English words on the Procrustean bed: Polysyllabic shortening reconsidered

Abstract

The polysyllabic shortening hypothesis holds that the duration of a primary stressed syllable is inversely proportional to the number of additional syllables within the word. We examine the evidence for this process in British English speech by measuring the duration of primary stressed syllables in monosyllabic, disyllabic and trisyllabic words, both right-headed series – e.g. mend, commend, recommend – and left-headed series – e.g. mace, mason, masonry. In contrast with some of the original studies of polysyllabic shortening (e.g. Lehiste, 1972), we record target words both when carrying nuclear pitch accent and when unaccented. As in previous studies, we find strong evidence of polysyllabic shortening in accented words, an effect of comparable magnitude in right-headed and left-headed words. In unaccented words, polysyllabic shortening is minimal or absent, but there is evidence, supporting previous studies, of domain-edge effects localised to specific sub-syllabic constituents. Unlike these effects, which occur on both pitch-accented and unaccented words, polysyllabic shortening of the primary stressed syllable in these data is confined to pitch-accented words.

Introduction

In Greek mythology, Procrustes was a bandit who dragged his kidnap victims back to his lair and tied them to the bed. He had a strict “one size fits all” policy: short captives were stretched to fit the length of the bed; tall ones had their legs cut down to size. The Procrustean bed approach has been much in evidence in theories of speech timing, particularly for Germanic languages like English. The isochrony hypothesis proposed that stress-delimited feet in English are of equal duration (e.g. Abercrombie, 1967), which requires stretching or compression where feet are composed of different numbers of syllables or segments. Similarly, the polysyllabic shortening hypothesis proposes that sub-constituents of words are stretched or compressed towards preservation of relative uniformity of word length (e.g. Lehiste, 1972).

There is a broad class of speech timing processes that are well-supported, which may be glossed as “lengthening at important points in the signal”. Lengthening at the edges of domains is well-attested, in particular, word-initial lengthening (e.g. Cooper (1991), Fougeron & Keating (1997), Oller (1973)) and phrase-final lengthening (e.g. Klatt (1975), Oller (1973), Wightman, Shattuck-Hufnagel, Ostendorf, & Price (1992)). Some studies have reported a word-final lengthening effect (e.g. Beckman & Edwards (1990), Klatt (1973), Oller (1973)), although whether this occurs in the absence of higher-level boundaries or phrasal stress remains uncertain (e.g. Harris & Umeda (1974), Turk & Shattuck-Hufnagel (2000)). Accentual lengthening – the greater duration of segments within phrasally stressed, pitch-accented words – may be counted another such process (e.g. Sluijter (1995), Turk & Sawusch (1997), Turk & White (1999)), affecting the stressed syllable and other parts of phrasally stressed words.

Procrustean processes such as stress-based isochrony and polysyllabic shortening posit durational adjustments of a different nature, essentially suggesting an inverse relationship between the length of some constituent, such as the word or the foot, and the duration of some sub-constituent, e.g. segment or syllable. A parsimonious model of speech timing should posit the minimum number of different types of durational effect necessary to explain observed durational data, so given that stress-based isochrony is not well-supported empirically (e.g. Dauer, 1983), the status of Procrustean processes in general may also be questionable. In particular, much of what has been described as polysyllabic shortening may be explainable in terms of other, better-supported effects, e.g. accentual lengthening, initial lengthening and final lengthening.

A primary purpose of this paper therefore is to reconsider the so-called polysyllabic shortening in contexts where other durational effects are likely to be absent or controlled. For example, in comparisons of disyllabic vs trisyllabic words, initial lengthening on [m] is unlikely in, e.g. commend vs. recommend, and accentual lengthening is absent when the word bears no pitch accent. A shorter [mεnd] duration in unaccented recommend as compared to commend, all else being equal, would be unambiguous evidence for polysyllabic shortening, lending support to the view that it is a fundamental process in English speech timing.

In addition, we consider how the lengthening of constituents in pitch-accented words (accentual lengthening) may mediate the relationship between word length and the duration of the primary stressed syllable, given that most previous findings of polysyllabic shortening have examined accented words only. We also assess the possible role of word-final lengthening in observed polysyllabic shortening. To do this, we compare the effect of word length on the primary stressed syllable in sequences with final stress (“right-headed” words – e.g. ˈmend, coˈmmend, recoˈmmend) and sequences with initial stress (“left-headed” words – e.g. ˈmace, ˈmason, ˈmasonry). In right-headed words, final lengthening of the primary stressed syllable is expected regardless of word length, and so should be a constant factor, whereas in left-headed words, final lengthening on the primary stressed (initial) syllable is expected to attenuate the further this syllable is from the end of the word (e.g. Wightman et al., 1992).

It has long been asserted that there is an inverse relationship between word size and sub-constituent duration in English speech. Jones (1942–3, p. 10) stated that the duration of English long vowels in primary stressed syllables is strongly affected by the number of following unstressed syllables within the word. Lehiste (1972) and Port (1981) found evidence for this, showing that the stressed vowel in words such as speed, speedy, speedily was progressively shorter as word length increased. (Except where otherwise stated, all findings discussed here concern American English speech.) These studies considered only left-headed words (i.e. those beginning with a stressed syllable) and utilised a fixed frame sentence, in which the only variation between successive sentences was the test word itself. For example, Port (1981) presented nonsense words such as dib, dibber and dibberly in the fixed sentence I say [target word] again every Monday. This context would be expected to elicit a nuclear pitch accent on the target word, as the only new information in successive sentences. Furthermore, Lehiste and Port – in common with polysyllabic shortening researchers such as Barnwell (1971), Klatt (1973) and Nakatani, O’Connor, and Aston (1981) – did not control phrase length or the alignment of the stressed syllable with constituent boundaries.

Given these aspects of experimental design, the interpretation of such results remains ambiguous. Firstly, where evidence regarding polysyllabic shortening is only obtained from accented words, the possibility remains that the effect depends on the presence of phrasal pitch accent. Secondly, because the word length manipulation in these experiments was accomplished by adding a syllable to the word, and thus also to the phrase and utterance, it is not clear whether results from these studies really demonstrate a Procrustean effect at the word-level. The domain of such an effect could equally be a higher-level constituent dominating the word, or indeed a lower-level constituent, such as a within-word constituent beginning with a stressed syllable. Thirdly, unless the alignment of the measured syllable with word and phrase boundaries is controlled, observed Procrustean effects may be confounded with domain-edge lengthening: for example, the difference between speed and speedy in a phrase-final context could be due to the relative attenuation of final lengthening in the stressed syllable in the context of the longer word.

Turk and Shattuck-Hufnagel (2000; see also Beckman & Edwards, 1990) addressed the potential confounds implicit in many earlier studies, looking for evidence of polysyllabic shortening in monosyllables and disyllables beginning and ending with matched stressed syllables (e.g. left-headed tune vs tuna; right-headed choir vs acquire). They attempted to control extraneous influences, such as phrase length and alignment with phrase boundaries, by placing these words in near-homophonous phrase pairs (e.g. tune acquire vs tuna choir), within carrier sentences. They found evidence for some polysyllabic shortening in the stressed syllables of both left-headed and right-headed words, although this effect appeared to be greater in left-headed (e.g. tuna) than right-headed words (e.g. acquire) and greater in all words when pitch-accented. They concluded that a combination of word-initial lengthening (Cooper (1991), Fougeron & Keating (1997), Oller (1973)), accentual lengthening (Sluijter (1995), Turk & Sawusch (1997), Turk & White (1999)), syllable ratio equalisation (Abercrombie (1965), Albrow (1968)) and polysyllabic shortening best accounted for their results in American English (see Turk & Shattuck-Hufnagel, 2000, for a full description and review of the evidence for these processes).

By examining both left-headed and right-headed words, and by controlling the length of the carrier phrase, Turk and Shattuck-Hufnagel's study specifically targeted the influence of word length on sub-constituent duration. Their results did, however, raise some questions deserving of further study, in particular, concerning the link between accentual lengthening and polysyllabic shortening.

When the primary stressed syllable of a word carries a pitch accent, it undergoes substantial lengthening, manifest on all sub-constituents of the syllable (Sluijter (1995), Turk & Sawusch (1997), Turk & White (1999)). Turk and White, in a study of contrastive pitch accent in Scottish English, found this lengthening was greater in a monosyllable than in a disyllable: thus, for example, [ni] shows 23% accentual lengthening in the monosyllable knee but only 16% in the disyllable kneecap. Given this attenuation of accentual lengthening in disyllables as compared to monosyllables, stressed syllables will be longer in pitch-accented monosyllables than in pitch-accented disyllables, even in the absence of a more general inverse relationship between word length and stressed syllable duration. Turk and Shattuck-Hufnagel (2000) also found that polysyllabic shortening appeared greater when words were pitch-accented, reinforcing the possibility of a link between accentual lengthening and observed polysyllabic shortening.

Durational mechanisms are traditionally described in terms of lengthenings and shortenings, terms which imply the adjustment of default durations. The adjustment-of-a-default view is not readily distinguished on the basis of data currently available from another type of view, in which surface durational patterns result from the simultaneous influence of a variety of abstract, structural factors (e.g. position, phrasal stress, number of syllables in a word, etc.). The main goal of our paper is to distinguish Procrustean effects, where surface durations depend on the number of subunits within a larger unit, from other types of effects, such as edge effects and durational patterns due to the presence of phrasal pitch accent. We frequently refer to these effects in traditional terms (e.g. polysyllabic shortening, final lengthening, etc.), but our use of these terms should not be taken to imply that we endorse an adjustment-of-a-default view.

A comparison between monosyllables and disyllables, such as in Turk and Shattuck-Hufnagel (2000), makes disentangling domain-edge and Procrustean effects problematic: if a stressed syllable is spoken in the context of a monosyllabic word, both initial and final lengthening may operate to increase its duration relative to a disyllable. For example, [tun] in tune is word-final and its rhyme may be longer than that of [tun] in tuna. Similarly, [kwair] in choir is word-initial, and therefore its onset is expected to be longer than the onset of [kwair] in acquire. In the current experiment, we aim to distinguish domain-edge durational effects from the influence of polysyllabic shortening by looking at sequences of monosyllables, disyllables and trisyllables, both when pitch-accented and when unaccented. Comparisons of disyllables and trisyllables in particular (e.g. commend vs recommend, mason vs masonry) will control for several potential word-edge confounds. Evidence of comparable polysyllabic shortening in right-headed and left-headed words, both when accented and unaccented, may be taken as a clear demonstration of a Procrustean process; specifically, an inverse relationship between word length and sub-constituent duration. If this relationship is found to be strongly influenced by the presence of pitch accent, or differentially distributed in right-headed and left-headed words, it weakens the status of polysyllabic shortening as a speech timing mechanism with general applicability.