LINGUIST List 9.1171

Fri Aug 21 1998

Sum: Phonological Clusters in Similar Words

Editor for this issue: Martin Jacobsen <>


  1. Gareth Gaskell, Phonological Clusters of Semantically Similar Words

Message 1: Phonological Clusters of Semantically Similar Words

Date: Thu, 20 Aug 1998 12:10:20 +0100
From: Gareth Gaskell <>
Subject: Phonological Clusters of Semantically Similar Words

On 17th July I posted the following question to the linguist list:

There are fairly well known clusters of similar sounding words that
also mean similar things, such as glimmer, glisten and glint or
sneeze, snort and snore. Does anyone know of any work carried out on
these clusters? I would be particularly interested to find out about
research looking at the prevalence of these clusters in the lexicon,
or their effects on new word formation, but any references to
linguistic or psycholinguistic research would be most welcome.

This provoked an excellent response, thanks to the following people:

Robin Allott <>
Rachel Rowlands <>
Conrad Charles Treff <>
Alan C. L. Yu <>
Nancy Frishberg <>
Dr. Rebecca Letterman <>
Monika Bruendl <>
Daniel Currie Hall <>
John M. Lawler <>
Philip Grew <>
Jack Quintero <>
C. Walter <>
Margaret Magnus <>

Philip Grew has already posted a valuable reply to this question
(Linguist 9.1106), so I shall try to avoid overlap with his posting in
my summary.

Conrad Treff provided a review of the various names for this and
related phenomena: phonosymbolism, phonosemantics, sound symbolism,
iconicity and phonaesthesia all provide useful search terms for a
literature search.

Six people pointed me towards the excellent website devoted to sound
symbolism--Margo's magical letter page--which contains a number of
useful resources and links:

Philip Grew also noted the sound symbolism discussion list:

Send email to <> with the following
command in the body of your email message:

	subscribe soundsymbol

Other useful references are: [and other papers on
Robin Allot's web site]

Bolinger, D. Forms of English: Accent, Morpheme, Order (Harvard
University Press, Cambridge, Mass., 1965)

Hinton, L et al (Eds). Sound Symbolism (1994).

Jakobson, R. & Waugh, L. The sound shape of language.

Lehrer, Adrienne (1996), "Why neologisms are important to study",
Lexicology Vol 2/1,63-73.

Waugh, Linda R. (1994), "Degrees of iconicity in the lexicon", Journal
of Pragmatics 22, 55-70 

Waugh, Linda R. & Madeleine Newfield (1995), "Iconicity in the lexicon
and its relevance for a theory of morphology", in: Landsberg, Marge
E., ed. (1995), Syntactic iconicity and linguistic freezes. The human
dimension, Berlin - New York., 189-221.

Finally, John Lawler sent me a posting that he sent to sci.lang on the
subject. The remainder of this summary is his posting (slightly edited
to avoid duplications):

- ---------------------------------------------------

Newsgroups: sci.lang
Subject: Re: English initial consonantal bi-phonemes.
Organization: University of Michigan

Dwight Bolinger introduced the term "assonance" for what you're
talking about:
 Bolinger, Dwight. 1950 Rime, Assonance and Morpheme Analysis. Word
> There are in English at least, three clear-cut cases, possibly more,
> of bi-consonantal initial phonemes beginning words in large clusters
> with similar meanings, even though not (all) etymologically cognate.
> (Dave Librik) writes:
>>Yes, this is a known systemic pattern. 
>>The best introductory articles on this are:
>> Rhodes, Richard A. and John M. Lawler, "Athematic Metaphors",
>> _Proceedings of the Chicago Linguistics Society [CLS] 17_,
>> pp. 318-342, 1981.
>> Rhodes, Richard A., "Aural Images", in _Sound Symbolism_, ed.
>> by John Ohala, Leanne Hinton, and Johanna Nichols,
>> pp. 276-292, 1994.
>> The first paper lays out the shape-classificatory system underlying
>> English monosyllables like stick, strip, spine, flap, skin, nut,
>> bump, notch, slump, etc. The second builds on this to describe the
>> sound- classificatory system heard in peep, bang, plunk, click,
>> crunch, chirp, thunk, whiz, zing, etc. Both make for very
>> compelling reading, though the first is probably more impressive.
Thank you. One additional paper:
 Lawler, John M., "Women, Men, and Bristly Things: The
Phonosemantics of the BR- Assonance in English," in Beddor (ed.),
Michigan Working Papers in Linguistics I:1, 1990.
>> The titles of the papers are both somewhat misleading : the
>> clusters described here are essentially _sub-morphemes_. They
>> carry meaning but are not separable nor predictable; they occupy a
>> point on the continuum between arbitrariness and regular morphemes.
>> They're the sort of rudiments, probably found in all languages,
>> which have developed in some languages into regular, patterned
>> "classifier systems."
Precisely. The principal insight of this research has been that
English phonesthemes form a classifier system. Some references on
 Adams, K.L. and N.F. Conklin, 1973. Toward a Theory of Natural
 Classification. CLS 9:1-10.
 Allan, Keith. 1977. Classifiers. Lg. 53:285-311.
Becker, Alton L., 1973. A Linguistic Image of Nature: the Burmese
Numerative Classifier System. International Journal of the Sociology
of Language 5:109-121.
 Conklin, Nancy F., 1981. The Semantics and Syntax of Numerical
 Classification in Tai and Austronesian. (Unpublished Ph. D.
 dissertation, University of Michigan.)
Denny, J.P 1976. What are Noun Classifiers Good For? CLS 12:122-132.
> The three clusters I know of are:
> 1) The "sn-" group. There are hugely many of these, almost all
meaning something nasty.
> snake, snail, snarl, sneak, snatch, sneer, snide, etc etc; indeed a
> quick look at a dictionary shows them almost without exception to be
> nasty things. (The only clear counter-example seems to be snug.)
Pejorative senses are very common with these. The SN- assonance has 43
members in our monosyllable database, available on the Web at
Most of these fall into two groups, both based on the same classifier
model: 3-D Convex/Concave, with different body-part foci
 A: Nose
 snivel snoop snoot snooze sneer sniff sniffle snigger snot
 snout snarl snub snuff snob snicker sneeze snore snorkel snort
 B: Other, principally Grasping Hand
 snatch snag snug snuggle snaggle snack snitch snap
 snip snippet snood snick snare
This illustrates several common phenomena of assonance submorphemes:
a) there are usually several semantic groups in a class
b) these cohere partially, but not fully
c) they have embodied models (my new motto is 
 "Ontology recapitulates physiology" :-)
d) there's a lot of metaphoric extension, most often pejorative
e) there is residue (for SN-, the first-cut residue is snazzy snocker
snark snide snit sneak snipe snake snail snow snooker)
> 2) The "sl-" group. These often mean something flat or smooth, that
> could get into narrow places.
> slink, slime, slide, slab, slat, slender, slice, and so on.
That's a nice one. I analyzed this in responding to another sci.lang
post a while back. Here's a precis:
 Like BR- and PR- (cf Lawler 1990), one of the fields is physical and
 experiential, and the other is human and evaluative. Together they
 form a radial semantic category accessed by the assonance.
 SL-1) An experiential physical field, like that evidenced by
Classifier systems in many languages. In this case, it is a species
of Liquid classifier, differenced from DR-, the principal Liquid
assonance in English (drink, drown, drip, dry, drool, etc.) by
reference to Liquid interface with Solid.
 SL-1 Liquid/Solid Interface
 SemiLiquids Liquid/Solid
 -------- ----------------
 slush slug sluice
 sludge slog slobber
 slur slick slather
 slurry sleek slime
 slag slough slurp
 slop sloop slosh
 sleet slake
SL-2) An evaluative (therefore inevitably pejorative for the most
part) Human classifier; in this case, a radial category consisting of
several species of metaphoric application of category (1) above to
humans, their activities, their (fancied) properties, and their (low)
degree of desirability and tastefulness.
 SL-2 Terms of Pejoration non-Human
 Insults Actions Abuse Physical Properties Nouns
 ------- ------- ----- -------- ---------- -----
 slug slink slam slender slack slag
 slut slobber slander slight sleazy slop
 slob slog slang slim slick sludge
 slouch slosh slap slip sluggish slush
 slime slump sling slow sly slum
 slave slur sloth
 slope slurp
 slant sluff
> 3) The "gl-" group. Though not as widespread as the last two, there
>are still sufficiently many for it to be clearly beyond random
>chance. They often mean things only-just-seen.
> glimpse, glimmer, glance, glisten, glow, gloom, (and perhaps a
> little off-definitional) glare and glower.
There are 42, almost the same as SN-. Briefly (and roughly; I haven't
looked at these closely),
A: Reflected Light (Eye)
 glamor glance glare glass glaze gleam glimmer glimpse glint glisten
 glitter gloaming gloom gloss glow glower glum
B: "Sticky" (very loose, sorry)
 gland glob globe glom glop glub glue glug glut gluten glutton
 glad glade glaive glean glebe glee glen glib glide
 glitch gloat glory glove glyph
> Question 1
> ==========
> Do other languages have the same phenomenon? Presumably they
> couldn't occur in Sinic languages with their paucity of consecutive
> consonants; but what about Semitic and other IndoEuropean tongues?
>Peter Daniels <> writes:
>>They're an important part of the grammar of various Southeast Asian
>>languages, and Gerard Diffloth calls them ideophones, I think.
They're all over the map in Japanese, too, where they're usually
called "ideophones". Mark Rosenfelder points out some Quechua cases.
And Indonesian has them as well. A *very* thorough analysis of the
latter was done by Keith McCune, under Rich Rhodes and me:
McCune, Keith Michael. 1983. The Internal Structure of Indonesian
Roots (Vols I and II). Ph.D. Dissertation (Linguistics), University of
Michigan. Available from University Microfilms.
I would be very surprised if that were all.
> Question 2
> ==========
> Are there any cases in English other than the three given above?
Lots. Most initial CC- and CCC- clusters have senses (though some
don't; for instance FR- doesn't cohere, as far as I can tell).
Also, some rimes have senses, notably -UMP (3-D: rump, stump, lump,
hump, dump, slump, clump, etc.), -AP (2-D: flap, strap, map, slap,
wrap, lap, etc.), and some more. But most don't.
> Question 3
> ==========
> The biggy. Why are they there?
> What mechanism can have led to the appearance of this phenomenon?
> (Mark Rosenfelder) writes:
>>I don't know, but I'd speculate that analogy working on a few chance
>>instances could soon produce a fair number of such words.
And once it's produced, it will attract more. It becomes a stable
semantic lexical nexus (a Lexus? :-). Many of these can be traced
back to Old English and beyond, though few of the words in them are
the same now as they were then.
More ontogenetically, the hypothesis that words of a feather flock
together is a very obvious one, and children hit on it
immediately. The fact that it doesn't work most of the time eventually
leads to other strategies for vocabulary learning and storage, but --
and this is a point often overlooked -- these don't *replace* early
hypotheses so much as overlay them, and to the extent the simpler
hypotheses actually work, they're reinforced very strongly. They work
just fine with these classes, and this keeps the semantic clustering
>Neil Coffey <> writes:
>>You're probably making more of this "phenomenon" than there really
>>is to make of it. Consider the following:
>>(1) You say that there are sufficiently many words for it to be
>> "clearly beyond random chance". But what do you mean by
>> "random chance"? -- word formation is probably never random,
>> so is this really a good comparison to make?
>> And how do you know that there are 'sufficiently many' to
>> make any given comparison let alone this one?
This is the same kind of insoluble statistical problem that leads to
arguments like those on how likely it is that words in two unrelated
languages will have both the same shape and the same meaning (e.g,
English 'hole' vs Yucatecan 'ho:l'; Latin 'dua' vs Indonesian 'duwa',
etc). We can quantify the phonological search space, but not the
semantic one; we don't even know how many dimensions there might be.
>>(2) You've got to admit that some of your correlations are a bit 
>>arbitrary, and you happen to have chosen examples which fit well 
>>these descriptions -- there are probably just as many that don't. 
>>For example, "These often mean something flat or smooth, that could 
>>get into narrow places" hardly defines 'slob', 'slag', 'sleep', 
>>'slack', 'slum', 'slow', 'slur', 'slope' etc etc.
If you look at *all* the words in a class, you find that the assonance
classes that cohere semantically typically include between 70 and 80
percent of the words in the coherent class. This is so far beyond
*any* definition of "chance" that one needs no statistical measures to
see it. I'd add that this *doesn't* happen with random sets of words
-- the database produces those, too, subject to the whims of the
TurboPascal rand() function.
>>(3) Do you really think that "'sl' = flat or smooth" is part of
>> the intuition of native speakers of English?
I can't speak for Bill, but *I'd* say that the semantic
characterization of the SL- class in appropriate terms (*not* 'flat or
smooth') definitely is part of the intuition of native speakers of
English. It's clearly not arbitrary; what else could it be?
>>(4) The cluster [sC] seems more readily onomatopeic than other
>> clusters. One road you might go down is to see what percentage
>> of these words can be traced back to onomatopoeia.
There's significant onomatopoeia, of course, but that only applies to
words that *refer* to sounds. Rich Rhodes treated that system in the
article cited by Dave Librik above.
>>(5) Instead of trying to come up with one semantic correlation 
>>between every single word beginning with a given cluster, you might
>>look instead at correlations between smaller numbers of words
>>(e.g. 'slime'/'slush'/'sleet', 'slash'/'slay'/'slaughter' -- though
>>I think that 'slay' and 'slaughter' are derived from the same verb
>>-- this is one of the problems which your comparison with 'random
>>chance' seems to ignore).
That's right. As Dave points out (that we pointed out), it's not as
coherent and regular as a morpheme, but not as arbitrary as a phoneme,
either. It's intermediate between them, and the species of meaning it
displays is of a different kind, and refers to more rudimentary
There's lots more to say about this topic, but this is way too long

Dr. Gareth Gaskell
- -------------------------------------------------------------------
MRC Cognition and Brain Sciences Unit Fax: 01223 359062
15 Chaucer Road Phone: 01223 355294 xt 620
Cambridge CB2 2EF Email:
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue