Date: Fri, 25 Feb 2005 11:13:47 -0500
From: Mike Maxwell <email@example.com>
Subject: The Phonological Spectrum, vol I: Segmental Structure
EDITORS: van de Weijer, Jeroen; van Heuven, Vincent J.; van der Hulst, Harry
TITLE: The Phonological Spectrum
SUBTITLE: Volume I: Segmental Structure
SERIES: Current Issues in Linguistic Theory 233
PUBLISHER: John Benjamins
Michael Maxwell, Linguistic Data Consortium / University of Pennsylvania
The preface (p. vii) to the two volume work (only the first volume is reviewed
here) states the aim as "giving a comprehensive overview of current
developments in phonological theory." That is a tall order, and not one that
this volume delivers with respect to segmental phonology. Only a few
themes are covered: three papers each on nasal harmony and voicing
assimilation, and a single paper each on diphthongs, lenition and signed
languages. While some of these topics might serve as stand-ins for others
(nasal harmony for vowel harmony, for instance), in fact the range of
approaches discussed is rather limited. If you want a "comprehensive
overview", you will be better served by a handbook.
Moreover, the papers on nasal spreading are flawed by misreadings of the
primary data and by focusing on un-affixed words, as I will discuss below.
But if this work does not succeed in its avowed aim, there are still some
interesting papers. The editors give a rather short but useful overview of
the papers in this volume; I will attempt a closer look at some of the issues
The first two papers, by Paul Boersma and Rachel Walker, discuss nasal
harmony systems, and in particular the question of so-called 'transparent'
segments. Languages allow a range of nasal spreading varying across a set
of segments ranging from none (languages which do not evidence nasal at
all) through adjacent vowels, glides, liquids, and voiced stops. In each case,
all the intermediate segments are nasalized; spreading can thus be viewed
as local. But an issue arises in the extreme case in which spreading is
alleged to cross voiceless obstruents, but where these obstruents remain
unaffected: they are apparently transparent to nasal spreading.
Under the theory of autosegmental phonology, spreading between non-
adjacent phonemes has often been accounted for by assuming that at the
relevant feature nodes are in fact adjacent at some level. For various
reasons, Boersma and Walker prefer a different solution. Boersma
approaches the data from the standpoint of his theory of "Functional
Phonology", which relies on the difference between production and
perception. In one group of languages, ostensibly united by a constellation
of properties, nasal spreading is said to be based on spreading of the
articulation, while in a second group of languages, with different
properties, spreading is said to be based on perceptual harmony. It is in
the latter set of languages that so-called transparent segments are found.
Boersma makes an analogy to vision: if we see a car behind a lamppost, we
don't perceive two cars, but only one. Nasality is the same: we hear
nasalization on two non-adjacent phonemes, with a non-nasalized stop
between, and the stop is perceived as not breaking the nasalization.
For her part, Walker uses an extended version of Sympathy Theory (itself an
extension of Optimality Theory) to argue that nasal spreading does target
voiceless obstruents in the optimal candidate, but that in fact the obstruents
are skipped by virtue of a Sympathetic candidate.
Walker considers a possible counter-argument to her analysis: why are the
only attested transparent segments voiceless obstruents, which are the only
consonants that would be unnasalizable? That is, why are there no
languages where, say, voiced obstruents do not block nasal spreading, but
also do not nasalize? She attributes this apparent coincidence to the rarity
of opacity in general, which I do not find convincing.
Boersma and Walker both draw heavily on the literature of Tucanoan
languages, particularly Tuyuca. But in my opinion, both analyses have a
serious flaw: the authors restrict attention to monomorphemic words.
Tuyuca, like the other Tucanoan languages, is highly agglutinative, and
nasal spreading as a process is only observable in suffixed forms. What
Walker and Boersma refer to as nasal spreading in Tuyuca is really only a
generalization about the form of monomorphemic stems.
Nasalization in Tuyuca is described in detail in a paper by Barnes 1996,
which is cited by Walker, but not by Boersma. Briefly, the facts are these:
stems, which are virtually all bisyllabic, are either wholly nasal or wholly
oral. Nasalization tends to spread to the right from a nasal stem onto
suffixes. (Tucanoan languages have few if any prefixes.) But suffixes,
unlike stems, have a three-way contrast: some suffixes are inherently nasal,
and condition nasal spreading to their right, regardless of the presence or
absence of nasalization to their left; some suffixes are inherently oral, and
block further spreading of nasalization; and some undergo nasalization
when they appear to the right of a nasalized vowel, allowing that
nasalization to spread further. If this three-way contrast were
phonologically predictable (e.g. by the segmental composition of the
suffixes), this would be of interest, but not perhaps surprising.
But -- and this is the crucial point -- the membership of a suffix in one of
these classes is largely unpredictable, as Barnes takes pains to document.
In suffixes, voiceless obstruents plus /b/ and /d/ block spreading of
nasalization from morphemes to their left (although some affixes
beginning with voiceless obstruents have an inherently nasal vowel to the
right of that obstruent, giving the illusion that the obstruent is
transparent). Beyond this generalization, nothing more can be said.
Suffixes which consist of a single vowel, with no consonant at all, may in
fact resist nasal spreading. Minimal pairs, and even minimal triplets, of
suffixes exist, where the only difference lies in the nasality or nasalizability
of the suffixes. Nor do such distinctions as inflectional vs. derivational
suffixes have any predictive power.
The process of nasal spreading across morpheme boundaries onto suffixes
in Tuyuca thus differs from the putative spreading within monomorphemic
stems in three ways:
(1) Whether spreading affects a suffix cannot, in general, be predicted by
the phonological content of the suffix, with the following two exceptions:
(2) Voiceless obstruents are not transparent to spreading; rather, all
suffixes containing voiceless obstruents block the process of spreading.
(3) Not all consonants which have nasal counterparts allow spreading. In
particular, the voiced stops /b/ and /d/ block spreading in all suffixes in
which they appear, despite the existence of the phonemes /m/ and /n/.
(The velar stop /g/, on the other hand, does allow spreading in a lexically
determined set of suffixes.)
Similar facts are well documented in several Tucanoan languages, and pose
a problem for any theory which attempts to derive the facts of nasal
spreading across morpheme boundaries from the phonological form of
affixes. The failure of either Boersma or Walker to address the process of
nasal spreading in suffixes is thus a puzzling omission. In fact their
analyses account for nothing more than the nasalization properties of
stems, facts which at least potentially require no explanation at all, since
there are no alternations, and the stems could simply be lexically listed with
nasalization on all the relevant segments. Alternatively, one might account
for the uniform properties of stem nasality by means of Morpheme
Stefan Ploch addresses a different issue, claiming that the phonologically
nasality of a phoneme cannot be determined simply by phonetic
measurement. Rather, the property of being nasal is a cognitive property,
and the status of phonological nasalization can only be determined from
the contrasts a language employs, and from its phonotactic constraints.
Ploch casts his argument in the context of the 'Element Theory' of
phonology (Kaye et al. 1985), but the conclusion is compatible with a
variety of theories, including structuralist phonology.
While I am in essential agreement with Ploch on the status of phonological
features such as nasalization, there are a number of problems with his
arguments. First, one of Ploch's arguments draws on work by Entenman to
the effect that in order to be perceived as nasal, low vowels need a larger
velopharyngeal opening than do high vowels. From this he concludes that
nasalization cannot be defined as velic opening. While this does imply that
the property of being nasalized cannot be derived from a single phonetic
property, it is unclear that phonological nasalization could not be derived
from a constellation of phonetic properties.
Ploch also argues against a claim by Kawasaki which calls on perceptual
cues to explain the origin of certain phonological processes of
denasalization. Specifically, since the oral/nasal distinction is hard to hear
on vowels, the (partial) denasalization of nasal consonants before oral
vowels enhances the detection of the oral/nasal distinction on the following
vowel, and is therefore to be expected.
Some of the data Ploch relies on to counter-exemplify Kawasaki's claim
comes from an analysis of the Ecuadorian language "Auca" (now known
as 'Waorani' or 'Wao'), done by Ken Pike and Rachel Saint in 1962. Since this
analysis has been cited elsewhere, it is worth pointing out that Pike and
Saint relied (out of necessity, at the time) on a young speaker of the
language, Dayuma, who had fled from her language community as a girl,
and had been living in a different language community (Quichua) for
several years. When fluent speakers of Waorani became available several
years later, it turned out that nasal spreading in the language was more
pervasive than had been apparent from Dayuma's speech. (A revised
analysis of Waorani phonology has never been published, but an accurate
description is given in the pedagogical grammar Peeke 1979.) In particular,
one of the examples from Pike and Saint that Ploch relies on here is flawed.
It purports to show a nasal stop preceding an oral vowel. But with the
better data now available, this word turns out to have a nasal vowel
instead. Fortunately, it is possible to replace this incorrectly transcribed
word with correctly transcribed ones that make the same point, so Ploch's
point here stands.
Ploch's overall argument may not fare as well. The problem with Ploch's use
of Waorani to counter-exemplify Kawasaki's claim that a language can "use"
nasalization on consonants to enhance perception of nasalization on
vowels, is that in Waorani, obstruents following (not preceding) a nasal
vowel have precisely that effect. Specifically, in the environment following a
nasal vowel, voiceless obstruents are prenasalized, and may become voiced,
while voiced obstruents become nasal. Kawasaki's more general point --
that nasalization on consonants often gives clues to the nasalization on
vowels -- still holds, except in the opposite direction: in Waorani, the clues
are to be found on the consonant following the vowel, rather than the
consonant preceding the vowel. This would be an extension of Kawasaki's
argument, to be sure, but not an implausible one. At the same time, Ploch
is correct to point that Kawasaki cannot predict what happens, he can at
best offer probabilities.
While I am on this topic, I will make a more general point: much current
work is being devoted to moribund languages, in some cases attempting to
document languages based on the recollection of elderly speakers who
have not had the opportunity to talk with other fluent speakers of their
languages for many years. This is laudable, but it is incumbent on the
documenters of such languages curb their enthusiasm and make it clear
how reliable the data is, lest future generations of linguists be misled into
making false generalizations.
Another problem with the primary data appears in an argument on page
95, in which Ploch states that nasalization in Cubeo does not spread
beyond a single onset-nucleus pair. This is incorrect; the error is based on
a misreading of Salser 1971. Salser, writing in an American structuralist
phonemic framework, limits his discussion to allophonic rules; he does not
discuss nasal spreading beyond the syllable, because this would be a
morphophonemic process in Cubeo. Nasal spreading across morpheme
boundaries in Cubeo in fact functions much the same as the nasal
spreading in Tuyuca described above (for more detail, see Morse and
I now turn to the three articles that discuss voicing. Mirjam Ernestus
discusses syllable-final and word-final devoicing of obstruents in Dutch,
and the devoicing of fricatives following obstruents. While obstruent
devoicing seems to feed fricative devoicing, Ernestus shows that final
devoicing exhibits the properties typical of phonetic processes (such as
variability and gradient effects), whereas fricative devoicing exhibits the
properties of phonological processes -- an apparently impossible situation
in which a phonetic process feeds a phonological one. This apparent
contradiction is resolved by assuming that coda obstruents come to lack a
value for the phonological feature 'voiced' at the phonological level, so that
their actual voicing is free to be determined by the phonetic component. In
effect, this makes voice a ternary-valued feature: plus, minus, or unmarked.
Moreover, it requires two markedness constraints on the voicing feature in
obstruents: one to the effect that "no obstruent has a [voice]-feature",
deriving the lack of a feature value for voicing, and a second (higher
ordered) constraint saying that "an obstruent in a cluster is not voiced."
Caroline Féry examines obstruent devoicing in German. The descriptive
generalization is that while obstruents contrast for voice in syllable-initial
position, they are voiceless in word-final position, and most -- but not all --
are voiceless in syllable-final position. Two general sorts of explanations
have been proposed: either voicing is licensed in syllable onsets, or it is
forbidden in syllable codas.
One might hope that the issue could be resolved by looking at ambisyllabic
stops, which are presumably both syllable- initial and syllable-final.
(Ambisyllabicity is of course controversial; Féry argues briefly for it on the
basis of the distribution of stressed lax vowels, which do not appear in
German in unambiguously open syllables. Some unfortunate typos in her
example (10a) mar the presentation, if not the actual argument.) But as it
turns out, German behaves inconsistently: native words have only voiceless
ambisyllabic obstruents, while some loanwords have voiced stops in that
position. This contrast would be consistent with a two-level stratification of
the German lexicon, something that has been proposed in the past; but
Féry argues that there are further idiosyncrasies of the lexicon which
demand a hierarchical stratification into multiple levels. Féry develops an
Optimality Theory (OT) analysis of the facts, although it is not clear that an
OT analysis would have any advantage over a traditional rule- based
There are apparently very few voicing alternations which hinge on
alternations in ambisyllabicity; Féry mentions one, an alternation between
two forms of a 'strong' verb: schneiden ~ geschnitten. But the vowel
changes of German strong verbs render them essentially irregular, meaning
that the allomorphs must be stored in the lexicon. So it is unclear whether
the voicing alternation in this case needs to be accounted for as a
synchronic phonological process. Moreover, it is not obvious that the
intervocalic obstruent in schneiden is not ambisyllabic. There is no hard
evidence, since the preceding tense vowel need not appear in a closed
syllable; but neither is there any evidence that the obstruent is not
ambisyllabic. The conclusion that the voicing alternation is driven by an
alternation in whether the obstruent is in a coda is therefore doubly weak.
Apart from these few putative alternations, then, it could plausibly be
argued that there is no evidence from alternations for an active process of
devoicing ambisyllabic obstruents. Under a derivational theory of
phonology, German speakers might simply have chosen underlying forms
which reflect the surface voicing of those obstruents.
Féry addresses a similar issue in a footnote concerning word-initial [ts] vs.
[s] (the latter occurs in loanwords, but not in native vocabulary), but argues
that because speakers will have heard both pronunciations of loanwords,
they must allow for both inputs, and their grammar must choose the correct
output. While this argument goes through for OT (assuming "richness of
the base"), it does not go through for traditional phonological theories,
where speakers are free to choose an appropriate underlying form for their
In summary, for the native speaker, the division of German words into two
sets, one with and one without voiced ambisyllabic obstruents, might simply
be a random division, no more interesting than the division of non-native
words into those which happen to have or not have voiced ambisyllabic
obstruents, or for that matter the set of words which have or do not have
the phoneme /m/. Without knowing what else follows from the division, it
is impossible to tell.
K.G. Vijayakrishnan assumes that asymmetries in the typology of
phonological processes -- in this case processes of weakening in Tamil --
are to be accounted for by universal conditions on constraint ordering.
While I suspect that the explanation for such asymmetries is to be found
elsewhere (in the phonetic processes which are the diachronic sources of
phonological processes), explanations based on universal orderings are
common in much recent OT work. At any rate, the interest in this paper
may lie in the different approach to the same problem as that discussed by
Féry, namely the problem of phonological strata. Rather than assuming
different constraint hierarchy orderings, Vijayakrishnan suggests that the
underlying representations are different: older words in the history of the
language, which are more prone to weakening, are underlyingly unspecified
for the variable features, and therefore more subject to markedness
constraints, while prespecification blocks such weakening in newer words.
While this runs contrary to most work in OT, with its assumption of
the "Richness of the Base", it is nevertheless an interesting proposal,
potentially tying together issues of irregular forms with research into
the "emergence of the unmarked".
Eon-Suk Ko's study concerns the effect of the phonation type of a prevocalic
consonant on the F0 of the following vowel in Korean, which has a three-
way contrast, traditionally referred to as lenis-aspirate-fortis (the precise
meaning of these terms has been debated). Previous studies have shown
that the F0 is higher after aspirate and tense consonants than after lenis
consonants; it had been suggested that this high tone had been
phonologized in Korean to a high tone, on the grounds that the rise in F0
observed in Korean was substantially greater than the analogous change in
English or French. Ko argues that the change is merely allophonic. She has
several arguments, which I will not attempt to describe, but which seem
Markus Hiller discusses a phonetically small, but nevertheless phonemic,
contrast in Swabian German between two diphthongs. At first glance (and
based on instrumental analysis), the difference appears to be a minute
timing difference. But if the phonology needs access to such low-level
phonetic detail, it brings into question the phonology-phonetics
distinction. Hiller considers several ways the contrast might instead be
represented, ultimately opting for treating the diphthongs as complex units
consisting of a consonantal part and a vocalic part (with low vowels
counting as consonantal in some diphthongs).
Philipp Strazny investigates why a set of consonants in Zulu causes lowering
of a following high or low tone. The set of tone-depressing consonants
does not form a natural class under traditional feature systems. (The
presentation is confused by the inconsistent use of orthographic forms and
IPA transcription. Another source of confusion is what appears at first
glance to be labels on a scale of F0 values at the bottom of a sound
spectrogram on page 224; the labels turn out to be measurements of F0 at
certain points along a time axis whose tick marks correspond to unlabeled
The solution Strazny proposes is the creation of another segmental
phonological feature (or actually, a pair of features, the second feature
being motivated by an analogous phenomenon in another language). This
feature is based on the articulatory gesture of vocal cord tensing; the more
traditional 'H' and 'L' features are then viewed as abstract features: cover
terms for a set of gestural features, one or more of which an individual
language may use to produce certain tones. The fact that certain Zulu
consonants cause tone lowering then arises out of an interaction between
gestural features, some of which are carried by the consonants in question,
while others serve as the realization of the abstract features of tone. As a
welcome side-effect, this explanation accounts for the fact that tone
depression is strictly local (adjacent to the consonant that causes it), while
high tones themselves may float some distance away from their source.
Strazny's proposal is related to the idea that a feature like voice may be
differently implemented in different languages. Furthermore, it is a small
step from here to the notion that the cover terms, such as 'H' or 'L', or
even 'voice', are language-particular, rather than universal.
Finally, Onno Crasborn and Els van der Kooij argue that the configuration of
the base joints of the fingers (the joints closest to the hand) are never
phonologically significant in the Sign Language of the Netherlands (NGT).
(Poor quality in reproduction has resulted in the crucial labels in one
photograph being virtually impossible to read, and the contrast is marginal
in some of the other pictures.) While there are obvious differences in the
flexure at the base joints among signs, Crasborn and van der Kooij present
convincing reasons to think that most of the differences are free or
conditioned variation. (They do not use the term "allophonic", nor do they
state the conditioning in terms of rules or constraints, but that seems to be
what they mean.) A small set of words whose base joint configuration
cannot be accounted for in these terms is said to be iconic (similar to
onomatopoeia in spoken languages).
Crasborn and van der Kooij briefly consider a few other sign languages,
which seem to be like NGT in not using the base joint flexure contrastively.
The authors predict that this is likely a universal of signed languages,
presumably because it is innate. Of course, the discovery that some
language did use the basal joints contrastively would not undermine
Crasborn and van der Kooij's claim for NGT, only the universal claim. At any
rate, I am skeptical that feature systems for signed languages (or spoken
languages, for that matter) are either universal or innate. Nevertheless,
following up on this study in other signed languages would be a fascinating
At $95, the book is rather expensive for what it provides. There is a web
where one can read it, and for a nominal price copy or print pages. If you
are only interested in one or two of the articles, this is probably a better
bet. However, the special browser software runs only under Netscape
Communicator 4 and Internet Explorer, under Microsoft Windows.
Barnes, Janet. 1996. Autosegments with three-way contrasts in Tuyuca.
International Journal of American Linguistics 62:31-58.
Kaye, Jonathan, Lowenstamm, Jean, and Vergnaud, Jean-Roger. 1985. The
Internal Structure of Phonological Elements: A Theory of Charm and
Government. Phonology Yearbook 2:305-328.
Morse, Nancy L., and Maxwell, Michael B. 1999. Cubeo Grammar: Studies in
the Languages of Colombia, 5. Dallas, TX: Summer Institute of Linguistics.
Peeke, M.Catherine. 1979. El idioma huao: Gramática pedagógica, tomo 1:
Cuadernos Etnolingüísticos, 3.Quito: Instituto Lingüístico de Verano.
Salser, J. K. 1971. Cubeo Phonemics. Linguistics 75:74-79.