Beruflich Dokumente
Kultur Dokumente
Phonology......................................................................................................................1
1 Phonology..............................................................................................................1
2 Phonology and phonetics.......................................................................................2
3 Syllables and segments..........................................................................................3
3.1 Syllables.........................................................................................................3
3.2 Syllable parts..................................................................................................5
3.3 Segments........................................................................................................5
4 Phonotactics...........................................................................................................7
4.1 Dependent syllables and syllable classification.............................................7
4.2 Dependent consonants....................................................................................8
4.3 Adjacency.......................................................................................................8
4.4 Ambisyllabicity and floating segments........................................................10
5 Realisation............................................................................................................10
6 Sandhi...................................................................................................................10
7 Optimality............................................................................................................10
1 Phonology
This draft paper is an attempt to fill the gap in Word Grammar theory where there
should be a theory of phonology – none too soon, either, because a theory of language
which doesn’t cover phonology is pretty seriously deficient. The gap is especially
important because some linguists (e.g. Halle & Bromberger, 1989) claim that
‘phonology is different’; e.g. even if the rest of the grammar is declarative, at least
phonology is procedural. One of the main claims of WG is that the whole of language
is formally similar, so phonology is a potential challenge which needs to be addressed
at least in outline.
What I shall suggest below extends the formal and logical apparatus of
semantics, syntax and morphology down into phonology, so – if it works – it does
answer this challenge. Here in outline are the main ideas:
Phonology is itself the interface with phonetic ‘substance’ – acoustics and
motor programs – without any separate mediating level of phonetics. See 2.
Phonological structure includes syllables; in fact, syllables are the basic units
of phonology, with a status comparable to the word in syntax. E.g. {hack} is
realized directly by the syllable /hak/, and not by its constituent segments. See
3.
Phonological structure is defined by independent ‘phonotactic’ rules – i.e. in
Jackendoff’s terms, phonology is a ‘generative’ level like syntax. This is
where we define the notion ‘possible word’ (except that the units concerned
are actually forms, not words). E.g. in English, syllables like /hak/ are ok,
but /ha/ or /hah/ are not.
Forms are realized by particular examples of phonological structures.
(Remember that in WG, phonology realizes forms, not words.) Since these
structures are generated by phonotactic rules they are fully ‘explained’, and
the only arbitrary element is the realization relation. E.g. {hack} is realized
by /hak/.
Although forms are realized by permitted phonological structures, they may
not fit the default patterns well, or even at all. For such cases there are
‘phonological rules’ which replace the normal realization by something
which fits the default more closely. These non-default patterns arise in two
ways:
o at the boundaries between adjacent forms. E.g. {feel} is realized by
/fi:l/, but the /l/ ‘prefers’ to be an onset so if it is immediately before a
vowel it moves into the following syllable, as in selling (with clear [l]).
o Even among the permitted phonological structures, some structures
may be ‘better’, i.e. nearer to the default prototype, than others, so the
phonological rule may apply optionally. E.g. even in rhotic accents of
English, /r/ is often omitted before a consonant as in Labov’s famous
fourth floor; presumably this shows that coda /r/ is non-default.
These phonological rules are the biggest challenge for a declarative theory
such as WG, because they seem to involve procedures for changing the normal
pronunciation into a different one. However, I shall suggest that they can be
handled by means of default inheritance.
Although these ideas have never before been applied to WG, none of them are new or
original. For example, the idea of combining realization rules with phonotactic rules
goes back (at least) to Lamb’s early stratificational grammar.
Unfortunately, I shall have virtually nothing to say about intonation. I’m
aware of this gap, and regret it deeply, but we have to start somewhere and intonation
is even harder to understand than segments and syllables.
In conclusion, therefore, the units of phonology are sounds which are fully
specified for both acoustic and articulatory properties, but which in non-default
situations may acquire different properties.
3.3 Segments
In the simplest kinds of syllable, the onset and the rhyme are each just one segment
long, so a simple syllable consists of just one consonant followed by one vowel.
Furthermore, almost every consonant can be used as an entire onset (exception: those
that can only appear in coda position) and most vowels can be used as an entire rhyme
(exception: vowels that require a coda consonant.) For example, the syllable /pɜ:/ has
a single-segment onset /p/ and a single-segment rhyme /ɜ:/.
One way to interpret this structure is in terms of a strict hierarchy of parts, in
which the syllable has two parts, each of which in turn has one part. In this
interpretation, the syllable-parts remain conceptually distinct from the segments.
However, this is not how I shall interpret it. Instead, I shall assume that each part of
the syllable is just a single segment, and any additional segments are attached as
dependents to one of these; for example, in /spɜ:t/ the onset is a /p/ which supports an
/s/ and the rhyme is an /ɜ:/ that supports a /t/. I explain this approach below in 4.
Given this analysis, segments and syllable-parts are the same thing, just as syllables
and poly-syllables are the same and, in syntax, words and phrases are the same thing.
Moreover, in spite of ‘polysystemic’ approaches which stress the different
options available at different structural points, there does seem to be a strong tendency
for a language to draw on a single common list of consonants in the onset and coda of
a syllable; for example, the lists are almost the same in English (with exceptions – for
me, /h/ and /r/ are only in onset and /ŋ/ only in rhyme), and Arabic has a curious lack
of /p/ in both places. These facts seem to suggest that a language has a single list of
vowels and a single list of consonants which are in some sense ‘the same’ when found
in different positions. But nothing much turns on this, and some languages may
indeed have quite different choices in onset and coda positions, suggesting two
fundamentally different lists of consonants. An isa hierarchy can accommodate this as
easily as a single list.
Are segments psychologically real? The evidence from literacy research
(mentioned above) suggests that segments only become real as a result of learning an
alphabetic writing system, but this evidence may only reveal how accessible
segmental structure is to conscious introspection. Internal structural evidence seems to
suggest that segments are psychologically real in any language, regardless of its
writing system (or lack of one). For example, Beja, which has no writing system at
all, includes phonological rules which make specific reference to segments; e.g. one
such rule suppresses a pre-consonantal /n/ in the onset except where it can be attached
as coda to a preceding vowel: so /nga:l/, ‘one’, is pronounced /ga:l/ in /tak ga:l/, ‘one
man’, but the /n/ surfaces after the definite article /u:/ in /u:tak u:nga:l/, ‘the one man’.
In this rule the /n/ is a ‘floating’ segment which has to belong to a preceding syllable
based outside the form that it helps to realize. It is hard to imagine how a floating
segment could be handled if segments are not available for rules.
The conclusion so far, then, is that a language’s phonology includes the
following units:
a full inventory of syllables, classified according to parameters such as
prominence and openness;
a full (but much shorter) inventory of the segments that act as syllable-parts,
classified as consonants and vowels and possibly according to other
parameters such as tone and length and also for the familiar contrasts of voice,
place, and so on.
This analysis is summarized schematically in Figure 1. In words, a form’s realization
(more precisely, its pronunciation) is a syllable, which may support other dependent
syllables and which has two parts, an optional consonant as onset and an obligatory
vowel as rhyme, each of which may also support dependent segments.
F orm
p r o n u n c iatio n • C onso nant
o nset
• Syllable Segm ent
rhym e
dependent 1 Vow el dependent
• •
/b n : n /
e e
/ l k t r f k ei n/
3 2 3 3 1 2
The main point of the discussion is that a syllable may have one or more
dependent syllables. These dependencies are themselves properties of the syllables
concerned, which distinguish these syllables from other instances of the same
syllable-type; for example, the syllable /lɛk/ in electrification isn’t simply the
syllable /lɛk/, but an example of this syllable with /ɪ/ as its ‘pre’ and /trɪ/ as its ‘post’;
and similarly, the main syllable is a very specific example of /keiʃ/ this particular kind
of /lɛk/ as its ‘pre’ and an example of /ə/ as its ‘post’. What this means is that the main
syllable ‘stands for’ the entire structure in just the same way that the syntactic root of
a sentence stands for the entire sentence; so as far as the form {electrification} is
concerned, the answer is simply this one syllable. In a network analysis this is an
important conclusion, because the realization must be a single node rather than a
collection of them.
/tr n d / // /l i /
/t r n d l i/
+1 +1 +1 +1 +1 +1
+1
4.3 Adjacency
One characteristic of phonology which makes it different from other levels is its role
as an interface with acoustic and motor programs. One consequence is the importance
of mere adjacency of segments, since adjacent segments inevitably influence each
other’s acoustic and articulatory implementation. For example, even though onsets
and rhymes combine freely, every rhyme is influenced by its onset (and vice versa)
because of the transition from one segment to the next.
Most of these details can presumably be left to the acoustics and to
articulatory programs, but some of them seem to belong to phonology. For example,
when a nasal assimilates to the place of the following consonant, this is rule-governed
rather than mechanical because not every nasal does assimilate in this way - /n/ does,
but /m/ doesn’t.
These interactions between adjacent segments cannot be handled in terms of
dependency structures because the segments concerned may have no direct
dependency links. For example, in thin people, the /n/ and the following /p/ are not
even in the same syllable and there is no reason to assume any kind of dependency
between them; and yet the /n/ assimilates to the /p/ giving /mp/ instead of /np/.
Whatever rule is responsible for this assimilation must be able to recognise adjacent
segments, so this information must be made available in the phonological structure
alongside the dependencies discussed above.
What is needed is a relation between adjacent segments: ‘next’ (labelled ‘+1’
in diagrams). If dependency relations are themselves directional (thanks to the
contrast between ‘pre’ and ‘post’), these links are strictly redundant, but they
nevertheless need to be included for the sake of rules that are sensitive to adjacency.
Adjacency links are certainly learnable – indeed, they are much more learnable than
dependency relations, which may be based on mere adjacency – so I assume that they
are also part of a form’s stored realization, along with its dependency and
constituency structure. The phonological structure for trendily is given in Figure 5 is
complete, except for the segmental transcription which I have simplified by omitting
the isa links to the segment types which I discuss below in 5.
/t r n d l i/
+1 +1 +1 +1 +1 +1
+1
/j l/ /l ou/
dep
/j/ / / /l / /o u /
onset
dep •
#
• d a r k /l/
0
The English /r/ is even more fussy about its syllabic status. In all accents it
prefers to be an onset, and even in rhotic accents it is subject to deletion in coda
position (as in Labov’s famous fourth floor investigation); but in non-rhotic accents it
refuses to be anything but an onset. In the latter accents we also find ‘intrusive r’ in
examples like the idea of it or saw it, so the best analysis is a purely phonological one
which inserts /r/ after ‘r-coloured’ vowels such as /ə/ and /ɔ/. The ‘rule’ is expressed
in Figure 8, which combines an r-coloured vowel with a ‘floating’ /r/, a segment
which has no dependency relation to this vowel but which is simply the next segment,
attached as onset to the next syllable. When this is not possible, the /r/ can be omitted.
/r /
+1 onset
r -c o lo u r e d • •
vow el
+1
r -le ss • C o nsonant
5 Realization
Phonotactic rules are purely phonological, so they need to be complemented by a set
of realization rules which show how the permitted phonological structures are
exploited by the forms that need a realization. (Strictly speaking, the realization rules
that relate form to phonology are ‘pronunciation’ rules, a sub-case of ‘realization’, in
contrast with ‘spelling’ relations from form to orthography and ‘formation’ relations
between words and forms; but I shall use ‘realization’ here.)
5.1 Forms
The forms that need realization are generated by morphology, so they include not only
simple forms such as {sell} but also more complex ones such as {{sell}{ing}} or
even {{{de}{{stabil}{iz}}}{ing}}. The largest structures are called ‘wordforms’
and, of course, the smallest are called morphemes (whose traditional signal is the
curly brackets that I use for all forms). Wordforms include not only the conventional
roots and affixes that we separate by word-spaces, but also clitics that are attached to
the word concerned. However, roughly speaking, a wordform is the form of a word.
Separate wordforms are not related to each other at this level except in terms
of adjacency, but internally they have a rich whole-part structure shown by the
brackets. This structure is important for the rules of morphology, but has no
significance for phonology so realization rules simply ignore it, paying attention only
to adjacency. For example, since {sell} is realized by /sɛl/ and {ing} by /ɪŋ/, {{sell}
{ing}} merely concatenates these two syllables as shown in Figure 9 (which does not
show the effects of the ambisyllabification discussed in 4.4). In words, {{sell}{ing}}
contains two morphemes, and example of {sell} and {ing} respectively, each of which
is realized by an example of its default realization, which in turn is an example of the
syllable indicated. The order of the syllables follows that of the morphemes and is
indicated by the ‘+1’ (‘next’) relation.
syllable
/s l / / nj /
{sell} {ing }
r e a l i za ti o n
p ar t2
p ar t1 • •
{{sell}{ing }} • •
r e a l i za t i o n
• •
+1
5.2 Realizations
At least in simple cases, a form is realized by a single unit of phonology. By default, a
morpheme corresponds to an entire syllable, but exceptions are found in both
directions. Some morphemes are realized by less than a syllable (e.g. {s}) and many
more are realized by a ‘modified’ syllable – i.e. a syllable with dependents.
Of course, not all of morphology involves simply combining one morpheme
with another so we have to be able to accommodate patterns in which morphology
affects phonological structure directly – e.g. by changing a vowel (hide ~ hid),
shifting the main stress (convict as verb or noun), affecting the voicing of a consonant
(breath ~ breathe) or even determining the entire syllable structure as in semitic
interdigitation mentioned earlier, whereby the root /ktb/ surfaces as /katab/ in one
form and as /ka:tib/ in another. These changes are phonological, but the rules
responsible for them are morphological.
Moreover, such changes are not mere lexical freaks, like the suppletion of go
and went; there are important generalizations which a good analysis will reveal. For
example, although English irregular verbs show a messy and unpredictable spread of
vowel changes, the important generalization is that what changes is always the main
vowel (and sometimes its coda, as in stand ~ stood). This suggests that the main
vowel is somehow more accessible to the morphology than the rest of the
phonological structure. One way to explain this is to postulate a direct link between
the form and the vowel; for example, the form {hide} has a direct ‘vowel’ link to /ai/,
which would otherwise be linked to it only indirectly (via realization, isa and onset
links), as shown in Figure 10. The extra link is labelled simply ‘vowel’, but it can be
analyzed as a particular specialization of the more general ‘realization’ relation. I
assume that it can be inserted by a very general rule so that it is available to every
form, and not just for those that undergo vowel alternation.
{hide} /h a i d /
r e a l i za ti o n o nset
• • • •
vow el
/h / /a i / /d /
Figure 10: The sound /ai/ has a special relation to the form {hide}
What this special realization relation allows is realization by vowel change,
which greatly simplifies the statement for any irregular form. Past-tense forms are
derived in WG by a morphological relation called ‘ed-variant’, and by default the ed-
variant of a form is another form whose first part is a copy of that form, with an
example of {ed} as its second part. Figure 11 shows how easily an irregular ed-
variant can be added to this network. (The relation ‘copy’ is defined elsewhere as the
relation between two forms that share the same realization; it is tempting to replace it
by a simple ‘isa’ relation, but this produces logical contradictions.)
e d -var ian t
fo r m •
copy p ar t1 p a rt2
• •
e d -var ian t
{hide}
copy •
vow el vow el
• •
/a i / //
This section has briefly reviewed the realization relations that complement the
purely phonological phonotactics. Since every realization involves a selection from
the structures permitted by phonotactics, the latter acts like the well-formedness
conditions of old-fashioned generative phonology. However, we also saw that
phonological rules such as those for /l/ and /r/ build directly on the phonotactics, so
the phonotactics provides a motivation for all the details of realization and
phonological rules.
6 Optimality
Finally, why am I not using some version of Optimality Theory, like most real
phonologists? It would have been much more comfortable to follow the herd, just as
in syntax it would have been easier to use phrase structure rather than dependency, but
I having too many reasons for rejecting OT. None of them are original, but here they
are:
To my mind, the most attractive feature of OT in phonology is the list of very
general constraints on phonological structure. As a summary of known
universals, or universal trends, this is very helpful. But why assume that this
list is part of linguistic competence?
o The constraints all tend to help in processing, so they can be
interpreted as functional pressures on how grammars evolve; why
should we assume instead that they are part of competence?
o If they are not functional pressures, they must be innate; but they face
all the same objections as the supposed innate principles of syntax,
which I find totally implausible.
One of the basic principles of OT is that the best candidate wins, so there will
always be a winner. It may be no coincidence that this theory started in
phonology, because so far as I know there are no unpronounceable words; the
whole point of phonology is to provide strategies for circumventing problems
so that every word does in fact have a pronounceable realization. But the same
is not true in morphology and syntax: there are genuine gaps where there is in
fact no realization for a particular combination of inflectional properties (my
favourite example is the impossible form *amn’t in English English) or for a
particular interaction of two constructions (e.g. how do you pluralize car in
They have too big a car?). If OT demonstrably fails in morphology and
syntax, it looks less attractive in phonology.
I find it very hard to imagine how we could apply an OT phonology in actual
processing, given the need in at least some versions of OT to generate an
open-ended set of candidates and then pick the winner. In contrast, a network
theory such as WG provides a very plausible basis for a processing model
using spreading activation to select the winner by the Best Fit Principle.
I don’t see how OT could be reconciled with network architecture. Since I do
see how network architecture accommodates every level of language, and does
so at least as well as any other theory I know, I don’t see any reason to jettison
networks in favour of OT.
As I said above, these complaints are neither new nor original, and advocates
of OT have no doubt grappled with them, so it may be easier than I think to
incorporate OT into my framework of ideas. And of course, there may be
overwhelming evidence in favour of OT that I’m not aware of, in which case my
current ideas may be seriously wrong. But as things stand, these are my reasons for
preferring a network theory and for thinking it is a better candidate than OT.
Reference List