emptiness in phonology - pázmány péter … in... · intuition that an element of phonological...
TRANSCRIPT
EMPTINESS IN PHONOLOGY
KATALIN BALOGNÉ BÉRCES
1. Introduction: emptiness and abstraction1
The present paper is meant to provide a brief, only moderately theoretical
overview of possible interpretations of emptiness in phonology, i.e., in (the
study of) the sound pattern of language. While most examples and
illustrations of the phenomena discussed will come from English, the
claims are supposed to hold universally, as is suggested by the diversity of
the languages alluded to in similar contexts in the sources used (and
referred to, accordingly, below).
One might think that talking about emptiness in the sound system of
languages is at best contradictory. How could, after all, invisible and mute
objects exist in the system of our vocalizations? However, we should bear
in mind that (current) phonological theory is primarily concerned with the
formal modelling of the cognitive processes that are manifested by those
vocalizations, rather than the physical manifestation itself. This latter area
is tackled by what is called phonetics—the study of speech sounds as
physical (articulatory, acoustic or auditory) objects. Indeed, in phonetics
emptiness is synonymous with real silence, i.e., the absence of verbal
activity, and as such it may be referred to as a “pause”, which corresponds
in phonological descriptions to the edge of the so-called “utterance” (the
stretch of language said at one go, both preceded and followed by a
pause). Phonology, however, deals with how what happens in the physical
world of utterances can be represented in a mental model, in such a way
that the model provides explanations to why what happens could not
happen differently, and it also makes predictions which may or may not be
supported by a new set of data.
Therefore, phonology may sometimes be forced to conclude that the
observed phenomena or processes can (only) be modelled with recourse to
1 The paper, as well as the author, has immensely benefited from the work, in both
research and teaching, of John A. Goldsmith, Jonathan Kaye, Jean Lowenstamm,
Jean-Roger Vergnaud, John Harris, Stephen R. Anderson, Monik Charette,
Michael Kenstowicz, Miklós Törkenczy, and Péter Szigetvári, among others, some
of whom the author is honoured to know, and to have been taught by, in person.
All the errors and misinterpretations are hers.
Katalin Balogné Bérces
a theoretical tool which has little or no physical reality. Take the very
simple case of the syllable as an example. There is massive speaker
intuition that an element of phonological organization grouping sound
segments into units, popularly dubbed “syllable”, exists beyond doubt.
Moreover, there is a long tradition in phonological descriptions to treat
certain regularities as governed by syllable structure.2 And yet, no
phonetic study has succeeded in specifying either the articulatory or the
acoustic correlate(s) of the syllable—apparently, it is more of a
psychological construct than a truly physical object.
Modelling, that is, describing and explaining physical reality, then,
may involve a great degree of abstraction. In chemistry, water, one of the
most ordinary and basic substances, is modelled as H2O. In physics, the
fact that free objects fall downwards on the surface of the Earth (rather
than float in the air like astronauts in space) is due to gravity, i.e., an
invisible force of attraction determined by the mass of the two objects in
relation. What is more, the same force is said to be responsible for the fact
that the Earth goes round the Sun (and not the other way round, as
previously thought)—a claim that obviously goes against our everyday
observations. Science is very often far removed from what is simply
observable or perceptible.
In a similar vein, phonology will be shown below to frequently
manipulate entities not necessarily rooted in physical representations. In
particular, we will argue that emptiness, i.e., categories not present in the
phonetic signal, are detectable at all levels of phonological structure. As
Honeybone (2010: 5) explains:
(1) If phonology is seen as a cognitive entity (or even if it is conceived of as a
purely theoretical instrument), it should be clear that, while phonetics can
be seen to have an influence on phonology, there could be purely
phonological entities which are not founded on phonetics; these could
only be discovered by reasoning and analysis—not by phonetic
investigation; and there may even be phonological entities which have
only an indirect influence on speech.
It will be argued below that, similarly to the other components of linguistic
knowledge and their respective studies, especially syntax and morphology,
phonology also has to assume emptiness (“embodying” the absence of
2 In modern, generative phonological theory, this view goes back at least to Kahn
(1976), the symbolical break with the linear, segment-based representations of
Chomsky & Halle (1968) (SPE).
Emptiness in Phonology
information) whenever a property appears to be inert; empty/zero/“covert”
categories are posited whenever an entity with no apparent physical body
exhibits some systematic behaviour, that is, manifests itself. In a sense,
then, empty phonological objects are very much like ghosts (or, more
precisely, poltergeists): what we observe as taking place in sound structure
in a given situation gets an adequate explanation (only) with recourse to a
hidden agent.
The paper is structured as follows. Section 2 deals with emptiness in
linguistic structure, in syntax (2.1) and morphology (2.2). Section 3, as the
main body of the paper, argues for assuming empty categories in
phonology, too. First, autosegmentalism, the separation of the dimensions
of phonological representation is introduced (3.0), and the claim is made
that autosegmental representations do not only facilitate the expression of
emptiness but they directly lead to its necessary assumption. Then, Section
3.1 considers empty categories in suprasegmental phonology: empty
consonantal positions as they arise in the analysis of liaison (3.1.1) and
hiatus-filling (3.1.2); and empty vocalic positions word-finally (3.1.3) as
well as in other word positions (3.1.4). Finally, Section 3.2 states that
emptiness is present in subsegmental structure, too, in the form of, e.g.,
underspecification, and Section 4 concludes.
2. Evidence for empty categories in linguistic structure
This section presents a brief sketch of empty categories as they are used in
current syntactic and morphological theory. Rather than give an exhaustive
survey, it aims to highlight, with the help of just a few, well-known,
classical textbook examples, that zero elements are an integral part of both
syntactic and morphological descriptions. For more detail, the reader is
referred to the sources indicated.
2.1 Syntax
One of the distinctive features of transformational grammar, the most
popular of present-day syntactic theories, is the abundance of empty
categories. Moreover, it is also claimed that non-overt elements have to be
classified into subtypes, e.g., a whole typology of empty pronouns (PRO,
pro) and traces of moved elements has been set up (cf., e.g., Haegeman &
Guéron 1999: 383–405). There are arguments of all sorts (syntactic,
morphological, phonological, semantic—see Radford 1988: 466–479) for
Katalin Balogné Bérces
the analysis of transformations (such as passivization or the formation of
direct wh-interrogatives and relative clauses) as resulting in so-called
“chains” of moved categories and their traces, of which I will choose only
one, wanna-contraction. The point illustrated here is that the trace left
behind by the movement of a wh-phrase from its extraction site to its
position in surface structure forms a kind of invisible barrier, which is able
to prevent the phonological contraction of two, apparently adjacent
elements, for example the verb want and the following infinitival to
introducing its complement. The examples (in (2)–(4) below) are
reproduced from Radford (1988: 475–476).
The story is the following. In colloquial English, want can contract to
wanna, as shown in (2).
(2)
a. I want to win
b. I wanna win
Wanna-contraction is also possible in sentences like the one in (3), where
a wh-phrase (who) has moved from the object position after beat to the
front. Notice that who, ensuring the interrogative force of the sentence by
taking the initial position, also serves the role of object at the same time,
since (i) it can be replaced by the accusative form whom, (ii) it appears in
the object position when it stays in situ in, e.g., so-called echo-questions
like You want to beat who(m)?, and (iii) the same position is taken by the
phrase representing the missing piece of information in responses, e.g., I
want to beat Jim. Empty categories, like the trace of who in (3c), will
henceforth be denoted by a boldfaced zero.
(3)
a. Who do you want to beat?
b. Who do you wanna beat?
c. Who do you wanna beat Ø?
As shown in (4), however, wanna-contraction cannot take place (indicated
by the asterisk in front of (4b)) when who originates from between want
and to, that is, when it is subject in a subclause complementing want.
Consequently, in such cases (i) it cannot be replaced by whom (cf. *Whom
do you want to win?), (ii) it appears in the embedded subject position
when it stays in situ, e.g., You want who to win?, and (iii) want and win are
separated in the same way in responses, e.g., I want Jim to win.
Emptiness in Phonology
(4)
a. Who do you want to win?
b. *Who do you wanna win?
c. Who do you want Ø to win?
Notice that the most evident explanation to why contraction is acceptable
in (3) but blocked in (4) is that the non-overt element, this abstract entity
with no physical body, serves as a barrier between the two words
preventing them from contracting. That is, want and to are not adjacent in
(4). The boldfaced zero, our syntactic ghost, only has mental reality
(existing in the cognitive system constituting the speakers’ linguistic
capacity) but no physical reality; still, it is able to exhibit systematic
behaviour.
2.2 Morphology
Zero morphs have an even longer history. Back in the 1940s and 1950s,
scholars posited empty inflectional affixes in examples like plural men or
past participle come corresponding to the -s in boys and -ed in sailed,
respectively, with no unanimous support from fellow researchers (cf.
Matthews 1991: 123–125). Certain cases, however, are less debatable,
such as the zero plural of nouns in English. As you can see in (5), the
plural of reindeer is not marked phonologically, that is, it is only marked
by a non-overt, abstract allomorph.3
(5)
a. All of the other boy-s used to laugh and call him names
b. All of the other reindeer-Ø used to laugh and call him names
It is important to see that one cannot analyze reindeer as having no plural
form since it is clearly different from words like information, which do
exemplify that case. While five reindeer, that is, five reindeer-Ø (zero
plural form), is analogous to five boys, neither *five informations nor *five
information is acceptable, i.e., the construct *five information-Ø is non-
existent, and the contrast between reindeer and information is expressed
by the reference to a non-overt, ghostly element acting as the plural
marker in the former example.
3 For a discussion of the principle constraining the use of zero allomorphs in
morphological theory, see Aronoff & Fudeman (2005: 16–17).
Katalin Balogné Bérces
The other area where zero categories are posited in English
morphology is derivation, where a word-formational operation having no
visible reflex in the derivative, traditionally referred to as conversion,
illustrates the point. It changes the lexical category of a word without
changing its phonological shape, and appears to be highly productive in
English.4 A noun like text, for instance, can be converted into a verb
meaning ‘send a text message’ without any affixes attached to it, as in He
is texting, and as such the word can serve as the base for further derivation
as in He is the fastest texter in the world. We need to posit the non-overt
derivational affix in such cases (i.e., text-Ø), since there is no visible
element which could be held responsible for the evident semantic and
morpho-syntactic change, apparent in other examples like courage ~
encourage or idol ~ idolize. In addition, in languages like Hungarian,
where noun-to-verb conversion is not an option, this transformation from
one word class to the other is always overtly marked. Therefore,
Hungarian SMS ‘text message’ expands into SMS-ez-(ik) ‘send a text
message’ with the denominal verb-forming suffix -ez- (plus person and
number inflection) added. English (s/he) text(s) will therefore receive a
perfectly parallel treatment, text-Ø-(s), where a covert morpheme
functions to encode crucial semantic and morpho-syntactic information.
3. Empty categories in phonology
I hope to have shown above that in both syntactic and morphological
descriptions, assuming empty categories is not only inevitable but is also
fairly accepted as part of the theoretical machinery, at least in mainstream
frameworks. Very often, we observe phenomena (e.g., the constraints on
wanna-contraction or the well-formedness of five reindeer) which only
receive proper explanation once we resort to this extremely abstract
device: an invisible agent can be shown to exhibit systematic behaviour.
And the behaviour of an entity is evidence of the existence of the entity.
In this section, which constitutes the major part of the paper, I will
argue that empty categories play a similar role in phonological theory.
4 For more detail, see, e.g., Aronoff & Fudeman (2005: 109–110, 133–137), Bauer
(1983: 32–33, 226–230) and Spencer (1991: 19–20), as well as the references
therein.
Emptiness in Phonology
3.0 A detour: autosegmentalism
Before we can embark on the discussion of emptiness in phonology, we
need to take a brief detour and introduce autosegmentalism.
Autosegmental phonology (Goldsmith 1976) claims that phonological
representations are multidimensional, i.e., their components are arranged
on separate, autonomous dimensions or levels called tiers (hence the
name). A major observation, to be illustrated below, is that phonological
rules can apply independently to the (component parts of) segments at
these autonomous levels. A rough sketch of such a model is given in (6),
consisting of a syllabic tier (encoding the number of syllables, traditionally
abbreviated to σ), a skeletal (or timing) tier (for sheer quantity), and a
segmental (or melodic) tier (for phonetic substance). Please note that this
representation is only an illustration, an oversimplified sample to suit our
present purposes; however, both the syllabic and the melodic tier are in
current phonological theory conceived of as decomposable into sub-levels.
(6)
syllable σ σ σ ef ho 2 2 skeleton x x x x x x x x g g yt g g g g melody t r i l e t
tree letter
A crucial feature of the model is the representation of long segments, e.g.,
the /i:/ in tree, which contain a single melodic element pronounced long,
i.e., taking up roughly twice as much space in an utterance as their short
counterparts. This is indicated in (6) by the single melody /i/ being
associated to two timing slots denoted by x’s, rather than one x,
characteristic of short segments (cf. all the others in (6)). The skeleton,
then, encodes length/quantity, separated from melody/quality. In addition,
it serves as an anchor for the rest of the autosegmental dimensions,
similarly to the spiral binding keeping together the pages of a notebook.
One of the most persuading pieces of evidence for tier autonomy
comes from a process called compensatory lengthening, whereby, as a
reaction to the deletion of a segment, a neighbouring segment lengthens to
compensate for the loss of the one deleted. Consider the data in (7) below.
It shows the way vowel length is affected by the presence vs. absence of
R-deletion (traditionally referred to as R-dropping) in non-rhotic vs. rhotic
Katalin Balogné Bérces
English, respectively, in the relevant environments, i.e., when the
orthographical R is either word-final (e.g., car) or pre-consonantal (e.g.,
harsh). Being a rhotic accent, standard American English pronunciation
called General American (GA) represents the pre-R-dropping situation
(historically), while Received Pronunciation (RP, the standard Southern
British accent) is non-rhotic and exhibits the effects of R-deletion.
(7)
GA RP
car more sir harsh fork
bird
A consequence of R-dropping in the RP pronunciations is that the vowels
preceding the deletion site systematically lengthen: in GA, there are short
vowels followed by pronounced R’s, whereas in RP, there are long vowels
(and no R’s, of course). In autosegmental terms, the long vowels of RP are
worth two short segments, i.e., two x’s—exactly the same as the short
vowel+R sequences of GA. This explains why the vowels lengthen in RP
after R-dropping: the quantity of segments within the syllable should
remain stable. This is shown for car in (8).
(8)
a. b. c.
syllable σ σ σ egi egi egi skeleton x x x x x x x x x
g g g g g g yt
melody k r k k
As you can see, the relevant portion of the pronunciation of car, -ar,
occupies the same number of skeletal slots, two, in both the R-ful (8a) and
the R-less (8c) pronunciations. The whole process may be modelled,
therefore, as R-dropping deleting the melody of the R only and leaving the
rest of the structure intact; this results in an empty x slot in the syllable (cf.
(8b)), which gets subsequently occupied by the melody of the vowel
Emptiness in Phonology
spreading from the left, to compensate for the loss of the R. That explains
the name compensatory lengthening. Most importantly, we only arrive at
this analysis if we assume the autonomy of the autosegmental tiers,
allowing R-dropping to manipulate melody only.
Below, I will argue that empty categories are a direct consequence of
autosegmentalism. A model that adopts the latter will automatically get
endowed with the former. This also follows from a fundamental property
of autosegmental representations: the fact that the skeleton–melody
relationship is not necessarily one-to-one. While it is in the case of
simplex short segments (9a), it is not in the other logically possible
combinations, all of which correspond to attested structures in languages.
Many-to-one correspondence is exemplified by so-called contour
segments (affricates like /t/ as in chin, or the short diphthongs of
French—(9b)); one-to-many correspondence, on the other hand, is
illustrated by long segments (9c), as we have already seen.5 But then the
question arises why we could not as well assume zero-to-one and one-to-
zero correspondences, zero being just like any other number. This means
that we assume the possibility of empty skeletal slots (9d) on the one hand,
and of unassociated, “floating” segmental melodies (9e) on the other.6
(9)
a. b. c. d. e.
skeleton x x x x x g 2 yt melody i t i r
As we will see below, classical syllabic theory, conforming to
autosegmentalism by definition, has managed to integrate this view into its
model to a certain extent. It recognizes the existence of certain empty
categories in certain phonological positions. However, it fails to bring the
logic of its own principles to its necessary conclusions and find that,
similarly to syntax, there exists a whole typology of phonological empty
categories, and all the types may, under limited circumstances, occur in all
phonological positions.
5 The issue of multiple branching, e.g., overlong segments, will be left aside here. 6 In fact, we have seen an empty timing slot in (8b), as part of the intermediate
stage in the derivation of compensatory lengthening.
Katalin Balogné Bérces
3.1 Emptiness above the segment
In this section we look at structure above segmental melody, i.e., we
concentrate on empty skeletal/syllabic positions7 and their subtypes, as
alluded to above. Some of the ideas presented below originate from (early)
Government Phonology (GP—Kaye et al. 1985, Kaye et al. 1990, Kaye
1990, Charette 1991, Harris 1994, Kaye 1995, etc.), but, as also mentioned
in the previous section, most of them are not advocated in mainstream
phonological theory.
We will start with rather standard autosegmental accounts of two types
of phenomena, liaison and hiatus filling.
3.1.1 Liaison
The term liaison refers to cases when a linking consonant is appended to
certain vowel-final words when a vowel-initial morpheme follows. What
may serve as a base for the epenthesis of this consonant may be
circumscribed lexically (as in French, for example—see (10)) or
phonologically (as generally assumed for English R-liaison—see (11)). In
French, a lexical property of certain vowel-final stems (e.g., les ‘the, Pl.’,
grand ‘great’, tout ‘every’, faux ‘false’) is their ability to “regain” their
final consonants, lost historically but still preserved by spelling, in the
situation in question.
(10) les les /z/ amis ‘the friends’
les /z/ enfants ‘the children’
grand grand /t/ homme ‘great man’
tout tout /t/ homme ‘every man’
faux faux /z/ amis ‘false friends’
Since the quality of the intruding consonants varies from stem to stem and
is therefore unpredictable8, their analysis as floating consonants lends
itself as an obvious theoretical choice. These final consonants, then, are
part of the lexical representation of the stems but not of their
7 The difference between empty skeletal positions and empty syllabic constituents
(with no skeletal slots) will not be discussed here. See Charette (1991). 8 Spelling is not considered to be a factor capable of influencing phonology. This is
further supported by non-historical intrusive consonants like Intrusive-R in certain
accents of English, in examples like vanilla /r/ ice or Gloria /r/ Estefan.
Emptiness in Phonology
suprasegmental structure, and are thus unpronounceable unless they can
spread forward to become the first consonant of a following morpheme.
The case of R-liaison in (most) non-rhotic accents of English is similar,
though there has been a long theoretical debate over its being either
lexically (the R is underlying) or phonologically (the R is inserted as a
reaction to certain specific vowel qualities to the left of its site)
conditioned. In either case, in a non-rhotic accent like RP morpheme-final
R’s are dropped when followed by a consonant or a pause in speech but
pronounced when a vowel-initial morpheme follows (cf. (11)). This latter
instance of the R is called Linking-R.
(11)
RP
car car is
more more ice
sir Sir Allen
If we opt for the analytical solution with the R stored in the lexical
representations of the stems, the process can be modelled analogously to
French liaison: the R is a floating consonant (cf. car is in (12a)), which is
only interpreted (= pronounced) when a target skeletal slot is available, to
which it spreads to become incorporated into phonological structure (12b).
(12)
a.
syllable σ σ egi egi skeleton x x x x x x g yt g g
melody k r z
b.
syllable σ σ egi egi skeleton x x x x x x g yt g g
melody k r z
Katalin Balogné Bérces
Note, however, that such an analysis does not only support
autosegmentalism, but it also crucially hinges on the theoretical
recognition of both floating melodies and, even more importantly, empty
skeletal slots, that is, phonological boldfaced zeros, as added to our
representation in (13).
(13)
syllable σ σ egi egi skeleton x x x x x x g yt g g
melody k r Ø z
Since liaison applies in the same way with all vowel-initial words
irrespective of the quality of the initial vowel, the morphosyntactic
properties or the meaning of the words, etc., we are led to conclude that all
surface vowel-initial words in fact start with an empty consonantal
position. Moreover, languages abound in similar linking phenomena, so
the observation seems to be rather universal. This means, then, that in all
languages, all words have to start with a consonantal (henceforth C)
position, which may or may not be lexically occupied by melodic material.
Crucially, this condition is automatically satisfied by lexically filled initial
C’s and no empty slots are allowed in such cases (e.g., *Øfz fizz). As a
consequence, the model predicts that linking is only possible with surface
vowel-initial words as only then do floaters find empty slots to dock
onto—a prediction born out by the facts. Therefore, our description of the
process provides an explanation to why the way it happens is the only way
it can happen.
3.1.2 Hiatus-filling
Hiatus is the name given to the sequence of two vocalic segments9. As the
name (Latin for ‘gap’) suggests, the universal (human) intuition is that
there is a yawning gap between the two vowels, since in the normal,
unmarked sequence of segments a consonant is sandwiched between them.
Therefore, various “repair strategies” exist in languages to avoid it,
9 They are situated in two separate syllables. When two vocalic elements are found
in the same syllable, they form a diphthong.
Emptiness in Phonology
including hiatus-filling, whereby a consonant or consonant-like element is
inserted between the vowels. Very frequently, the segments appearing in
place of the hiatus are glides (or semivowels—called so exactly because
phonetically they are intermediate between consonants and vowels), like in
English, where it applies in such a way that /j/ is used if the first of the two
vowels is high and front, while /w/ is inserted if the first vowel straddling
the hiatus is high and back. A few examples are given in (14), where the
underscores ( _ ) mark the potential insertion sites.10
(14) (a) hiatus potentially
filled by /j/(b) hiatus potentially
filled by /w/
ski_ing
play_a tune
fly_ing
boy_ish
my_idea
Woody_ Allen
so_exciting
allow_ing
Jew_ish
too_old
go_away
New_ England
The most straightforward explanation why hiatus-filling is not only
possible but cross-linguistically common assumes that there is an empty
consonantal slot between the two vowels, that is, the second syllable starts
with an empty C, e.g., ski-Ø-ing. Hiatus-filling, then, is possible because
the position is lexically present—all that happens is the phonetic
interpretation thereof. In addition, this interpretation is a frequent process
in languages because it is a simple way to resolve hiatuses, which are
cross-linguistically dispreferred and tend to be avoided exactly because
empty C’s are dispreferred—emptiness is unnatural and marked in
phonology. Therefore, we, again, do not only find a possible analysis but a
possible explanation, too. And the hiatus really is a hiatus: a gap, an empty
prosodic position.
Notice that this description of hiatus-filling, at least in cross-word
examples like Woody Allen, logically follows from the way we analysed
liaison in the previous section. If surface vowel-initial words like Allen
start with empty C’s phonologically (cf. the Linking-R in, e.g., Mister
Allen), then that empty C will get sandwiched between two vowels when
the preceding word ends in one. But from our analyses of liaison and
hiatus it also follows that all surface vowel-initial syllables start with an
empty C, whether they are word-initial (e.g., Allen) or word-internal (e.g.,
hiatus). A further consequence is that in our model, all syllables start with
a C, which may or may not be filled lexically. In the former case, we get
10 For more examples and a detailed discussion, see Balogné Bérces (2011).
Katalin Balogné Bérces
surface consonant-initial syllables, whereas the latter construction
generates surface vowel-initial syllables.
The claims above find additional support in cross-linguistic
observations. For instance, it is well-known that there are two types of
languages, (i) where syllables and/or words have to start with a consonant,
and (ii) where syllables and/or words can start with a vowel.11
In our terms
this means that type (i) does not tolerate empty C’s, whereas type (ii)
does—another explanation to the facts. Furthermore, this model also
predicts that there can not be a language where words have to start with a
vowel (recall, emptiness is unnatural and marked), which is borne out by
the data.
3.1.3 Are there any empty vocalic positions?
Once we introduce emptiness into our model in the form of empty
consonantal positions, the question arises whether their vocalic
counterparts also exist. For the system to be symmetrical and as general as
possible, it is in fact desirable that they do. This section and the next argue
for the theoretical advantages of positing empty V’s word-finally and
word-medially, respectively.
Let us first assume that the phonological patterning of any given
segment derives from its intrinsic nature (determined by, e.g., its internal
structure12
), therefore the same phonological object within the same
system always receives the same treatment. This means that whenever two
segments are combined in the same order, they always enter into the same
relationship. So whatever representation we render to, e.g., the /tr/
sequence in tram, it will be the same in betray. Crucially and non-trivially,
syllable structure depends on the melodic make-up of segments, and
seemingly minor differences, like that between the two liquids /l/ and /r/,
may lead to fundamentally different parsings to the consonant clusters they
produce. When they combine with a preceding /t/, for example, the
sequence with the /l/ is always heterosyllabic (e.g., Atlantic, which
syllabifies into At-lan-tic), while the one with the /r/ is always
11 Word-initial and medial onsetless syllables are separately evaluated. This in fact
generates four logical possibilities for language types, each of which is attested—
cf. Balogné Bérces (2006) and (2008: Ch.7). 12 The subsegmental property in charge is charm in early GP (cf. Kaye et al. 1985)
and complexity in later versions (e.g., Harris 1994).
Emptiness in Phonology
tautosyllabic (e.g., betray, giving be-tray)13
. That these two consonant
clusters differ in structure will also account for why their distributions do
not coincide, either: /tl/ is only found word-medially (there are no English
words starting or ending with it), whereas /tr/ is found word-initially, too
(cf. tram and betray above).
Consider now the syllabic affiliation of the consonants in the /nd/
sequence in abandon. However uncontroversial the example may look,
with a-ban-don as the result, other words like band will cause a theoretical
problem. Namely, if we stick to our principle above, then the /nd/ should
receive the same, heterosyllabic analysis in both abandon and band: if the
/n/ is syllable-final and the /d/ is syllable-initial in abandon, so are they in
band: ban-d. Here, however, the /d/ can only be conceived of as an onset
to a degenerate syllable, with no overt vocalic element, traditionally
considered as the obligatory centre or nucleus of syllables. We conclude,
therefore, that band ends in an empty vowel: ban-dØ, which is referred to
in GP as a final empty nucleus (FEN).
The case of band, however, is neither idiosyncratic nor exceptional. In
fact, all the final two-member consonant clusters of English are also found
word-internally, where—with the exception of sp, st, sk, cf. lisp, fist,
risk14
—they are always unambiguously heterosyllabic. This means that
whenever a word ends in two consonants, the same argumentation leads us
to positing a FEN. Moreover, notice that single consonants do not pattern
differently: the word-final ones (e.g., abandon) are also found in internal
intervocalic environment (e.g., money)15
, where they are usually treated as
belonging to the following syllable.16
This in turn entails that words
ending in a single consonant on the surface also end in a FEN
13 Certain authors (e.g., Kahn 1976) have suggested that the /t/ in words like petrol
is ambisyllabic, that is, it simultaneously belongs to the first and second syllables.
We will not consider that option here. 14 Clusters composed of /s/ plus a voiceless plosive have problematic phonotactics
since they seem to be unconstrained: they are found in all possible positions, word-
initially (cf. sport, stick, skirt), word-internally (cf. Aspen, foster, muscat) and
word-finally (cf. lisp, etc.) as well. None of the other consonant sequences are so
free in their distribution, that is why this kind of phonotactic liberty is usually
considered abnormal, and as a consequence, s+C clusters will be ignored in the
present discussion. 15 Not necessarily also the other way round, though. Typically, word-final
possibilities are more restricted and form a subset of intervocalic ones. 16 Certain authors (e.g., Kahn 1976) have suggested that the /n/ in words like
money is ambisyllabic, that is, it simultaneously belongs to the first and second
syllables. As with the petrol case above, we will not consider that option here.
Katalin Balogné Bérces
phonologically (e.g., abandonØ). It follows, then, that there are no
phonological consonant-final words—whenever such a word is
pronounced, our model posits a FEN in the representation.
But up to this point, we have only seen one kind of argument,
consonant distribution, for the existence of FENs, and the question arises
if there is anything else that they can buy us. There are two sorts of further
support for FENs, coming from the behaviour of final consonants on the
one hand, and from language typology on the other.
First, FENs can explain a long-observed peculiarity of final
consonants, traditionally referred to as extrasyllabicity. This means that
word-final consonants tend to act as if they were invisible to phonology,
that is, as if they were not part of the syllable that they apparently belong
to (hence the name). For example, while in English there are no long
vowels before a tautosyllabic consonant17
(e.g., words like raptor, with a
short vowel before the /p/ closing the syllable, abound, but a word like
*rayptor would be ill-formed18
), there are numerous examples when a
long vowel appears before a seemingly tautosyllabic consonant word-
finally, cf. gape, keep, rote, right, etc. Is the final consonant an exception
to the regularity? Is it extrasyllabic, that is, not part of the syllable
(ga<pe>, kee<p>, ro<te>, ri<ght>, etc)? Note that extrasyllabicity is just
a label and the angled brackets do no more than formalize an observation.
If we assume the FEN, however, i.e., kee-pØ, then it becomes evident that
the consonant in question is not at all tautosyllabic to the vowel in
question, so our expectation that they should interact is ungrounded. Note
that while in kee-pØ the /p/ is not a syllable-final consonant, it is one in
kep-tØ, which makes this latter example totally analogous to raptor.
Another area where consonant extrasyllabicity manifests itself is word
stress assignment. In English, stress rules scan the syllables of the word
from right to left, starting with the last syllable and moving step by step
towards the first one. They are quantity-sensitive, which means that they
prefer to assign (primary) stress on syllables that contain at least a long
vowel, or a short vowel plus a consonant – crucially, syllables ending in a
short vowel are subminimal and tend to be unstressed. The stress patterns
of cá-rry and de-ný differ (the accents denote stress) due to the difference
in the underlined final syllable, which is too “light” in carry but has a long
vowel (a diphthong) and is therefore optimal in deny. Now, if we look at
words ending in a single consonant, such as abandon or astonish, we will
17 This regularity is sometimes referred to as closed-syllable shortening. 18 There is only a handful of exceptions with a well-defined set of consonant
sequences, e.g., angel, chamber, RP master.
Emptiness in Phonology
notice that they are systematically stressed on the second-last syllable (like
carry) as if the final consonant was not there: a-bán-do<n>, a-stó-ni<sh>.
It is not true that final syllables cannot be stressed (cf. dený), neither is it
the case that consonant-final last syllables are never stressed: they are, as
long as they weigh enough even without the final consonant, cf. u-ní<te>,
a-chíe<ve>; tor-mén<t>, e-léc<t>. It is again the word-final consonant
that is “invisible”, i.e., extrametrical in traditional terminology. If we
assume FENs, however, the word-final consonant is not at all contained
within the syllable in question; our syllabification is a-ban-do-nØ, where
the first stressable syllable really is -ban-.
In the model with FENs at the end of all surface consonant-final words
we can make the generalization that all words end in a vocalic position,
which may or may not be empty. It also follows that word-final
consonants are never syllable-final, but they are initial to syllables headed
by FENs (cf. kee-pØ, a-ban-do-nØ); you can only find consonant-final
syllables word-internally, where they are always followed by another
consonant (cf. kep-tØ, rap-tor, a-ban-do-nØ). A significant consequence
is that syllable-final and word-final consonants are two distinct objects,
which has to mean that they pattern in languages independently of each
other. This prediction is borne out by the data of language typology:
according to whether they allow for syllable-final and word-final
consonants, languages exemplify all the four logical possibilities (see
Kaye 1990). In addition, in languages without word-final consonants (i.e.,
without FENs), originally consonant-final loanwords are typically
“augmented” with a vowel at the end—that is, the original FEN gets
phonetically interpreted. In Italian, for instance, where syllable-final
consonants are, but word-final ones are not licit, monosyllables of English
origin with a single final consonant tend to get adapted with an extra
schwa at the end (plus a geminated consonant), e.g., /ikkə/ ‘chic’, /tattə/
‘chat’ (Morandini 2007: 19).19
3.1.4 Are there any empty vocalic positions word-internally?
Let us make the hypothesis that rules (= processes, changes) do not
(normally) delete or insert material, but they involve (a) the movement of
an element already present in the representation to another position already
present in the representation, or (b) the reinterpretation of an entity already
19 For a recent well-presented argument for final empty nuclei, see Cyran (2008).
Katalin Balogné Bérces
present in the representation.20
For example, the adjective separate has
two alternative pronunciations, one is schwa-ful, the other is schwa-less in
the second syllable. In traditional analyses this is accounted for with
recourse to schwa-deletion: the schwa-ful, three-syllable pronunciation is
the basic form of the word, which may lose the second vowel and become
disyllabic (sep’rate). This, however, violates our hypothesis. Moreover,
schwa-ful and schwa-less pronunciations frequently alternate in English,
and it is not always straightforward which of the two should be considered
basic: every, for example, is trisyllabic historically, disyllabic in present-
day speech, but may be trisyllabic again in singing or verse. Words like
umbrella and lovely have consonant sequences /br/ and /vl/ in most
varieties of English which are broken up by a schwa in Cockney (cf. the
famous song Loverly in the musical My Fair Lady). Do we assume
deletion in separate to produce /pr/ but insertion in the /br/ of umbrella? Is
historical deletion in every reversed to insertion in present-day singing? Is
there a difference between the /vr/ of every and the /vl/ of lovely? The
intuition is that all of these cases are simply vowel–zero alternations,
taking place in basically the same environment, and the direction (deletion
vs. insertion) is not of primary importance. Consequently, we will opt for
an analysis in which a vocalic position is posited between these
consonants, and the (non)pronunciation of the vowel is a simple matter of
interpretation.21
It is a word-medial empty nucleus that is either interpreted
(as a schwa) or remains uninterpreted.22
We arrive at representations like
se-pØ-ra-tØ and lo-vØ-ly, and the above hypothesis holds, inasmuch as
nothing is added or deleted; neither pronunciation is derived from the
other, but both are derived from the same abstract form.
Sometimes we find phenomena where neither the deletion nor the
insertion analysis is adequate and the only possible account is the one
20 Similar principles in Government and Binding syntax are called Structure
Preservation (Radford 1988: 538ff) and Projection Principle (Radford ibid: 548ff),
introduced to constrain derivations. 21 This way of describing vowel–zero alternation goes back to Anderson (1982)’s
analysis of French schwa. 22 Both, esp. the latter, happen under strictly defined conditions, cf. separate (v),
where the conditions on non-interpretation are not met and the disyllabic
pronunciation is not an option. In fact, empty vowels in general do not come
unrestrained—see GP literature, esp. Charette (1991), Kaye (1990), Kaye et al.
(1990).
Emptiness in Phonology
referring to empty vowel slots. An example comes from Hungarian.
Consider the data in (15).23
(15) torony ‘tower’ tornyok ‘towers’
What we see here is two forms of the word (more precisely, morpheme)
meaning ‘tower’, the singular with two o’s, and the plural ending in -rny
(torny-ok). The question is, which is the base form? We may be tempted to
say that it is the singular, and in the plural the second o (boldfaced in (15))
is deleted from between the r and the ny. The problem such an analysis has
to face is that there are other morphemes with exactly the same vowel and
consonants in which the o does not alternate, see (16).
(16) szurony ‘bayonet’ szuronyok ‘bayonets’
A way out would be to posit the plural (torny- and szurony-) as the basic
form of Hungarian nouns, and assume a rule that inserts an o between the
consonants in torny- in singular, when the -rny sequence would otherwise
be final. But then, how do we distinguish torony from words like szárny
(in (17)), where the same -rny sequence is final but unbroken?
(17) szárny ‘wing’ szárnyak ‘wings’
Whatever strategy we choose, stems of the torony type need to be marked
for their alternating vowel, or else they collapse either with szurony-type
nonalternating stems or with szárny-type stems with a final consonant
cluster. The most logical way of marking, then, is simply designate the
position of the alternating vowel, in the form of an empty nucleus—in the
same way as we did for separate and company above. Consequently, the
two consonants will not be adjacent in torØny and szurony, unlike szárny.
The conclusion of the above discussion of empty nuclei is the
following: all words end in a vocalic slot (which may or may not be
empty; if it is, it is called a FEN); in addition, certain words also have
empty vowels in medial position. The obvious question arises if there are
any empty vowels in the third possible position, word-initially. In fact, it
has been suggested (by Kaye 1992, and then by Gussmann 2002: 114–
115) that certain languages that do not tolerate word-initial /s/ plus
23 Most of the letters in Hungarian orthography (roughly) correspond to the IPA
symbols of the sounds they represent. The other graphemes in our examples: <ny>
= //, <sz> = /s/, <s> = //. Accents denote vowel length.
Katalin Balogné Bérces
voiceless plosive clusters (cf. footnote 14) insert a vowel to their left
exactly because such clusters are always heterosyllabic, even at the
beginning of the word, and the /s/ is final to a syllable headed by an empty
nucleus. It is that empty nucleus that is somehow permitted24
in languages
like English or present-day Hungarian, but has to get phonetically
interpreted in Spanish, (Old) French, or in the Slavic loanwords of Old
Hungarian. Cf. (18).
(18)
a. Spanish and (Old) French
Lat. status Sp. estado, Fr. état ‘state’, etc.
Sp. España (cf. Eng. Spain), etc.
b. Hungarian
iskola ‘school’ (cf. Ru. школа ‘ibid.’)
István ‘Steven’
asztal ‘table’ (cf. Ru. стол ‘ibid.’), etc.
Accordingly, it is possible to assume initial empty nuclei in words like
Latin Østatus or English Østate, ØSpain, Øschool, ØStephen, etc. We
conclude, then, that the existence of empty vocalic positions seems to be
justified in all possible word positions: final, medial, and initial.25
3.2 Emptiness below the segment
This section deals with the possibility of having empty components in
subsegmental, that is, melodic structure (cf. the definition of the melodic
tier in Section 3.0 above). It is a phonological commonplace that segments
are not atomic, but rather they are composed of so-called features or
primes. The classical view is that these primes are binary (= two-valued),
i.e., segments may be specified either as having (indicated by +) or lacking
(indicated by –) a given phonetic property. Nasal articulation, for example,
is expressed with reference to the feature [nasal], in such a way that nasal
sounds like n are [+nasal], while oral sounds like d are [–nasal]. In the
24 Kaye (1992) dubs it “magical licensing”. 25 For the claim that all surface consonant clusters contain an empty V and all
surface vocalic sequences (including long vowels) contain an empty C, see
Lowenstamm (1996), Scheer (2004), etc. For the claim that all words (may) start
with an empty CV-unit, see Lowenstamm (1999), while Szigetvári (1999) claims
that all words begin with an (empty or nonempty) vocalic position.
Emptiness in Phonology
same vein, vocal cord activity (voicing) will have two basic states, sounds
like d are specified as [+voice], whereas sounds like t are [–voice].
There is a third option26
, though, and that is when a given segment is
neither + nor – for a given category. This means that the sound remains
unspecified for the feature: it is missing from its internal structure because
it is a feature that is associated with a kind of phonological patterning
which is irrelevant for the given segment. This state of affairs is well-
known in mainstream phonological theory and is referred to as
underspecification27
: the sound segment remains underspecified during
phonological computation, and the missing values are only adduced by a
late phonetic spell-out component.
We will bring a very simple example from Hungarian voicing
assimilation. Ignoring a few complications, we can say that in Hungarian
all sequences of obstruents (plosives like /t/, fricatives like /s/, and
affricates like /ts/) must agree in voicing. Should feature values clash,
upon morphological concatenation for instance, the consonant on the left
assimilates in voicing to the one on the right, that is, so-called regressive
voicing assimilation takes place. As is shown in (19a–b), stem-final /s/
becomes /z/ before /b/, while /z/ merges with /s/ before /t/.
(19)
a. mész ‘whitewash’ /-s/; méz ‘honey’ /-z/
b. -ben ‘in’: mészben = mézben /-zb-/
-től ‘from’: mésztől = méztől /-st-/
c. -nél ‘by’: mésznél /-sn-/ ≠ méznél /-zn-/
Certain consonants like nasal /n/, as well as all vowels, (i.e., the so-called
sonorants), however, are unspecified for [voice]: nothing happens in (19c),
the /n/ is unable to transfer its voicing onto the preceding consonant, and
as a result the contrast between /s/ and /z/ is maintained. That is, [voice]
for such sounds is phonologically inactive or inert. The standard
explanation for this is that they are unspecified for that feature: sonorants
are neither [+voice] nor [–voice] phonologically, therefore they are unable
to participate in voicing assimilation, and the vocal cord vibration which
accompanies their articulation, “spontaneous voicing” as it is commonly
referred to, only arises late, out of phonetic necessity (since the vocal
cords do need some kind of instruction during phonetic implementation
26 The fourth combination, when a segment is both + and –, is ruled out by the
physical impossibility of simultaneously having and lacking the same property. 27 See Steriade (1995) for an overview.
Katalin Balogné Bérces
after all, and what seems compatible with the relatively unobstructed
airstream mechanism of sonorants is vocal cord vibration). Therefore, as
shown in (20), besides the + and – feature specifications we can assume
the boldfaced zero as a third option.
(20)
sz (/s/) z b t n ...
[voice] – + + – Ø
...
4. Conclusion
In this paper I hope to have provided an overview of (some of) the
possible interpretations of emptiness in phonology, tackling the issue of
the real sounds of silence, the real silence of sounds. The major conclusion
is intended to be the claim that phonology is abstract enough to be able to
accommodate entities with no phonetic embodiment, and empty categories
are present in phonology, similarly to morphology and syntax, in all forms
(vocalic or consonantal) at all levels (subsegmental or suprasegmental).
We have to accept that phonetic/physical reality may diverge from
phonological/mental/analytical reality: what you see may not be what you
get in linguistics, just as in natural science. Most specifically, like in e.g.
syntax, where surface and underlying sentence boundaries do not
necessarily coincide (cf. Who do you want to beat Ø), we have seen that
syllable and word edges do not necessarily coincide, either (cf. Øs.te.pØ
for step). While classical syllabic theory has managed to integrate this
view into its model to a certain extent and recognizes the existence of
certain empty categories in certain phonological positions (esp. empty
syllable onsets and floating consonantal melodies), and underspecification
has been around for half a century, some of the empty categories presented
above (esp. empty nuclei) remain in the toolbox of non-mainstream
theories like Government Phonology.
WORKS CITED
Anderson, Stephen R. “The Analysis of French Shwa: Or, How to Get
Something for Nothing.” Language 58.3 (1982): 534–573.
Aronoff, Mark and Kirsten Anne Fudeman. What is Morphology? Oxford:
Blackwell, 2005.
Bauer, Laurie. English Word-formation. Cambridge: CUP, 1983.
Emptiness in Phonology
Balogné Bérces, Katalin. “What’s Wrong with Vowel-initial Syllables?”
In: SOAS Working Papers in Linguistics 14 (2006): 15–21.
———. Strict CV Phonology and the English Cross-word Puzzle.
Saarbrücken: VDM Verlag Dr. Müller, 2008.
———. “Filling the Hiatus: A Changing Face of English.” In: Balogné
Bérces, Katalin, Földváry, Kinga, Schandl, Veronika (eds.) Faces of
English. Pázmány Papers in English and American Studies. Volume 5,
30–41. Pázmány Péter Catholic University, Piliscsaba, 2011.
Charette, Monik. Conditions on Phonological Government. Cambridge:
CUP, 1991.
Chomsky, Noam and Morris Halle. The Sound Pattern of English. New
York: Harper and Row, 1968.
Cyran, Eugeniusz. “Consonant Clusters in Strong and Weak Positions.”
In: Brandão de Carvalho, Joaquim, Tobias Scheer and Philippe Ségéral
(eds.) Lenition and Fortition, 447–481. Berlin: Mouton de Gruyter,
2008.
Goldsmith, John A. Autosegmental Phonology. PhD dissertation, MIT,
1976. (Published by Garland Press, New York, 1979.)
Gussmann, Edmund. Phonology. Analysis and Theory. Cambridge: CUP,
2002.
Haegeman, Liliane and Jaqueline Guéron. English Grammar. A
Generative Perspective. Oxford: Blackwell, 1999.
Harris, John. English Sound Structure. Oxford: Blackwell, 1994.
Honeybone, Patrick. Handout to course on Modern English Phonology.
Week 2. University of Edinburgh, 2010.
Kahn, Daniel. Syllable-Based Generalizations in English Phonology. PhD
dissertation, MIT, 1976. (Published by Garland Press, New York,
1980.)
Kaye, Jonathan. “‘Coda’ Licensing.” Phonology 7 (1990): 301–330.
Kaye, Jonathan. “Do you Believe in Magic? The Story of s+C Sequences.”
SOAS Working Papers in Linguistics and Phonetics 2 (1992): 293–
313.
Kaye, Jonathan. “Derivations and Interfaces.” In: Durand, Jacques and
Francis Katamba (eds.) Frontiers of Phonology: Atoms, Structures,
Derivations, 289–332. Harlow: Longman, 1995.
Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud. “The
Internal Structure of Phonological Representations: A Theory of
Charm and Government.” Phonology Yearbook 2 (1985): 305–328.
Katalin Balogné Bérces
Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud.
“Constituent Structure and Government in Phonology.” Phonology 7
(1990): 193–231.
Lowenstamm, Jean. “CV as the Only Syllable Type.” In: Durand, Jacques
and Bernard Laks (eds.) Current Trends in Phonology: Models and
Methods, 1996. European Studies Research Institute, University of
Salford Publications: 419–442.
Lowenstamm, Jean. “The Beginning of the Word.” In: Rennison, John and
Klaus Kühnhammer (eds.) Phonologica 1996. Syllables!?, 1999. The
Hague: Holland Academic Graphics: 153–166.
Matthews, Peter H. Morphology. Second edition. Cambridge: CUP, 1991.
Morandini, Diego. The Phonology of Loanwords into Italian. MA
dissertation, University College London, 2007.
Radford, Andrew. Transformational Grammar. Cambridge: CUP, 1988.
Scheer, Tobias. A Lateral Theory of Phonology. Vol 1: What is CVCV, and
Why Should it Be? Berlin: Mouton de Gruyter, 2004.
Spencer, Andrew. Morphological Theory. Oxford: Blackwell, 1991.
Steriade, Donca. “Underspecification and Markedness.” In John A.
Goldsmith (ed.) The Handbook of Phonological Theory, 1995. Oxford:
Blackwell: 114–174.
Szigetvári, Péter. VC Phonology: A Theory of Consonant Lenition and
Phonotactics. PhD dissertation, MTA/ELTE, Budapest, 1999.