8.2 PHONOLOGICAL PRELUDE: AUTOSEGMENTAL PHONOLOGY
My intention is not to turn you into an accomplished autosegmental phonologist in these short paragraphs, but rather to provide you with a basic grasp of the issues in the model of 'phonology with tiers' that is a pre- requisite to understanding prosodic morphology.
Autosegmental phonology has several precursors, as John Goldsmith, the originator of the model, points out (1990: 1-4). It is a direct descendant of the theory of generative phonology, given its fullest expression in The Sound Pattern of English (SPE) by Chomsky and Halle in 1968. Auto- segmental phonology continues to regard as central the goals of phonologi- cal investigations set out in SPE, although it has rejected many of the original assumptions, formalisms and principles of SPE.
Furthermore, the work of the London School on prosodies (see Firth, 1948; Palmer, 1970) and that of some leading American structuralists such as Hockett (1947, 1955) and Harris (1944) on long components foresha- dows the central tenet of autosegmental phonology to some extent. Like linguists of these earlier schools, autosegmental phonologists hypothesise that phonological representations consist of several independent, parallel tiers, i.e. levels of representation. Autosegmental phonology articulates this insight more clearly than its precursors. It also provides the formal means of representing these autonomous tiers and of showing how they link up with each other.
154
Phonological Prelude: Autosegmental Phonology 155 In our discussion of English stress, we have already implied a multi- tiered analysis. We have assumed that stress is a prosody, i.e. it is a phonological element that is not an inherent feature of consonants or vowels but rather a property of the entire word. Hence, it can hop from syllable to syllable when certain stratum 1 suffixes are added as in [8.1]:
[8.1] 'democrat 'pompous 'adjective
demo 'cratic pom'posity adjec'tival
If stress were an integral part of the segments in a syllable it would not enjoy such mobility and independence.
8.2.1 Autosegmental Phonology: Mapping Principles
The theory of autosegmental phonology was initially used to describe tone.
Although it has been extended to cover other phenomena, its essential properties are still best illustrated using tone. So, before seeing how the theory was used later to describe morphological data we will come to grips with its main characteristics through a brief account of its application to phonological phenomena, in particular tone.
As we saw above, stress is independent of consonants and vowels.
Similarly, in tone languages, tone can be shown to be an independent prosody, and not an integral part of vowel and consonant segments. Tones are represented on the tonal tier and vowels and consonants on the segmental tier. Processes affecting elements on one tier may in some cases have no impact on elements on a different tier. For instance, when vowels are deleted, the tones associated with them are not necessarily lost. Thus, as seen below, in Luganda, when two vowels are adjacent, the first is deleted unless it is high (in which case it becomes a glide [y] or [w]) but the tone associated with it is not necessarily affected:
[8.2] Segmental tier: kusa ebyo ~ kuse:byo 'to grind those'
I I I
Tonal tier: LH L H L H LH
Note: Land H stand for L(ow) and H(igh) tone. The lines link vowels with tones. An unbroken line indicates pre-linking and a broken line shows the establishment of a fresh link (see [8.4] below).
156 Prosodic Morphology
Observe that when the vowel at the end of kusa 'to grind' is deleted, its high tone is preserved and inherited by the e of ebyo 'those'. If tone were an inherent property of the vowel, the deletion of the vowel would mean the demise of the tone as well. The survival of the tone bears testimony to its independence from the vowel.
The choice of the name autosegmental phonology for this model is intended to reflect the fact that phonological representations consist of segments like stress, tone, vowels and consonants that appear on auton- omous tiers. On their tiers, all phonological elements, be they tones, consonants, vowels, etc., behave as though they were 'segments'. They cannot occupy the same space. They must follow each other in a linear sequence.
The tiers in terms of which phonological representations are organised are independent, but they are not isolated from each other. On the contrary, they are linked in complex hierarchical structures and are capable of interacting with each other (cf. Goldsmith, 1976, 1979, 1990;
Clements and Keyser, 1983). In this theory articulation is seen as the bringing together of different autonomous gestures resulting, for instance, in voicing, lip rounding, high pitch that make up a speech sound.
Tone (and stress) are best represented on a separate tier from vowels and consonants which belong to the segmental tier. Mediating between these tiers is the skeletal tier (or CV-tier). The skeletal tier holds together the different phonological tiers in the same way that the binding holds the pages of a book together. Association lines indirectly link stress or tone to actual vowel and consonant sounds through the skeletal tier, as shown in [8.3] below.
In [8.3a] the skeletal tier is represented by x's (with each x standing for a segment like a consonant or vowel (cf. Hyman, 1985). An alternative representation, with the skeletal tier represented by C and V slots (as in [8.3b], is proposed by Clements and Keyser (1983).
[8.3] a. Tonal tier: H L L H
I I I I
Skeletal tier: X X X X X X
I I I I I I
Segmental tier: a f a f
b. Tonal tier: H L L H
I I I I
Skeletal tier: v c v v c v
I I I I I I
Segmental tier: a f a f
The rest of the notation used in autosegmental phonology is explained in [8.4]:
Phonological Prelude: Autosegmental Phonology 157 [8.4] a. An unbroken association line indicates pre-linking, i.e. prior
association in the lexicon of elements on separate tiers
b. A broken association line indicates linking, i.e. the creation of an association line
c. 1 A crossed-through association line shows delinking, i.e. the sever-
T ence by rule of an association line linking elements on different tiers.
d. [] A left bracket shows the left boundary and a right bracket shows a right boundary. (In this book these brackets will only be used where their presence is particularly important for the point under consideration.)
e.® A circle around an item indicates that the item has been deleted.
The theory does not require a one-to-one association of elements on different tiers. Elements at any one tier may be linked one-to-many with elements at another tier. This can be seen in the tonal examples in [8.5a] of rising tones or falling tones. Such tones are called contour tones. They are created when independent high and low tones are simultaneously linked to a single vowel (or to a single tone-bearing consonant).
[8.5] a. b
H b. k:S
belt mbu mba
bclc mbu mba rna
y 1\ f\
L H L L H
'war' k:Sma
'trousers' btltma
'owl' mbtima
'rice' mbiima
'on war' 'on trousers' 'on owl' 'on rice'
(Mende data from Leben, 1978) Note: H marks a high tone (')
- - L marks a low tone (')
HL marks a falling contour tone (A) LH marks a rising contour tone C)
Blank marks 'no tone
Conversely, the theory allows elements at one tier to remain unassociated with any item at another tier. Thus, in tone languages, sometimes a syllable representing a particular morpheme may not be associated with any tone in the lexicon. Instead, it may acquire later the tone of an adjacent syllable.
In Mende, as seen in [8.5b], the morpheme -ma 'on' is underlyingly
158 Prosodic Morphology
toneless. It gets its tone when the last tone of the stem to which it is suffixed spreads to it.
If phonological representations consist of independent tiers, how are these tiers eventually brought together in the production of speech? Are there any constraints on the ways in which elements on different tiers can be linked? In order to account for the constraints on the linking of elements on various tiers, autosegmental phonology has always incorporated Mapping Principles, which are claimed to be part of Universal Grammar. Mapping principles go under the names of Universal Linking Conventions (Pulleyblank, 1986) and the WeU- formedness Condition (-Phonotactic) (WFC) (Goldsmith, 1976, 1990;
Oements and Ford, 1979).
The WFC (for tone) is expressed in this fashion by Goldsmith (1990: 319).
[8.6] 1. All vowels are associated with at least one tone.
2. All tones are associated with at least one vowel.
3. Association lines do not cross.
The WFC is concerned with safeguarding phonotactic constraints. Its role is to ensure that restrictions on the combination of phonological elements are not violated.
As we mentioned above, mapping principles are referred to by some not as the WFC, but as Universal Linking Conventions which determine how elements on different tiers may be associated. There are two contrasting versions of the Universal Linking Conventions or WFC in the literature.
One version, which has been proposed by Goldsmith (1976, 1990) and Halle and Vergnaud (1980) among others, assumes automatic spreading of tone melodies on to tone-bearing units (normally vowels) that are unasso- ciated with any tone. Thus, in [8.6] above, Goldsmith's version of the WFC requires (i) all tones to be associated with at least one vowel and (ii) all vowels to be associated with at least one tone. This means that in configur- ations like those in [8.7a], where a tone associated with a vowel is adjacent to another tone that is unassociated with a vowel, a universal convention automatically spreads the tone to the toneless vowel as in [8. 7b]:
[8.7] a. H L L H b. H L L H
v v
v v v v
By contrast, some linguists, like Pulleyblank (1986), assume that the spreading of melodies to toneless vowels is non-automatic. On this view, spreading only occurs if a specific rule in a given language sanctions it. In
Phonological Prelude: Autosegmental Phonology 159 many languages, a tone that is underlyingly unassociated with a vowel does not automatically spread to an adjacent vowel. Instead, it is deleted - unless a specific rule links it to a vowel. For reasons that will become clear later in this chapter, this kind of non-automatic spreading is what is required for morphological purposes.
We will be adapting the Mapping Principles outlined below for linking tones to tone-bearing units in our morphological analysis. These principles are based on Pulleyblank (1986) and Archangeli (1983).
[8.8] Universal Linking Convention
(a) Link a sequence of autosegments (e.g. tones) with a series of elements on the skeletal tier that are capable of bearing them (e.g. link tones with vowels) (see [8.3]);
(b) Perform the linking going from the beginning to the end of the word. Unless specific instructions are given in the gram- mar of the language to do otherwise, link autosegments (e.g.
tones) with units that are capable of bearing those autoseg- ments (e.g. vowels) in a one-to-one fashion.
(c) Association lines do not cross in the linking process.
The last point needs some comment because its significance is not obvious.
There is a good reason for the ban on crossed association lines. To see why, let us begin by observing that each autosegmental tier contains a linearly ordered sequence of 'segments'. In this context the term segment includes not only consonants and vowels, but also any other elements on autoseg- mental tiers such as C and V slots (or x's), stress and tone. The segments on separate tiers are linked by an ordered sequence of association lines.
Association lines indicate that a given segment, say the vowel Iii, is articulated in a temporal sequence linked to another specified segment, say a high tone (Goldsmith 1976: 28-30). The prohibition of crossed associ- ation lines reflects the fact that when an articulatory gesture is interrupted by another distinct gesture, if we want to resume the original gesture, we have to start afresh. This means that a representation like [8.9a) is ruled out because the same H tone is simultaneously associated with the first and third syllables since the association lines linking H and L tones to vowels end up crossing. In circumstances like these, we must posit two separate H tones linked with the first and third vowels, as in [8.9b):
[8.9] a. b. H L H
ba la ba ba la ba
160 Prosodic Morphology
The WFC does not police the system so effectively that it prevents any unlawful representations from ever arising. Violations of the WFC can arise. But if they do, the WFC triggers appropriate repair rules to rectify the situation. It is assumed that the optimal repair strategy in each case will be one that involves the smallest number of rules. This follows the gener- ally accepted principle of economy. (There is no virtue in using five rules to deal with a problem that can be solved using one rule.) The repair operation always makes the minimal number of changes required to satisy the WFC-phonotactic in [8.6]. In [8.10a] the repair operation associates unassociated vowels (V) with tone (T) while in [8.10b] it associates unasso- ciated tones with vowels:
[8.10] a. H b. H L
[''',,,',, l// I
v v v
As we have hinted, the Mapping Principles outlined above are applicable not only to tone and tone-bearing units but also to any other elements that are on separate tiers. They regulate the ways in which association lines may link any autosegments to elements that bear them. For a detailed dis- cussion of autosegmental phonology see Goldsmith (1976, 1990), Clements and Ford (1979), Clements and Goldsmith (1984), Pulleyblank (1986), and Katamba (1989).
8.2.2 The Skeletal Tier
The original impetus for autosegmental phonology came from the analysis of tone. But before long phonologists recognised that the essentials of this approach could be extended to other phonological phenomena. For instance, just as tones may be linked in different ways to tone-bearing units, segments may be linked in a variety of ways to the skeletal tier. This has important repercussions for morphological theory, as we will see below.
Normally, in the lexicon every vowel segment starts off being associated with a V-slot and every consonant with a C-slot on the skeletal tier (also called the CV-tier) (see [8.3]). We illustrate this with the Luganda word mukazi 'woman':
[8.11] Skeletal tier: c v c v c v
Segmental tier: m u k a z
Phonological Prelude: Autosegmental Phonology 161 Luganda has both long vowels and geminate ('double') consonants. The geminate consonants are longer and have a more forceful articulation than plain, short consonants. Clements (1985) and Katamba (1985) have shown that in Luganda, the ways in which vowel and consonant segments are linked by association lines to the skeletal tier throws light on their phonological properties.
Geminate consonants arise if, where there are more C (or x) slots on the skeletal tier than there are consonantal segments on the segmental tier, a single consonantal segment is simultaneously associated with two C-slots on the skeletal tier. You can see this in [8.12] where the geminate tt of tta 'kill' contrasts with the single t of ta 'let go'.
[8.12] c c v
v a tta c t v a ta
Similarly, although the norm is one-to-one linking of vowels and V -slots on the skeletal tier, it is possible to find more than one vowel segment linked to a V-slot. Vowels simultaneously linked to several V-slots are long.
Length is phonemic. The verbs siiga 'smear, paint' and siga 'sow, plant' are distinguished by the length of their first vowel, as seen in [8.13]:
[8.13] a. c v v c v b. c v c v
v
s g a siiga s g a siga
Observe that in [8.13a] compensatory lengthening takes place when a single vowel is doubly linked with two V slots in the underlying represen- tation. Thus /if surfaces as long [i:] in siiga.
Using the notation of autosegmental phonology, account for the 'compen- satory lengthening' in the following:
[8.14] a. /ba- e- labal
'they-themselves-see' b. /tu- e -labal
'we-ourselves-see'
[be:laba]
'they see themselves' [twe:laba]
'we see ourselves'
162 Prosodic Morphology c. fbi- e- labal [bye:laba]
'they-themselves-see' 'they (e.g. animals) see themselves' Observe that in [8.14a], if the first of two adjacent vowels is nonhigh, it is deleted. Then the vacant V slot left behind at the skeletal tier is inherited by the second vowel which is 'lengthened in compensation', as seen in [8.15].
[8.15]
cvvcvcv cvvcvcv
t cvvcvcv v
baelaba b@elaba b e I a b a [be:laba]
they see themselves they-themselves-see
However, if the first vowel is high, i.e. Iii or /ul (as in [8.14b] and [8.14c]
respectively), it is not deleted. Instead, it is associated with the C slot to its left and surfaces as a nonsyllabic glide /y/ or /w/. Again, the first V-slot which is vacated is taken over by the second vowel, which is 'lengthened in compensation':
[8.16] a.
cvvcvcv
t u e 1 a b a we-ourselves-see b. cvvcvcv
cvvcvcv cvvcvcv
t ~I\ V
t u e I a b a t w e 1 a b a [twe:laba]
we see ourselves
cvvcvcv cvvcvcv
t ~ /\ v
b i e I a b a b e 1 a b a b y e 1 a b a [bye:laba]
they-themselves-see they see themselves
The aim of this phonological interlude has been to give you a grounding in the essentials of autosegmental phonology. The discoveries about the
Prosodic Morphology 163 nature and organisation of phonological representations made in this model form the backbone of the prosodic morphology model to which the rest of this chapter is devoted.