Effects of Focus Distribution, Pitch Accent and Lexical Stress on the Temporal Organization of Syllables in Dutch

In recent developments in phonological theory two independent representations for prosodic prominence are needed in languages such äs Dutch and English. A nonculminative autosegmental structure with high and low tones accounts for pitch accents in focused constituents, whereas a culminative metrical structure which is phonetically coded in relative duration accounts for the lexical stress position in a word. The most farreaching consequence following from this proposal is that relative temporal structure of a word does not change if a pitch accent is shifted to an unstressed syllable. Our results show that, if a pitch accent is shifted (through focus manipulation) from the stressed onto the unstressed syllable, rhyme durations are more or less inverted. Therefore, the assumption of completely independent tonal and metrical structure is largely untenable. However, our results also show a small residual effect of the original stress pattern after the accent shift, which can be accounted for by a metrical grid representation. 1, Theoretical Background It is a general characteristic of many languages that certain syllables are feit to be more prominent than others, whether in isolated words or in continuous speech. Such prosodically prominent syllables stand out from their environment due to (among other things) pitch changes, increased vocal effort (intensity), and longer duration. Publications on prosodics in generative phonology are generally concerned with either metrical sturcture [Nespor and Vogel, 1986; Baart, 1987; Kager, 1989], or tonal structure The authors would like to thank A. H. Ncijt Received: and J. W. de Vries Tor ideas and discussion. March 9, 1993 Accepted· May 7, 1994 A. M. C. Sluijter © 1995 Holland Institute öl' Generative Linguistics, S. Karger AG, Basel Phonetics Laboratory 0031-8388/957 Leiden University, PO Box 9515 0522-0071 $ 8.00/0 [Pierrehumbert, 1987; Gussenhoven, 1988]. Implicit in such work is the hypothesis that there exists a notion of 'prosodic prominence', and that one representation, be it metrical or autosegmental, will be able to account for all types of prosodic prominence. More recently, linguists have come to realize that at least two types of abstract prominence structure have to be distinguished: tonal prominence (the abstract representation of pitch accents) and metrical prominence (the abstract representation of temporal organization) [Neijt, 1990]. It is theoretically unclear at this moment whether tonal and metrical prominence structures are independent or interact, and in the latter case, how. The present article explores some of the phonetic consequences of proposals that have been suggested in the phonological literature on possible interactions between tonal and metrical prominence. Less phonologically oriented phoneticians might be inclined to dismiss the need for the present study on the grounds that its results are obvious from everyday experience with speech signals. However, given the apparent ease with which phonologists have come up with a variety of prominence representations with various dependencies among them, we feel an urge for Substantive experimentation to provide a factual basis for such theoretical work. Tonal structure is the succession of high and low tones in a sentence [Gussenhoven, 1988]. If a particular tone is accent-lending, the syllable (or larger unit) carrying the accent is put in focus, i.e., made communicatively important. Phonetically, an accented syllable is characterized by a fast pitch movement (rise, fall or both) ['t Hart et al., 1990]. Moreover, all the Segments in the accented Version of a word are pronounced longer than in the unaccented Version, in stressed and unstressed syllables alike [Eefting, 1991]. Metrical structure is predominantly coded in the durational properties of syllable strings [Slootweg, 1988]. The lexically stressed syllable is the metrically most prominent syllable. This syllable has the longest duration (after normalization for inherent segment duration and linear position within the word). Prominence relations among syllables are expressed in one of two ways (or even both): äs strong and weak nodes in a tree structure [Kiparsky, 1979] or äs a metrical grid [Prince, 1983; Selkirk, 1984]. Most recently, Hayes [1993] argued that the optimal representation of metrical structure is the labeled grid, a compromise between grids and trees. Crucially, however, whether one uses trees or grids, all theories suggest, explicitly or implicitly, that metrical structure accounts for the relative duration of syllables (again disregarding influences of (co-)intrinsic segment duration and preboundary lengthening). 'The metrical grid alignment of a sentence is a representation in terms of which such things äs the isochrony of stressed syllables and more generally the relative durations of syllables might be expressed' [Selkirk, 1984, p. 12]. Combining the theoretical views expressed above on the representation of tonal prominence and of metrical prominence, the conclusion seems warranted, and has in fact been advanced by Neijt [1990], that metrical structure determines relative duration of syllables within words and that the only temporal contribution of a pitch accent is linear expansion of the entire word. Data described in Nooteboom [1972] support this conclusion. Nooteboom varied stress positions in unaccented Dutch three-syllable nonsense words like /paipaipaip/. A stressed syllable was always longer than its unstressed counterpart in the same position. Moreover, when the same words were accented all the constituent syllables were linearly expanded in time. Reanalysis of the data reveals that the relative 72 SJuijter/van Heuven Focus, Pitch Accent, and Lexical Stress syllable duration (percentage of word duration) remains almost identical in the conditions with and without an accent [Martens, 1992J. Neijt [1990] goes one Step further by concluding that the contributions of metrical structure and accent to the temporal organization are completely independent. However, this conclusion might be premature. The literature data that were used in support of Neijt's view were exclusively based on experiments in which pitch accents occur on lexically stressed syllables. The conclusion that the relative duration of syllables is independent of accentuation can be based only on speech material with accents on unstressed syllables äs well. The validity of this conclusion can only be tested if stress and accent are varied independently. There are three different linguistic views on the relation between duration and tone, each predicting a differenl outcome of such an experiment: (1) No Separate Representations for Meter and Tone; Meter Determines Tone. According to this view, pitch accents are always placed on the metrically most prominent syllable. Tone structure is not represented on a separate level but it is just anolher acoustic correlate of metrical structure. This Option was put forward by Chomsky and Halle [1968]. However, they based their rules only on neutral utlerances whose accents always occur in stressed positions. We wil l not go into this Option any further. The fact that there are constructions in which a contrastive accent is realized on an unstressed syllable [Bolingcr, 1961 ] renders it unviable; this position will therefore not be pursued any further in this article. (The alternative possibility, i.e. a single prominence representation in which the position of pitch accents determines meter, has never been advanced by any linguistic theory.) (2) Separate Representations for Meter and Tone; Levels Do Not Internet. According lo this view, duration and tone structures are represented on separate autonomous prosodic levels [Neijt, 1990]. Metrical constituent structure i s reflected by the relative duration of the syllables. Tone structures are generated by the rules and conventions of autosegmental phonology [Gussenhoven, 1988J. Tonal prominence is brought about by a pitch movement on a constituent that places that constituent, or a larger constituent of which it is the prosodic head, in focus [Baart, 1987]. The prediction of this view is that contrastive accents do not affect the relative duration of the syllables, no matter where an accent is placed. (3) Separate Representations for Meter and Tone; Tone Determines Meter. This view suggests that pitch accents are able to change not only absolute duration, but also the relative duration structure of words. Tone and duration therefore have separate prosodic levels but are not independent: lexically unstressed syllables carrying a pitch accent have to be made metrically prominent. This view predicts that the relative duration of syllables within a word is affected by shifting the accent onto an unstressed syllable. Selkirk [1984] endorses this view. In her opinion the metrical level accounts for the prominence relations and rhythmical organization of the various constituents. The assignment of a pitch accent, which may even be assigned to a weak syllable äs in cofFEE vs. cof'FIN, changes the metrical structure: 'an accented syllable is more prominent (on the grid) than any syllable that is not associated with a pitch accent' [Selkirk, 1984, p. 152]. The main aim of our experiment is to choose between the remaining views (2) and (3). 1t is therefore necessary to know what happens to the relative duration of syllables when a (narrow-focus) accent is realized on a lexically unstressed syllable. Accentuation is used to focus [Ladd, 1980; Baart, 1987], i.e., to highlight a word, a pari of