LINGUIST List 7.1698

Sat Nov 30 1996

Sum: Lgs with obligatory onsets & without distinctive length

Editor for this issue: Ljuba Veselinova <>


  1. Robert Ratcliffe, Languages with obligatory onsets and without distinctive length

Message 1: Languages with obligatory onsets and without distinctive length

Date: Sat, 30 Nov 1996 11:39:49 +0900
From: Robert Ratcliffe <>
Subject: Languages with obligatory onsets and without distinctive length

Thanks to all who responded to my query regarding languages with
obligatory onsets and without distinctive length. The responses
were many and interesting, and a specific thanks and summary
follows. However, since many people who responded, were curious
as to why I was asking, first let me explain:

The original reason for my query is this: I am trying to develop
the argument that in languages which both 1) require a default
onset (onset filling) and 2) show compensatory lengthening type
processes (coda filling), the two types of filling are
independent of each other. This may seem obvious to some, but at
least one influential line of research has tried to account for
both types of filling in these languages in terms of a single
directionality parameter (such as left-to-right or edge-in). It
occurred to me that it might be helpful to my argument to note
that some languages have neither type of filling and some have
one or the other but not both. I knew (or rather thought I knew,
see summary) some languages which have neither type. I also
think I know some languages which show compensatory lengthening
type processes without obligatory onsets. I didn't know any
languages which had obligatory onsets, but no comp.length.
processes. But I thought there must be some, hence! the query.

I now have serious doubts about whether such a neat phomemic
typology is in fact possible. I come back to this with a new
query after the summary.

	---THANKS and SUMMARY-----

First Thanks to the following for their responses:

Alice Faber, Haskins Labs; San Duanmnu, University of
Michigan;Tapani Salminen, Department of Finno-Ugrian Studies,
University of Helsinki;John Coleman, Oxford University Phonetics
Laboratory; Mel Resnick, The University of Tulsa;Carsten
Peust,Seminar of Egyptology and Coptology,Goettingen;Liz McKeown,
SOAS, London;John Atkinson, Australia;Joseph DeChicchis,
Hiroshima University; Waruno Mahdi, Berlin; and Wenchao, UK.

Next, since the responses principally made aware that the issues
involved were more problematic than I had imagined, I will
organize the summary around the problems raised.

PROBLEM 1-- default onsets-- phonetic? or phonemic?

I have to admit that in trying to keep my query as simple as
possible, what I asked for was not exactly what I was looking

As Joseph DeChicchis notes in his response "it is difficult to
know the extent to which your [query] is a question about
phonetic allophony or (morphophonemic) typology."" I am
interested in the phonemic typology aspect of the question. But
certainly it would be a mistake to exclude apparent cases of
phonetic allophony a priori. The phonetic and phonemic aspects of
the issue are very difficult to disentangle.

I was assuming that onset filling was a universal PHONEMIC
parameter (=Languages either allow V initial syllables or they do
not). However, it seems that languages (like English and
Japanese) which I assumed allow V initial syllables, do in fact
show a phonetic glottal stop onset before at least some of these

Mel Resnick notes for example that "No English *utterance* can
start with a phonetic vowel. If an *utterance* would otherwise
start with a vowel, a glottal stop is preposed.E.g., Question:
What grows on apple trees? Answer: [?aeplz]."

And according to Joseph DeCicchis, "many (most?) Japanese
dialects are also C initial: even words transcribed with initial
A, I, U, E, O begin with a consonant (usually the glottal
plosive, although the O words sometimes begin with a bilabial

So that instead of being due to a phonemic parameter, the
appearance of default onsets in some cases may be due to a
universal phonetic, i.e. physiological effect. Is there any valid
distinction to be made between a phonetic and a phonemic default?

The only answer I am able to think of to this question is that
the phonemic default must otherwise be a valid phononeme of the
language. This definition works for the Semitic languages which I
am familiar with, where the default onsets (w or glottal stop in
Arabic, y or glottal stop in Ge'ez) are otherwise phonemic,
eg. Cl.Arabic /bi?run/ "well" vs. /barrun/ "dry land"

This criterion would, I believe allow us to exclude languages
like Japanese and English where the glottal stop is not phonemic.

It also seems that the most promising languages to investigate in
search of a phoNEMIC default, might be those where the default
onset is something other than glottal stop, which seems to be the
universal phonetic default. For example,

In Tundra Nenets (according to Tapani Salminen,)

"The default initial consonant is ng [= velar nasal]; it is
regular in established loan-words, and appears through an
automatic adaptation process even with the latest borrowings,
like ngarmiya from Russian armija 'army'. There are certain
(minor) complications; please check";

In approximately 70% of Australian languages (according to John
Atkinson) single-consonant onsets are obligatory. He further
notes "I could not find any loan words in Dyirbal which start
with a vowel in the original language; however, forms like
"Englishman"=>"yingiliman", "orange"=>"ngarrinji" are normal in
other Australian languages which do not permit initial vowels.

In Mandarin Chinese (Wenchao notes)

"According to some models (e.g. Duanmu 1990), all syllables have
an obligatory onset. The obligatory onset is described as a velar
approximant, velar fricative, or glottal stop."

PROBLEM 2-- defective orthography vs. real default

A second problem, which the responses made me aware of, is that
of how to distinguish between a true phonetic or phonemic default
and a regular phoneme which simply isn't indicated in the
orthography. By a default consonant I mean a consonant that can
be interpreted as not part of the underlying structure of the
word, but which appears on the surface, either because it is
required by syllable structure constraints of the language
(phonemic default) or because it is a physiological effect of
speech production (phonetic default).

To begin with, in Classical Arabic, it is possible to distinguish
a default glottal stop from a(n underlying) phonemic glottal
stop. (The distinction is, or at least can be, indicated in the
traditional orthography.) For example in the word /?ismun/ "name"
we can assume that the /?/ is a default, because it only appears
if the word is not preceded by a prefix or word in close juncture
which ends in a consonant. Thus when we attach the proclitic
preposition /b-/ "in" or "by" we get/bismi/ "in the name of...".
But the /?/ in /?amrun/ "command" is a phoneme which is part of
the underlying representation of the word. It doesn't delete in
the above environments, thus /bi?amri/ "by the command of..."

It seems to me (though I am far from confident) that Carsten
Peust's comments re glottal stop in German represent arguments in
favor of a phonemic glottal stop in that language rather than for
a default glottal stop. He writes:

"Take German: There are many words which seem to start with a
vowel but this vowel is usually preceded by a glottal stop, so
you might say that there must be a consonantic onset. The glottal
stop can also occur word internally (but this is always at a
morpheme boundary). You could find minimal pairs such as

/?ain/ "one" versus /kain/ "none" versus /bain/ "leg" etc.,
/eR?ailen/ "to come over someone (said from fate, death or sim.) vs.
/eRtailen/ "to give/teach someone (a lesson, advice)" etc.etc.
The common analysis is not to posit /?/ for the phonemic level."

The glottal stop in Indonesian (Malay), which Waruno Mahdi was
kind enough to analyze for me, also seems to be a regular phoneme
which is defectively indicated in the writing system. Mahdi

In Indonesian (Malay) "Onset is never vocalic, but words spelled
with an initial vowel (and thus listed in dictionaries) are
pronounced with glottal onset, i.e. with initial glottal stop
(the glottal stop phoneme also occurs internally in intervocalic,
preconsonantal, and postconsonantal position -- the latter also
morph initially -- as well as morph and word finally):

_api_ /?api/ "fire" >>-> _berapi-api_ /br?api?api/ "fiery (of a speech)"
 (with prefix _ber-_ & reduplication)
 ? = glottal stop
  = schwa = mid central unrounded vowel=
 = IPA topsy-turvy _e_.

_ada_ /?ada/ "be, there be" >>-> _keadaan_ /k?adaan/ [k?ada?an]
 "situation, condition, circumstance"
 (prefix _ke-_ & suffix _-an_)
 Note: in the latter, the second glottal stop is phonetic,
 i.e. it is automatic, non-phonemic.

_koran_ /kOran/ "newspaper" >< _Kuran_ /kur?an/ "the Qoran"

 O = back mid-low rounded vowel = IPA topsy-turvy _c_,
 in the former, the syllable boundary precedes the /r/,
 in the latter, it follows upon it."

Most interesting from my point of view is Mahdi's comment:

"The retention of morph-initial glottal stop after a prefix ending in /r/
is productive, but in some petrified inherited derivations the glottal
is lost, resulting in the following case in a least pair:

_apa_ /?apa/ "what" >>-> _berapa_ /br?apa/ "have what; what does ... have"
 >< _berapa_ /brapa/ "how much, how many"

 The former is productive, the latter petrified inherited),
 the syllable boundary divides the intervocalic consonant
 cluster, and precedes the intervocalic lone consonant."

This pair looks to me a lot like the /bismi/<>/bi?amri/ Arabic
pair. I suppose it would be dangerous to extrapolate too far
from this datum. But is it possible that the glottal stop in
/?apa/ was originally a default glottal stop (hence /berapa/ in
the non-productive, hence older form) which has become
phonemicized (hence /ber?apa/ in the productive modern form)?

PROBLEM 3- When is length distinctive?

A second problem, which falls on the border between phonetics and
phonology, is how to determine whether length is distinctive or

In Japanese, for example, length is generally assumed to be
distinctive for both vowels and consonants, yet Joseph DeChicchis

"As for Mayan and Japanese dialects, the existence of a length
contrast will depend on the particular phonological
description. In the natural speech I have examined, I have failed
to find cases where a "phonologically geminate" contrast is
purely a length distinction. For example, in Chorti Mayan,
geminate vowels are often signalled by laryngeal creak; in
Japanese, geminate consonants have distinctive formant transition

In German, on the other hand, length is assumed not to be
distinctive, yet Carsten Peust writes:

"[In German]There are two classes of vowels, one being long and
one being short, but the former class also being different in
vowel quantity (in general more closed). So you may either posit
phonemic vowel quantity for German and treat the accompanying
articulatory diffferences as phonologically irrelevant, or vice
versa. A similar problem exists for English vowels."

PROBLEM 4- Coda filling and compensatory lengthening not equal

Finally, I had been assuming that coda filing was always
compensatory lenghtening-- i.e. that it always involved the
maintance of a distinctive weight contrast, and hence that it
should only be found in langauges with distinctive length. Liz
McKeown has, however, referred me to some interesting research on
Chinese which clearly goes against this assumption:

"Duanmu (1990) in his PhD thesis... argues that all Chinese
syllables are the same length, with one slot in the onset and two
in the rhyme. In a CVC (eg 'man') or CVV (eg 'hao') syllable,
the vowel (or vowels) will always be short, while in CV syllables
(eg 'ma') the vowel is long: this distinction is totally regular,
and since it is not contrastive, has not been noted before.

"Goh (1996 - 'The Segmental Phonology of Beijing Mandarin', PhD
Thesis, SOAS, London) also claims that all Chinese syllables are
the same length. He proposes a uniform template for the minimal
phonological string, consisting of two onset-nucleus pairs, where
if O2 is filled, N2 must be p-licensed, and if N2 is filled, O2
must be p-licensed. Where both O2 and N2 are empty, material
will spread from N1 into N2 to create a long vowel. There are no
branching constituents."


The best examples, of what I was actually looking for then, would
appear to be

1) Modern Hebrew (though not Biblical), whic according to Faber
has the following properties: "Permitted syllables are CVC, CV,
CCV, CCVC, CVCC. I can't think of any CCVCC syllables, but I
wouldn't be surprised if they existed in loanwords. Glottal stop
is phonemic (and may reflect historical /h/ and /9/ (voiced
pharyngeal approximant) as well as /?/). Borrowed words that are
vowel-initial in the source language would generally be nativized
in Hebrew with an initial /?/: English is /?anglit/. Biblical
Hebrew unambiguously had a contrast between single and geminate
consonants, a contrast that is still reflected in normative
orthography.And I suppose someone who wanted to write a fairly
abstract phonology for Modern Hebrew might conceivably
incorporate gemination in underlying representations. But there's
absolutely no basis for this in the modern language"

2) Several Australian languages including Dyirbal. According to
John Atkinson "Around half of the 200 or so Australian languages
would satisfy both your properties. Single-consonant onsets are
obligatory in perhaps 70% -- although in a few of these, "i-" and
"u-" occur on occasion as allophones of "yi-" and "wu-".
Borrowed words are normally adapted to fit -- thus "yinggilibi"
for the introduced European bee. Hardly any Australian languages
allow geminate or long consonants."

3) Tundra Nenets (according to Tapani Salminen), see above


Thanks again to all of the respondents and to this wonderful
forum for making me aware of aspects of this issue which, arising
from language data or research traditions with which I am
unfamiliar, I would never have discovered on my own. In the
short run I have decided to drop the phonemic typology line of
argument from the paper I am working on. But I am still curious.

For those of you who have followed me this far (and are not
uspest with me for being so slow to summarize), my new query is:
do you think my original typology is valid? That is: given the
two parameters obligatory vs. non-obligatory onset filling and
obl. vs. non-obl. coda filling, there are four ideal language
types. Are clear examples of each of the four types actually
found? The following are my examples for the first three (plus
the List's examples for the fourth). Please tell me if you think
they don't qualify and give me better examples if you can.

(Although I realize I was wrong in assuming that coda filling
processes were only found in languages with distinctive length, I
continue to assume that length is necesarily preserved where it
is distinctive. Thus presence of distinctive length is a good
preliminary indication that obligatory coda filling is also

1-onset not obligatorily filled/coda weight not obligatorily

English (but see summary), French

2.-onset obligatorily filled/coda weight obligatorily preserved

Several classical Semitic languages-- Classical Arabic, most
modern dialects of Arabic, Biblical Hebrew (though not modern see
summary), Ge'ez; several Chadic langauges, for example Pero
(Frajzyngier 1977, etc.), at least one Penutian langauge- Sierra
Miwok (Noske 1985)

3.-onset not obligatorily filled/coda weight obligatorily

Japanese (but see summary), Classical Greek, Classical Latin,
several Bantu languages, for example Luganda, as analyzed by
Clements (1986).

4.onest obligatorily filled/coda wieght not obligatorily

Modern Hebrew, Tundra Nenets, Dyirbal and some other Australian

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