages there are constraints against CVC roots where the C's
are too similar. The motivation for dissimilation here is clear from
many languages that lose (i.e. completely assimilate) vowels between
consonants that are too similar.

If "distant" dissimilations of consonants protect the sounds they
flank from complete assimilation, we wouldn't expect dissimilation
either (1) if the flanked sounds are so many or so different from the
flanking consonants as to resist assimilation, or (2) if there are no
flanked sounds. Therefore,

(1) we would expect l...l dissimilation to be most needed where a
 single vowel would be flanked by l's (polaris) or maybe even a
 sequence of vowels (familiaris) or a sequence of vowels and
 consonants similar to the flanking l's, e.g. dentals (lunaris,
 plantaris), because the entire sequence might be assimilated.
 With velars or labials (localis, globalis) this is less likely,
 or with long sequences of flanked sounds (fluvialis).

(2) we would NEVER expect dissimilation in mellis (*melris), where
 there is no flanked sound to protect from assimilation. (The only
 other reason for dissimilating two sounds is to keep them from
 assimilating to each other, but ll can't get any more similar!)

I doubt whether the forms, domains, or roles of dissimilations can be
specified except in terms of their phonetic functions. On function
and form, see Donegan's thesis (OSUWPL 23, 1978, or the Garland
reprint, and the insightful works of Grammont and Fouch\'e she cites).
On their domains, cf the paper by Donegan & Stampe in A. Bell & J.
Bybee Hooper, eds, Syllables and Segments, Amsterdam 1978. On their
role in derivations, see the latter, and also our paper in D. Dinnsen,
ed., Current Approaches to Phonological Theory, Bloomington 1979. On
their special role in phonological perception, see Donegan's paper in
a forthcoming Longman's volume on linguistic change edited by C. Jones.

David Stampe <>, <stampeuhunix.bitnet>
Dept. of Linguistics, Univ. of Hawaii/Manoa, Honolulu HI 96822
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 2:

Date: Thu, 2 Jul 92 11:19:30 CDT
From: David Stampe <stampeuhunix.uhcc.Hawaii.Edu>
<From: Eric Schiller <>

Re: Natural Phonology
Perhaps John Coleman should review the literature (e.g., the Natural
Phonology Parasession at CLS 1974) and discuss particular analyses
with which he disagrees and which point to the failing of the theory.

As someone who is still shopping for a decent theory of phonology,
I would like to see a bit more in the way of specifics.

Eric Schiller
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue