CHAPTER FIFTEEN

Nonsegmental Phonology

Sanjord A. Schane

Within standard generative phonology a phonological representation is depicted as a linear arrangement of segments, although each of the segments itself is composed of simultaneously occurring features. Phonological rules operate on such strings, deleting, inserting, or permuting segments, or changing one or more of their feature values. Because of its adherence to segment-sized units, the theory with its associated notational conventions has not always accommodated easily higher-level phonological constructs, such as the syllable, or ‘suprasegmental’ phenomena which may extend over a sequence of segments, such as pitch and stress contours. Recent proposals by Goldsmith, Selkirk, and myself have demonstrated the value of recognizing such nonsegmental entities and the necessity for formally incorporating them into phonological theory.

I

Some processes affecting syllable structure show that a segmental phonology does not always lend itself to the most insightful analysis and may even engender notational complexity. Consider, for example, the standard generative formulation of the processes which lengthen vowels in open syllables and shorten them in closed syllables.

Hooper (1972) and Vennemann (1972b) have convincingly argued for the syllable boundary in phonological rules. Rules mentioning open and closed syllables becomes notationally simpler, as reference to unrelated disjoint environments is no longer required.

  (2)   a. V → [+ long] / ______$    b. V → [-long] / ______C$

Although the recognition of the syllable boundary is a step in the right diréction, its status is no different from that of the other boundaries of generative phonology. It appears interspersed among the segments of the linear string.

There is reason to believe that the syllable should be regarded as a higher-level unit in its own right, separate from the segments of which it is composed, for it plays a significant role in determining certain phonological phenomena. A case in point is ‘closed-syllable’ adjustment in French, a process affecting the vowels [ǝ] and [e]. These vowels occur in open syllables; in closed syllables they both become [Ɛ].

To accommodate this alternation the standard generative rule must list three environments.

The first two contexts are typically cases where the consonant following ǝ or e closes the syllable. By employing the syllable boundary, we can easily restate these two environments as a single one. However, this reformulation cannot encompass the third context, where the following syllable contains schwa, as this environment by no means represents a typical closed syllable; hence, it necessitates a separate statement.

It would appear that the second part of the environment could perhaps be eliminated by allowing schwa deletion to apply first. Then the ə or e would truly be in a closed syllable. This ploy is not really effective, for in those styles where schwas are preserved, it is still the case that ə and e become [Ɛ]—e.g., [žƐlǝra], [žƐrǝra].

It is well known that French schwa has peculiar properties. Selkirk (1978) accounts for the strange behavior of schwa by referring to higher-level constructs. She proposes that segments be grouped into syllables. A word will have as many syllables as it has vowels. (Every schwa belongs to its own syllable.) Syllables in turn are then grouped into yet higher-order units, called feet. Normally, a foot is composed of a single syllable. However, within a word a non-initial schwa does not form a foot by itself; rather, it is part of the same foot as the preceding syllable. But a schwa in the first syllable can constitute a foot. A French foot then may contain at most two syllables, in which case schwa appears as the second member and is considered to be subordinate. The words of (3a) are analyzed into feet, with brackets used to denote the domain of each foot.

(6) geler [žǝ] [le]

(il) gèle [žǝlǝ]

(il) gèlera [žǝlǝ] [ra]

gel [žǝl]

We can now state the appropriate generalization for closed-syllable adjustment: 9 and e become e when followed by something else within the foot, whether a single consonant, a consonant cluster, or an entire subordinate syllable containing schwa.

(7)ǝ e → ε / [C^o________C X]

Selkirk also shows how the foot explains other processes associated with schwa: schwa deletion and stress assignment. The latter has traditionally been described as follows: The last vowel of the word is stressed, unless it is schwa, in which case stress will be on the penultimate. The following words, taken from (3), will have stress on the italicized vowel: geler, (il) gèle, (il) gèlera, gel. The standard generative rule is:

(8)V → (+ stress]/_____C^o(ǝ) #

A reformulation utilizing the notion of foot requires no mention of schwa. The new rule becomes:Stress the last foot of the word. (Whenever the final foot contains a subordinate schwa, stress will of course be assigned to the vowel of the dominant first member.)

(9)[ ] → [+stress] / ________ #

II

Suprasegmental phenomena also lend themselves to a treatment divorced from the consonants and vowels of the segmental string. Goldsmith (1976b and in this volume), in his ‘autosegmental’ phonology, demonstrates that tonal contours are not mere features of vowels but constitute a separate tier of representation to which phonological rules can appeal. Within his system every vowel bears at least one tone and every tone is assigned to at least one vowel. ‘Association lines’ match tones and segments.

In (10a) the first syllable carries a high tone and the second a low one, in (10b) a high contour is spread over the first two syllables, and in (10c) the syllable bears a ‘falling’ (high-low) tone.

Rules may apply separately to each tier without disrupting any of the entities on the other level, although the association lines may require reassigning. For example, assume that there is a process whereby a syllable without an inherent tone is affixed to the beginning of a word, such as (11) (= 10a).

Since every vowel must bear at least one tone, the two tones will have to realign themselves to cover three syllables. Because association lines may not cross, the resulting contour would be that of (10b). By the same token, a deletion rule applying to segments need not touch any of the tonal contour. Imagine a rule deleting the final vowel of (10a).

As every tone must be assigned to some vowel, once again there will have to be realignment of association lines. The result will be a contour such as (10c).

Rules might also insert or delete tones without disrupting the sequence of vowels and consonants. The necessary realignments would be similar to those already illustrated. Or consider tonal assimilations. Let us assume that the second syllable of (10a) assimilates the tone of the preceding syllable. The resulting contour will be a high tone followed by a falling one.

Goldsmith cites many interesting examples of these different possibilities. It is easy to see the advantages of his proposal and its facility in handling the various processes, most of which are considerably more complex than what has been discussed here. He also has some insightful suggestions for extending the framework to other aspects of phonology, such as vowel harmony.

III

Stress is another phenomenon amenable to a nonsegmental approach. In Chomsky and Halle (1968) stress is a feature of vowels and the stress rules assign integer values to this feature on the basis of properties of the strings of segments. In Schane (1978) I propose a radically different treatment, one where stress contours are represented as alternating patterns of strong (S) and weak (W) syllables.catͣ

These data reveal certain surface constraints on the distribution of Ss and Ws. We shall be concerned with two of these constraints.

(15) a. *S S (Two Ss are never contiguous.)

b. *# W W (A word may not begin with two Ws.)

Constraint (15a) entails that any two Ss must be separated by at least one W. Hence, a word containing two Ss must have a minimum of three syllables—e.g., Constraint (15b) means that a word may begin with at most one W—e.g., Thus, whenever two Ws occur they are always preceded by an S.

There are many examples of related forms with shifting stress configurations—e.g., solid: solidity, adore:adoration, etc. We shall view the more complex form as composed morphologically of the simpler (base) form with the addition of a derivational suffix. With the juxtaposition of the suffix, it may happen that one of the constraints of (15) becomes violated. The impermissible sequence will then have to be corrected. The adjustments change the first element of the sequence—that is, if in the course of a derivation two contiguous Ss should arise, the first one will be changed to W, and conversely, if there should be an unpermitted sequence of two Ws, the first one will become S.

In the following examples, observe what happens to a base form (line 1) when a suffix with its own accentual properties is added (line 2). The addition of the suffix creates sequences in violation of (15), which are then adjusted in conformity with (16) until no violations result.

First let us look at some examples where two Ss become juxtaposed, causing the first one to change to W. Consider the derivation of circulatory.

The stress configuration of circulate is SWS, to which is added the suffix -ory with its pattern SW. The final S of circulate, because it is now immediately followed by the S of -ory, will be changed to W, and as a consequence its vowel will be reduced to schwa.

A slightly different example is solidity.

To the base form solid (SW) is added the suffix -ity, which itself has the pattern WW. However, with this particular suffix S is assigned to the immediately preceding syllable. The initial syllable of solid must accordingly become W and its vowel will then reduce.

The other rhythmic constraint disallows words beginning with two Ws. Should such a situation arise, the first one will be changed to S. Note the derivation of adoration.

Here the base form adore (WS) is followed by the SW suffix -ation, and so there are two consecutive Ss. But the change of the first one to W now produces a word beginning with two Ws. As this sequence is unacceptable, the first W becomes S.

We have examined three different proposals for describing phonological data. Each of these approaches makes use of entities different from the traditional segments of generative phonology. Selkirk’s notion of foot entails a higher-level construct, comprising one or two syllables, where the syllables in turn are higher-order units composed of segments. Goldsmith envisages separate ‘suprasegmental’ (and by extension perhaps also ‘subsegmental’) sequences, constituting a distinct phonological tier, yet in alignment with the segmental layer. I have advocated rhythmic patterns of alternating strong and weak syllables. All of these proposals have a common trait: Phonological rules are able to refer directly to these nonsegmental entities, thereby bypassing reference to individual consonants and vowels. In each case there emerges a more insightful account of particular phenomena, which, in the standard framework, are accommodated clumsily. This line of investigation represents one of the truly exciting avenues of research currently unfolding within phonology.