Stress, Linguistic

In pronouncing the word àutobìográphic, English speakers make the odd-numbered vowels more prominent than the even-numbered, with greatest prominence going on the last odd-numbered vowel. In the traditional terminology of phonetics, the odd-numbered vowels are said to be stressed and the even-numbered, unstressed. Stress is commonly -- though not universally -- implemented phonetically by an increase in the pitch (fundamental voice frequency) of the vowel. (For details of English stress, see Pierrehumbert 1980.)

In many languages, word stress is predictable. The principles and rules that make this possible are somewhat separate from the rest of the PHONOLOGY. This quasi-independence of the stress rules is reflected in the fact that when adding to its vocabulary a word from another language, the borrowing language often preserves the phonemes (sounds) of the original word but modifies its stress contour. For example, the Russian borrowing babúshka is stressed in English on the second syllable, whereas in Russian this word has initial stress (i.e., bábushka). Except for the stress, the sounds of the Russian word are (quite) faithfully reproduced in English. As there has been no concerted effort by the schools or media to instill a particular stress pattern, the most plausible explanation for the babúshka stress is that English speakers assign stress to this new word by analogy with such English words as Aláska, fiásco. Implicit in this explanation is the assumption that English speakers have knowledge of the English stress rules and make active use of this knowledge. Although little is known at present about how this knowledge is used by speakers in the production of utterances, a great deal has been learned in the last quarter century about the nature of this knowledge.

The fundamental insight into the nature of stress is due to Liberman 1975. It was he who suggested that stress reflects the grouping (chunking) of the vowels of the word -- more exactly, of its stressable sounds -- into subsequences called feet. Research since 1975 has shown that the construction of feet -- and hence stress assignment -- is governed by a small number of rules belonging to a few schemata that differ with respect to a handful of binary parameters. Because this perspective on stress is not widely known and may well fly in the face of common-sense views, the rest of this article presents a detailed illustration of how feet are employed to compute the stress of words. The discussion here adopts the formalism of Idsardi (1992), because it reflects most clearly the role of feet in the computation of stress. (For discussion of alternative approaches, see Halle and Vergnaud 1987; Idsardi 1992; Kenstowicz 1993; and Hayes 1995.)

We begin with the very simple stress system of colloquial Czech. In Czech, odd-numbered vowels are stressed, and main stress falls on the vowel of the word-initial syllable, as for example in prábabìkamì in 'greatgrandmothers' (instrumental case) (Jakobson 1962). In most languages, vowels are stressable, but consonants are not. To reflect formally this difference between sounds that are and are not stressable, the Idsardi theory posits that metrical structure is computed on a separate plane onto which are projected all and only the stressable phonemes (usually the vowels) of the word. This is illustrated in (1) with the Czech word cited above.

In (1) each vowel projects an asterisk and the sequence of asterisks so generated is labeled line 0. Additional lines of asterisks are generated by devices explained below. The set of asterisk lines associated with a given word is its metrical grid.

The grouping of stressable elements into feet is notated here by means of parentheses: a left parenthesis groups into a foot the stressable elements on its right, whereas a right parenthesis groups the elements on its left (cf. below). The parentheses themselves are inserted by three types of rules.

The first of these is iterative foot construction (IFC) rules. An IFC rule inserts left or right parentheses into a sequence of asterisks beginning at either its left or its right edge and proceeding toward the opposite edge of the string, subject to the constraint that except for the initial parenthesis a substring of two or three asterisks must separate an inserted parenthesis from the nearest parenthesis on its left, where insertion is from left to right, and on its right, where insertion is from right to left. The Czech IFC rule is given in (2) and its effects are illustrated in (3).

(2) On line 0 insert left parentheses starting from the left edge at an interval of two asterisks

What differentiates various IFC rules is the replacement of one or more of the three variables by its alternative (i.e., left by right or 2 by 3). The theory of metrical structure admits therefore exactly eight different IFC rules, of which a given language characteristically uses one. (For illustrations, see References.)

The second type are head-marking rules. In each foot, one of the elements -- called the head -- is specially marked (stressed). The head of a foot is either its left-most or its right-most element, and the head element of each foot is projected onto the next higher line in the grid, thereby signaling its marked (stressed) status. In Czech, line 0 feet are left-headed. This is illustrated in (4).

The third category are edge-marking rules. As noted above, the main stress in Czech words falls on the vowel of the initial syllable. In order to mark this vowel, the theory makes use of a second type of parenthesis insertion rule, Edge Marking, which in Czech has the form in (5).

(5) Insert a left parenthesis to the left of the left-most asterisk on line 1.

Like the IFC rule (2), the Edge Marking rule (5) has three binary variables. There exist therefore eight Edge Marking rules for languages to choose from.

The Edge Marking rule (5) creates a foot that includes all asterisks on line 1. Like all feet, those constructed by (5) have a head, which in Czech must evidently be the left-most asterisk. This is illustrated in (6).

It is worth noting that in (6) the height of the asterisk columns reflects the relative degree of prominence of the different vowels in the word.

The stress rules of Czech are summarized in (7).

With one major addition detailed below, the machinery introduced above accounts correctly for the stress patterns of words of most (all?) languages. Two further examples are reviewed below.

Pintupi, an Australian language, differs from Czech in that the Pintupi IFC rule inserts Right rather than Left parentheses. The Pintupi word, like that of Czech, therefore has main stress on the initial vowel and secondary stresses on other odd-numbered vowels; Pintupi differs from Czech in that the word-final vowel is never stressed. This is illustrated in (8). (Capital letters represent retroflex consonants. For some additional discussion of Pintupi, see Kenstowicz 1994.)

To deal with our final example, that of Selkup, a language of Siberia, a third type of parenthesis insertion rule is needed. This rule inserts a left/right parenthesis to the left/right of asterisks projecting vowels of special syllables -- for example, syllables with long vowels. In Selkup, stress falls on the last long vowel of the word, but in words without a long vowel, stress is on the initial vowel. This migration of the stress toward opposite ends of the word is accounted for by the rules in (9).

As illustrated in (10a), these rules assign stress to the last long vowel, and, as shown in (10b), in words without a long vowel the rules assign stress to the word-initial syllable. In view of (9iv) and (vi), there is but one stress per word in Selkup.

Although Selkup stress is vastly different from that of both Czech and Pintupi, it is the result of the same rules (rule schemata) with different settings of the parameters. To explain this similarity, we hypothesize that the schemata are part of the innate cognitive capacities that normal children bring to the task of learning the language of their milieu. Because the rule schemata are part of the innate cognitive equipment of learners, the task of a child learning a language reduces to figuring out the setting of a dozen or so binary parameters. This hypothesis accounts for the speed and accuracy with which children normally accomplish this task. Finally, by positing that knowledge of rule schemata is part of the genetic endowment of humans, but not of other species, we explain also why language is a uniquely human trait.

See also

Additional links

-- Morris Halle


Halle, M., and J.-R. Vergnaud. (1987). An Essay on Stress. Cambridge, MA: MIT Press.

Hayes, B. (1995). Metrical Stress Theory. Chicago: University of Chicago Press.

Idsardi, W. J. (1992). The Computation of Stress. Ph.D. diss., MIT. Distributed by MIT Working Papers in Linguistics, Cambridge, MA.

Jakobson, R. (1962). Contributions to the study of Czech accent. In Selected Writings 1. S-Gravenhage: Mouton and Co., pp. 614-625.

Kenstowicz, M. (1994). Stress. In Phonology in Generative Grammar. Oxford: Blackwell, pp. 548-621.

Liberman, M. (1975). The Intonational System of English. Ph.D. diss., MIT.

Pierrehumbert, J. B. (1980). The Phonology and Phonetics of English Intonation. Ph.D. diss., MIT.