patterns of accentual lengthening in english four-syllable words

16
Patterns of accentual lengthening in English four-syllable words Snezhina Dimitrova a,b , Alice Turk a,n,1 a Linguistics and English Language, University of Edinburgh, Dugald Stewart Building, 3 Charles Street, Edinburgh, EH8 9AD Scotland, UK b Department of English and American Studies, University of Sofia ‘‘St Kliment Ohridski’’, 15 Tsar Osvoboditel Blvd., Sofia 1504, Bulgaria article info Article history: Received 19 October 2010 Received in revised form 23 February 2012 Accepted 24 February 2012 Available online 20 March 2012 abstract Previous work on English disyllabic and trisyllabic words cannot distinguish two types of views on accentual lengthening: (1) phrasal accent affects a single, multisyllabic domain (the entire word), vs. (2) phrasal accent affects multiple, potentially separate, domains (e.g., the primary stressed syllable and the final syllable). In the present paper, we distinguish these views by examining the effect of phrasal accent on the durational patterns of English four-syllable words. We studied words of three types, with different positions of primary and secondary lexical stress: pattern 1000 (e.g. " presidency), pattern 2010 (e.g. "demo " cratic), and pattern 1020 (e.g. " suffo" cating). Our results show that accent-related lengthen- ing can affect multiple, potentially distinct, sites: the primary-stressed syllable, the secondary-stressed syllable rhyme (if the word has secondary stress), the onset of the word-initial syllable, and the final syllable. In addition, lengthening can ‘‘spill over’’ from a primary-stressed syllable onto a following unstressed syllable. Patterns of accent-related lengthening on onsets vs. rhymes are qualitatively different for stressed vs. word-edge sites, suggesting separate lengthening mechanisms. & 2012 Elsevier Ltd. All rights reserved. 1. Introduction This paper investigates the distribution of accentual lengthening in contrastively accented four-syllable words of varying lexical stress patterns in Scottish Standard English. Previous work has shown that in an English disyllabic word such as bacon, both syllables are longer when that word is contrastively accented, as compared with the same word when unaccented. In a trisyllabic word such as property, accent-related lengthening has been found to extend throughout all three syllables (Turk & White, 1999), although lengthening on the second syllable is reduced in magnitude as compared to the first and third syllables. These findings do not make it possible to distinguish between two views: (1) that accentual lengthening affects a continuous domain, e.g. the whole word (Sluijter, 1995; White, 2002), or a domain extending from a pitch accent to the end of a word (Turk & White, 1999), or (2) the alternative view that accentual lengthening affects a number of distinct, separate domains, such as the lexically stressed syllable and the unstressed syllable immediately following it, as well as the final syllable in a word. We therefore investigate patterns of durational adjustment under contrastive focus in words of more than three syllables, in which lexical stress is further removed from the end of the word. Prominence, or the (perceptual) salience of certain parts of an utterance, is observed in many languages of the world, and is associated with acoustic correlates such as greater acoustic inten- sity, less spectral tilt, f0 peaks or valleys, more peripheral vowel quality, and longer duration. The terminology used when discussing prominence phenomena in speech is often ambiguous. In this paper, lexical stress is used to describe the phonological strength patterning of syllables within a polysyllabic word (note that this is often referred to as word stress). We distinguish three degrees, or levels, of lexical stress: primarythe main stress in a word, secondarythe next level of stress below the primary, and unstressedthe weak stress on syllables not having primary or secondary stress. The surface phonetic properties of these syllables will depend on the prosodic structure of the utterance in which the word occurs. For example, if the word bears phrasal prominence, an intonational pitch accent will be associated with a lexically stressed syllable, and parts of the word will be lengthened. Several different terms are used in the literature to denote phrasal prominence. We use the term accent to refer to abstract phrasal prominence, which can have many physical correlates. Our focus in this paper is on the durational correlate of the type of accent illustrated in (1), i.e. on contrastive accent (‘‘narrow focus’’). In particular, we are interested in the distribution of this correlate throughout the accented word. (1) A: John lost his keys yesterday. B: John lost his wallet yesterday. In (1B), where contrastive phrasal accent occurs on wallet, special highlighting is used to convey contrast with the word keys. In particular, the primary lexical stress wa- of wallet bears an Contents lists available at SciVerse ScienceDirect journal homepage: www.elsevier.com/locate/phonetics Journal of Phonetics 0095-4470/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.wocn.2012.02.008 n Corresponding author. Tel.: þ44 131 650 3483; fax: þ44 131 650 6883. E-mail address: [email protected] (A. Turk). 1 The authors are listed in alphabetical order. Journal of Phonetics 40 (2012) 403–418

Upload: snezhina-dimitrova

Post on 13-Sep-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Journal of Phonetics 40 (2012) 403–418

Contents lists available at SciVerse ScienceDirect

Journal of Phonetics

0095-44

doi:10.1

n Corr

E-m1 Th

journal homepage: www.elsevier.com/locate/phonetics

Patterns of accentual lengthening in English four-syllable words

Snezhina Dimitrova a,b, Alice Turk a,n,1

a Linguistics and English Language, University of Edinburgh, Dugald Stewart Building, 3 Charles Street, Edinburgh, EH8 9AD Scotland, UKb Department of English and American Studies, University of Sofia ‘‘St Kliment Ohridski’’, 15 Tsar Osvoboditel Blvd., Sofia 1504, Bulgaria

a r t i c l e i n f o

Article history:

Received 19 October 2010

Received in revised form

23 February 2012

Accepted 24 February 2012Available online 20 March 2012

70/$ - see front matter & 2012 Elsevier Ltd. A

016/j.wocn.2012.02.008

esponding author. Tel.: þ44 131 650 3483; fa

ail address: [email protected] (A. Turk).

e authors are listed in alphabetical order.

a b s t r a c t

Previous work on English disyllabic and trisyllabic words cannot distinguish two types of views on

accentual lengthening: (1) phrasal accent affects a single, multisyllabic domain (the entire word), vs.

(2) phrasal accent affects multiple, potentially separate, domains (e.g., the primary stressed syllable and

the final syllable). In the present paper, we distinguish these views by examining the effect of phrasal

accent on the durational patterns of English four-syllable words. We studied words of three types, with

different positions of primary and secondary lexical stress: pattern 1000 (e.g."presidency), pattern 2010

(e.g. "demo"cratic), and pattern 1020 (e.g.

"suffo" cating). Our results show that accent-related lengthen-

ing can affect multiple, potentially distinct, sites: the primary-stressed syllable, the secondary-stressed

syllable rhyme (if the word has secondary stress), the onset of the word-initial syllable, and the final

syllable. In addition, lengthening can ‘‘spill over’’ from a primary-stressed syllable onto a following

unstressed syllable. Patterns of accent-related lengthening on onsets vs. rhymes are qualitatively

different for stressed vs. word-edge sites, suggesting separate lengthening mechanisms.

& 2012 Elsevier Ltd. All rights reserved.

1. Introduction

This paper investigates the distribution of accentual lengtheningin contrastively accented four-syllable words of varying lexicalstress patterns in Scottish Standard English. Previous work hasshown that in an English disyllabic word such as bacon, bothsyllables are longer when that word is contrastively accented, ascompared with the same word when unaccented. In a trisyllabicword such as property, accent-related lengthening has been found toextend throughout all three syllables (Turk & White, 1999), althoughlengthening on the second syllable is reduced in magnitude ascompared to the first and third syllables. These findings do not makeit possible to distinguish between two views: (1) that accentuallengthening affects a continuous domain, e.g. the whole word(Sluijter, 1995; White, 2002), or a domain extending from a pitchaccent to the end of a word (Turk & White, 1999), or (2) thealternative view that accentual lengthening affects a number ofdistinct, separate domains, such as the lexically stressed syllable andthe unstressed syllable immediately following it, as well as the finalsyllable in a word. We therefore investigate patterns of durationaladjustment under contrastive focus in words of more than threesyllables, in which lexical stress is further removed from the end ofthe word.

Prominence, or the (perceptual) salience of certain parts of anutterance, is observed in many languages of the world, and is

ll rights reserved.

x: þ44 131 650 6883.

associated with acoustic correlates such as greater acoustic inten-sity, less spectral tilt, f0 peaks or valleys, more peripheral vowelquality, and longer duration. The terminology used when discussingprominence phenomena in speech is often ambiguous. In this paper,lexical stress is used to describe the phonological strength patterningof syllables within a polysyllabic word (note that this is oftenreferred to as word stress). We distinguish three degrees, or levels,of lexical stress: primary—the main stress in a word, secondary—thenext level of stress below the primary, and unstressed—the weakstress on syllables not having primary or secondary stress. Thesurface phonetic properties of these syllables will depend on theprosodic structure of the utterance in which the word occurs. Forexample, if the word bears phrasal prominence, an intonationalpitch accent will be associated with a lexically stressed syllable, andparts of the word will be lengthened.

Several different terms are used in the literature to denotephrasal prominence. We use the term accent to refer to abstractphrasal prominence, which can have many physical correlates.Our focus in this paper is on the durational correlate of the type ofaccent illustrated in (1), i.e. on contrastive accent (‘‘narrowfocus’’). In particular, we are interested in the distribution of thiscorrelate throughout the accented word.

(1)

A: John lost his keys yesterday. B: John lost his wallet yesterday.

In (1B), where contrastive phrasal accent occurs on wallet,

special highlighting is used to convey contrast with the word keys.

In particular, the primary lexical stress wa- of wallet bears an

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418404

intonational pitch accent, and both syllables of the word aretypically longer than they are in a non-phrasally prominentcontexts, e.g. as in (2B).

(2)

A: John lost his wallet yesterday. B: John found his wallet yesterday.

Our strategy is to compare the durational patterns of contras-tively focused four-syllable words with the same words whenthey are unaccented. The phenomenon of de-accenting in alanguage like English typically affects contextually repeated(‘‘given’’) information and is characterised in acoustic terms bya shift of the pitch accent away from the lexically stressed syllableof the repeated unit. In (2B), the word wallet will be unaccented asevidenced by the shift of the pitch accent to the preceding verb.The unaccented rendition of a word thus provides a baselineagainst which temporal adjustments of the same word whenproduced with a contrastive pitch accent can be compared.

Some authors distinguish the ‘‘pragmatic focus domain’’ – thelinguistic unit which is being contrasted or corrected, and the‘‘durational adjustment domain’’ – the linguistic unit whoseduration is adjusted under contrastive focus (Chen, 2006), alsocalled the ‘‘locus’’ by White (2002). While virtually any linguisticunit—a sound, a syllable, a morpheme, a word, etc., can becontrasted, the durational effects induced by such contrast arenot necessarily co-extensive with that unit.

It is traditional to refer to durational effects of prosodicstructure (prominence and boundaries) as lengthening, as if toimply that there is a default, shorter, durational representation inmemory that speakers adjust (cf. e.g. Byrd & Saltzman, 2003).Although we do not want to commit ourselves to such a view, wewill nevertheless follow tradition in referring to durational effectsof accent as lengthening.

We are interested in relatively long, four-syllable wordsbecause they will allow us to distinguish the two theoriesmentioned above, namely (1) that contrastive accent targets acontinuous domain (the Continuous Domain hypothesis) vs. (2)that contrastive accent targets multiple targets within the word(the Multiple Site hypothesis) (Turk & White, 1999). We discussthese alternatives in more detail below.

2. Background

Previous research on contrastive accent has shown that thedurational effects of accent are not confined to a single segmentor syllable (Chen, 2006 for Chinese; Suomi, 2007 for Finnish;Heldner & Strangert, 2001 for Swedish; Bouchhioua, 2007 forTunisian Arabic; Cambier-Langeveld & Turk, 1999; Eefting, 1991;Nooteboom, 1972; Sluijter, 1995; Sluijter & van Heuven, 1995 forDutch; Sluijter, 1995; Turk & Sawusch, 1995, 1997; Turk &White,1999 for English). In English, all segments in monosyllabic CVCwords (e.g. bake) can be longer when the word is accented thanwhen unaccented, although the effect of accent can be greater onthe onset than on the coda (Sluijter, 1995; Turk & Sawusch, 1995,1997; Turk & White, 1999). Both syllables of disyllabic words (e.g."bacon, en

"force) can be longer when accented as compared to

unaccented, although magnitudes of lengthening on non-primarystressed syllables differ depending on position: e.g. –on of bacon

(final and immediately following the primary stress, 13% length-ening), shows greater lengthening effects than en- of enforce

(initial and immediately preceding the primary stress, 5% length-ening, only reliable for some speakers).

The effect of accent appears to be influenced by word boundarylocation in two ways. First, the magnitude of accentual lengtheningis greater on the primary stressed syllable of monosyllabic words

than it is on the primary stressed syllable of disyllabic words.Studies of the effect of accent on identical phoneme sequences suchas bake enforce vs. bacon force and there foursquare vs. therefore

square show 23% lengthening on monosyllabic words, e.g. bake inbake enforce, there in there foursquare, and 16% lengthening on thesame syllables in disyllabic words, e.g. bac- in bacon force and there-

in therefore square (Turk & White, 1999; White & Turk, 2010).Secondly, the stretch of speech affected by accent is influenced byword boundary location. In these homophonous sequences, themagnitude of lengthening on segments or syllables adjacent to theaccented vowel is significantly diminished if a word boundaryintervenes. For example, Turk and White (1999) showed a dura-tional difference of only 4% on e.g. en- of bake enforce when bake wasaccented as compared to when the sequence was unaccented. Incontrast, they found a much larger durational difference when noword boundary intervened: e.g. –on of bacon in bacon force was onaverage 13% longer when bac- was accented, as compared to whenthe sequence was unaccented. Studies of the effects of accent onpoly-morphemic words, e.g. therefore suggested that neither foot normorpheme boundaries block these effects; -fore of therefore showedcomparable accentual lengthening effects to –on of bacon (13% inTurk & White, 1999).

These results show that words are implicated in some way in thedescription of the stretch of speech affected by contrastive accent inEnglish words. However, characterizing this stretch of speech isdifficult for several reasons. First, the affected segments are to someextent speaker-specific: one of six speakers in Turk and White’s(1999) study showed no evidence of accentual lengthening one.g. –on of bacon, and magnitudes of lengthening varied considerablyacross speakers. The small lengthening effects on initial syllables(e.g. en- of enforce, four- of foursquare) were only present for somespeakers and, unlike effects for final syllables, were not reliable inby-Speaker analyses. Secondly, word boundaries appear to attenu-ate, but not block, effects of accent on adjacent syllables. Smalleffects on adjacent syllables across a word boundary have beencalled ‘‘spill over’’, or ‘‘residual’’ effects (Chen, 2006; Turk & White,1999; White, 2002). When a lengthening effect is relatively small inmagnitude (ca. 5%), such as occurs for some speakers on initialunstressed syllables (e.g. en- of enforce), it is difficult to knowwhether it represents a type of ‘‘residual’’ effect (in this caseanticipatory), or should be evidence for a lengthening unit thatincludes the affected syllable (cf. Sluijter, 1995; Turk & White, 1999;White, 2002).

Another problem in characterizing the affected stretch ofspeech relates to the fact that current findings for disyllabic wordslike bacon cannot distinguish between two theoretical possibilities:(1) that accent affects a single, continuous polysyllabic domainsuch as the word, or a unit that extends from the primary stressedsyllable until the end of the word (the Continuous Domainhypothesis) or (2) that accent affects two potentially separatelengthening sites, i.e. the primary stressed syllable and the finalsyllable (the Multiple Site hypothesis). On the latter view, the roleof accent is to highlight the word in two ways: (1) by making theprimary stressed syllable more prominent, and (2) by clearlyindicating the word’s boundaries, in this case through initial andfinal lengthening. Turk and White’s (1999) study of trisyllabicwords like catapult was designed to test these possibilities, butresults were ambiguous. They found lengthening on all threesyllables of these words, with lower magnitudes on the secondsyllable. The fact that all syllables were lengthened seems tosupport the Continuous Domain hypothesis, but the separatelengthening site hypothesis is not ruled out, given the possibilityof spill-over effects on the second syllable. That is, contrastiveaccent may have targeted multiple sites: the primary stressedsyllable, and the final syllable, with an additional, lesser magni-tude, spill-over effect on the second syllable. White’s (2002) results

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418 405

for e.g. "recom"mend and "predi

"spose were similarly ambiguous. He

found significant effects of accent on the first and last syllables,with a small effect (5%) on the medial syllable that was insignif-icant by Subjects, but significant by Items. On the multiple siteview, the small effect on the medial syllable could have been due toeither (1) a spill-over effect from the initial, secondary stressedsyllable, and/or (2) anticipation of the upcoming primary stressedsyllable.

The current experiments were designed to distinguish theContinuous Domain vs. Multiple Site hypotheses using longerwords. Words such as

"presidency, with primary lexical stress on

the first syllable, have a longer stretch of unstressed syllablesbetween the primary stress and the final syllable. A finding of nolengthening on one or both of the intervening syllables wouldsupport the Multiple Site hypothesis. On the other hand, a findingof reliable lengthening on all four syllables would be moreconsistent with the Continuous Domain hypothesis.

In the experiments reported in this paper, we test the dura-tional effect of accent on two types of four-syllable wordscontaining a secondary stress, "demo

"cratic-type words, with

secondary stress on the first syllable and primary stress on thethird,2 and

"suffo" cating-type words, with secondary stress on the

third syllable and primary stress on the first. These words willshow the extent to which the durational implementation ofphrasal accent differentiates between syllable types (primary vs.secondary stress, secondary stress vs. unstressed). Turk andWhite’s (1999) study of initial unstressed and initial secondarystressed syllables (e.g. initial unstressed: en- of enforce, initialsecondary stressed: four- of foursquare) showed no evidence ofdifferent effects of accent across unstressed vs. secondary stressedsyllables, but the small lengthening effects on these syllableswere found only for some subjects, and could be interpreted asanticipatory effects from a lengthening domain that begins withthe primary stressed syllable. Alternatively, the similar behaviourobserved on the initial syllables in Turk and White (1999) acrosssecondary stressed vs. unstressed categories might be due to ageneral tendency for English unstressed initial syllables to adoptcharacteristics of secondary-stressed, full-vowel syllables (manyunstressed initial syllables have secondary stressed pronunciationvariants, e.g. /i/ or schwa in the first syllable of report). Materialsin the current experiment were therefore designed to examinethe effect of accent on non-word-initial secondary stressedsyllables (e.g. the 3rd syllable in suffocating), in addition to theeffect on word-initial secondary stressed syllables (e.g. the firstsyllable in democratic), as tested in previous experiments.

In addition, if evidence favouring the Continuous Domainhypothesis is found for presidency-pattern words, words likesuffocating and democratic will allow us to test possible definitionsof this domain, e.g. that the accentual lengthening domain startswith the primary stressed syllable and extends to the end of theword. On this view, we would expect little or no lengthening onthe first two syllables of democratic, since primary stress is on the3rd syllable –cra-, but lengthening on all syllables in suffocating,where primary stress is on the first syllable su-. If the accentuallengthening domain is the whole word, all word-types shouldshow accent-related lengthening on all syllables.

Finally, we also measure sub-syllabic durations in order to testthe view that lengthening demarcates the edges of focusedconstituents: we call this the Edge Marking hypothesis. If therole of lengthening on the final syllable is to demarcate the word,we would predict that the distribution of lengthening on thatsyllable should mirror the distribution of lengthening found for

2 Preliminary results for democratic-type words were presented in Turk and

Dimitrova (2007).

phrase-final lengthening. That is, we would predict more length-ening on the rhyme of the final syllable than on its onset(Berkovits, 1993a, 1993b, 1994; Cambier-Langeveld, 1997;van Santen, 1994; Turk & Shattuck-Hufnagel, 2007; Wightman,Shattuck-Hufnagel, Ostendorf, & Price, 1992). This pattern ispredicted for all three types of words in our study. Sub-syllabicduration patterns may also reveal the presence of another type ofdemarcative effect, initial lengthening (Fougeron & Keating, 1997;Keating, 2006). Initial lengthening is expected to be localised oninitial consonants of the first syllable in our target words. Theseeffects may be confounded with accentual lengthening in wordswith initial lexical stress, e.g., presidency or suffocating, but may bemore evident in words with main stress on a non-initial syllable,e.g., democratic, where accentual lengthening may be less likelyon a secondary stressed initial syllable (Cho & Keating, 2009).

The results of these experiments have potential implicationsfor theories of phonetic implementation. If we find that lengthen-ing affects a continuous domain extending from the primarystressed syllable until the end of the word, our results will suggestthat a non-traditional prosodic unit, delimited on its left edge by aprimary stress and on its right edge by the end of the word, isimplicated in phonetic implementation. On the other hand, ifresults support the Multiple Site hypothesis, they may suggest arole for units approximately the size of syllables, or syllable sub-components, e.g. syllable onsets and/or rhymes.

3. Methods

Since the methodology used in the experiments reported inthe present paper was the same for both, we will first outline it,and then proceed to a discussion of the experimental results foreach experiment.

3.1. Speakers

The same group of six native speakers of Scottish StandardEnglish took part in each of the experiments. Two of the speakerswere male and four were female. Their age at the time ofrecording was between 19 and 51. They were paid for theirparticipation.

3.2. Test materials

Eleven four-syllable English words were selected for eachexperiment. The words belonged to different grammatical cate-gories and were not mono-morphemic, as previous researchsuggests that morphological boundaries do not affect the dis-tribution of accent-related durational effects (Turk & White,1999).

The words used in Experiment 1 have lexical stress on theirinitial syllable, followed by three unstressed syllables, e.g.,

"presidency.

Their lexical stress pattern can be represented as 1000. Experi-ment 2 focuses on four-syllable words with both primary andsecondary stress, e.g.,

"suffo"cating, "demo

"cratic. The lexical stress

patterns of these words can be represented as 1020 and 2010respectively, where 1¼primary lexical stress, 2¼secondary stress,and 0¼unstressed. Similar coding of lexical stress has been used by,among others, Nakatani and Aston (1978).

The contexts in which the test words occurred consisted of anitem number followed by two sentences: a background sentence,and a target sentence containing the target word. The aim of thebackground sentence was to prompt a natural reading with either acontrastive pitch accent on the target word, or with the target wordunaccented. We will refer to these two experimental conditions asthe ‘‘contrastive’’ or ‘‘accented’’, and the ‘‘unaccented’’ condition,

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418406

respectively. In the sentences in which the target words wereembedded, there were 4–6 syllables preceding, and 1–2 syllablesfollowing each of the target words.

3.3. Recording

All recordings were made in the recording studio of theDepartment of Linguistics and English Language, The Universityof Edinburgh. The recordings were made directly to disk, using ahigh-quality AKG CK98 hypercardoid condenser microphone.They were digitised at a sample rate of 48 kHz and had 16-bitquantisation.

Each item (item number, background sentence and targetsentence) was printed on a separate card, and the cards werestacked in blocks (‘‘contrastive’’ and ‘‘unaccented’’, respectively).The items were then pseudo-randomised within each block (bycard shuffling). Subjects read through two repetitions of the entiredata set with different randomisations for each repetition for eachparticipant.

Before the start of the recording session, the speakers weregiven instructions about the format and the manner of presenta-tion of the reading materials. They were instructed to read aloudall three items on each card at their normal tempo, trying not topause between words in the same sentence. The cards had nohighlighting of any kind to show the place of contrastive focus.Instead, speakers were told that it was especially important tocorrectly understand the intended relationship between the pairof sentences on each card. Finally, they were encouraged to askfor clarification or further instructions, and to repeat sentenceswhich they thought had not been produced properly.

Before recording the test materials, speakers read practicesentence pairs which were not part of the experimental materialbut contained words with the same number of syllables andlexical stress patterning as the target ones.

The speakers’ reading was monitored by the first author whosat in the recording booth with them, and asked them to repeat

Fig. 1. Segmentation and labelling of the ‘‘control’’ sequence and the word serviceable in

On Tier 2, O¼Onset, N¼Nucleus, C¼Coda, and on Tier 3 O¼Onset, R¼Rhyme. Tier 4 is

labelled.

items if pauses, incorrect accent placement or mispronunciationswere detected. In a few cases, it was necessary to record someadditional repetitions at the end of a block.

The overall number of tokens which was recorded for each ofthe three lexical stress patterns is 264: 11 words�2 con-ditions�2 repetitions�6 speakers.

3.4. Measurements

The target words were segmented and labelled by hand inPraat (Boersma & Weenink, 2009). Segmentation was done withreference to waveforms and spectrograms, accompanied by audi-tory evaluation of the recording. Cases of misplaced pitch accentwere evaluated aurally by both authors and discarded fromfurther analysis. The segmentation and labelling procedure isillustrated in Fig. 1.

Segmental boundaries were placed on the basis of identifyingoral constrictions and the intervals between them, as described inTurk, Nakai, and Sugahara (2006). Consonants were judged tobegin at the onset of constrictions, and to end at constrictionrelease. In cases where constriction release could not be identifiedreliably (as when fricatives may have been followed by loweramplitude aspiration noise, see /s/ in Fig. 1), the constriction wasjudged to end, and the following vowel was judged to begin, atthe onset of voicing for the vowel. Normally, however, vowelswere judged to begin at the release of a preceding consonantalconstriction, and to end at the onset of constriction for a followingconsonant, and thus included the VOT of a preceding aspiratedstop, when one occurred.

Segmental and syllable boundaries as well as syllable onsets,peaks, codas and rhymes were all labelled on separate tiers.Syllabification generally followed the Maximal Onset Principle. Forexample, in words like democratic, syllabification was de.mo.cra.tic.The segmentation and labelling was double-checked by the firstauthor. Where boundary placement was problematic or unreliable,data points were discarded, along with the corresponding data point

the sentence ‘‘The old tools will be serviceable for us.’’ Tier 1 is the Segmental Tier.

the Syllable Tier, and Tier 5 is the Word Tier, on which the control sequence is also

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418 407

from the compared condition. Finally, syllable and sub-syllabiccomponent durations were extracted, again using Praat.

In addition to the target words, in each target sentence acontrol sequence consisting of 2 to 4 sound segments was alsosegmented and labelled.

3.5. Analyses

Durations of all control sequences, syllables and sub-syllabiccomponents (onsets and rhymes) in the experiments reported inthis paper were analysed using linear mixed effects models (seeBaayen, Davidson, & Bates, 2008 for an accessible introduction).These analyses were implemented in R, an open-source languageand environment for statistical computing (R Development CoreTeam, 2009). The lme4 package was employed for parameterestimation and model evaluation (Baayen, 2008; Bates, 2005).Linear mixed effects models are types of linear regression modelscalled mixed because they can include both fixed and randomfactors. Similar inferences about factor significance can be drawnfrom linear mixed effects analyses as can be drawn from ANOVAor repeated measures ANOVA. One advantage of linear mixedeffects models over traditional (repeated measures) ANOVA isthat more than one random factor can be included.

In all analysis models, Speaker and Testword were used ascrossed random factors. In Experiments 1 and 2, where we wereprimarily interested in the possible effect of accent on syllablesand syllable sub-components, Accent was the only fixed factor. InSection 5.4, where we were interested in differences between therelative magnitude of the effect of accent on onsets vs. rhymes,we included the fixed factors, Accent and Onset_vs_Rhyme aswell as their interaction (AccentnOnset_vs_Rhyme).

Most of the analyses reported here use raw, non-log-trans-formed duration as a dependent variable. Log transformationsare sometimes used in the durational literature in cases wherethe distributions of durations are skewed (long-tailed), in orderto improve the normality of the duration distributions. However,for the majority of cases reported here, log transforming thedata did not significantly improve the normality of the durationdistributions. In cases where log transformations did improvenormality (1/3 of the cases), analyses of log durations showedsimilar results for the factor of interest (Accent) as the analyses ofraw durations.3

Before testing for the significance of fixed factors (e.g. Accent)on a given dependent variable (e.g. syllable duration), we firstassessed the possibility that the effects of fixed factors of interestwere significantly different for different speakers or test words.To do this, we compared the fits of a simple regression modelincluding Speaker and Testword as crossed random factors, aswell as one or more fixed factors, with the fits of more complexmodels that included additional random factors. When the onlyfixed factor was Accent, these additional random effects were(1) by-Speaker random slopes for Accent, and (2) by-Testwordrandom slopes for Accent. These slopes are analogous to interac-tions of e.g. (1) Subject by Accent, and (2) Testword by Accentin models where Speakers, Testwords and Accent are all fixedfactors. When the fixed factors were Accent, Onset_vs_Rhyme andAccentnOnset_vs_Rhyme (Section 5.4), the additional randomeffects were (1) by-Speaker random slopes for AccentnOnset_vs_Rhyme, and (2) by-Testword random slopes for Accent-nOnset_vs_Rhyme. These slopes are analogous to the three-wayinteractions of e.g. (1) Speaker by Accent by Onset_vs_Rhyme, and

3 In two cases (the onset of syllable 1 for 1000 pattern words, and the rhyme

of syllable 4 for 2010 pattern words), the log duration analyses showed no

evidence of a by-Speaker difference in the magnitude of the accent effect, whereas

the raw duration analyses had shown a by-Speaker difference in effect magnitude.

(2) Testword by Accent by Onset_vs_Rhyme in models whereSpeakers, Testwords and Accent are all fixed factors.

For each set of data, likelihood ratio tests were used tocompare the simple regression model (including the crossedrandom effects as well as the fixed factor(s) of interest), witheach of the more complex models. In these tests, twice thedifference between the log likelihood for the more complexmodel and the log likelihood for the simple model follows achi-squared distribution (Baayen, 2008). When random slopescontributed significantly to the regression model, the complexmodels including these slopes were used when assessing thesignificance of fixed factors. Otherwise, the simple regressionmodel was used. In all cases, the significance of the fixed factorswas assessed using t-tests within the pvals function of language R.

In Section 5.4, where one of our goals is to determine whetheronsets vs. rhymes show different amounts of accent-relatedlengthening, we report analyses using both raw and log10-trans-formed durations. The analysis of raw durations allows us tocompare the effect of accent on onsets vs. rhymes in absoluteterms. The analysis of log-transformed durations allows us toassess the effect of accent on onsets vs. rhymes in proportionalterms, because the logarithmic transform converts multiplicativerelationships into additive relationships (Napier, 1614; Wooldridge,2009). We use both types of analyses because it is unclear which isthe most appropriate, particularly when comparing effect sizeson different segment types (consonants in the case of onsets vs.primarily vowels in the case of rhymes).

Most analyses reported in Section 5.4 involved within-test-word comparisons and used the lmer procedure outlined above.However, one of the analyses in Section 5.4.1 involved a between-testword comparison of word-initial onsets vs. word-medialonsets in the same accent context, where the word-initial onsetsand word-medial onsets came from different word sets. For thisanalysis, we used a by-Subject repeated measures ANOVA, withDuration as a dependent variable, and Position (first vs. thirdsyllable), Accent (Accented vs. Unaccented), and Stress status(Primary vs. Secondary) as independent variables. For this analy-sis, we report only the effects of interest, that is, the interactionsbetween (1) Position (first vs. third syllable) and Accent (Accentedvs. Unaccented), and (2) Position, Accent, and Stress_status(Primary vs. Secondary).

Results of all analyses reported in the paper are presentedwithout any correction for multiple comparisons. While somestatisticians advocate such a correction (e.g. of the Bonferronitype), others recommend not using one in cases like ours (Bland &Altman, 1995), i.e. where comparisons are not independent ofeach other. For example, of interest here, comparisons of syllablerhymes are not independent from comparisons of whole syllables,nor arguably are any of the comparisons made on the samesubject’s data. Note that the lack of Bonferroni corrections for ouranalyses make it more likely that we will find significant differ-ences and therefore that we should find support for the Contin-uous Domain hypothesis.

3.6. Control sequence durations

First of all, control durations were analysed in order to ensurethat speakers had maintained a regular tempo across experimen-tal conditions. Comparisons of control sequence durations in thecontrastive target sentences with those in the respective unac-cented target sentences for the six speakers for all three groups ofwords used in this study did not show any significant differences.We therefore assume that any significant effects of Accent on thedurations of the target units (words, syllables and sub-syllabicconstituents) are not due to overall rate of speech differences.

Table 1Number of analysed syllables and sub-syllabic components (onsets and rhymes)—

1000 pattern words (e.g., presidency).

N

Syllables Onsets Rhymes

Syllable 1 196 200 192

Syllable 2 146 100 90

Syllable 3 154 124 126

Syllable 4 204 174 180

Syllables 2þ3 196 – –

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418408

4. Experiment 1

4.1. Introduction

Experiment 1 was designed to distinguish the ContinuousDomain hypothesis from the Multiple Site hypothesis. The Con-tinuous Domain hypothesis states that accentual lengtheningextends over a single multisyllabic domain beginning with theprimary-stressed syllable. On the other hand, the Multiple Sitehypothesis predicts that several distinct sites are potentiallyaffected by phrasal accent, such as the primary-stressed and thefinal syllable in a lexical word.

In order to investigate these two patterns of accentual length-ening we used 4-syllable words with primary lexical stress ontheir initial syllable. In these words, the primary-stressed syllableoccurred as far away as possible from the final syllable, withoutany intervening lexically stressed or full vowel syllables, e.g."presidency. We refer to these as the ‘‘1000 pattern’’, or the

‘‘presidency-pattern’’ words, where 1¼primary stressed syllable,and 0¼unstressed syllable.

4.2. Method

4.2.1. Test materials

In order to study the accentual lengthening behaviour ofpresidency-pattern words, we selected 11 four-syllable wordswith lexical stress on their first syllable, whose phonemic make-up was judged relatively unlikely to posit any major segmenta-tion problems. The 11 target words, along with the complete setof test materials, are given in Appendix A. The contexts in whichthe target words occurred in the experiment are illustrated below,using the word presidency:

4 Th

hard to

addition

thought

this was

6 speak

to be e

purchas

because

(3)

(contrastive) (It’s been a stable opposition for him.) It’s been a stable presidency for him.

(4)

(unaccented) (It’s been an awful presidency for him.) It’s been a stable presidency for him.

4.2.2. Measurements

Some of the 264 tokens recorded for this experiment had to beexcluded from the statistical analyses,4 because target wordswere produced with unexpected numbers of syllables or promi-nence patterns. In a few cases, further syllables or sub-syllabiccomponents had to be discarded because of segmentation diffi-culties. Table 1 summarises the number of analysed tokens(syllables and sub-syllabic components). The lower number oftokens for syllables 2 and 3 was due to segmentation difficulties.Therefore, we also included in the analysis the sum of thedurations of syllables 2 and 3, which was analysed for caseswhere the boundary was uncertain. Onsets and rhymes, however,were analysed separately, and only for those syllables for which aboundary had been reliably determined.

e first reason for this was that, in the recordings of several speakers, it was

find acoustic evidence for all 4 syllables of the target word seasonable. In

, when asked several weeks after the recording how many syllables they

they had in this word, speakers replied almost unanimously that for them,

a three-syllable word. This is why the word seasonable, as produced by all

ers, was excluded from the analysis. For two of the speakers, noticeable had

xcluded on the same grounds, and for one of those two speakers,

able and desolately were also excluded. A few further items were excluded

of wrong pitch accent placement.

4.3. Results

4.3.1. Syllable durations

The pooled data from all six speakers for the presidency-patternwords (Fig. 2) supports the Multiple Site hypothesis. Phrasal accentsignificantly affected the durations of Syllable 1 (the initial, primarystressed syllable) and 4 (the final syllable) only. Syllables 2 and 3, onthe other hand, were unaffected by Accent, either when analysedseparately or together. Syllable 1 was 21 ms (10.2%) longer when theword was contrastively accented as compared to unaccented,t(195)¼�4.974, po0.001, and Syllable 4 was 8 ms (4.6%) longer,t(203)¼�2.908, p¼0.004.

Our data also showed that the effect of Accent on Syllable1 differed by Testword. We found a significant contribution of by-Testword random slopes for Accent condition (contrastive vs.unaccented) for Syllable 1 (w2(2)¼6.1846, N¼196, p¼0.045). Forexample, in testword 1 candidacies, Syllable 1 was unexpectedlyslightly longer (by about 2 ms) when the word was unaccented ascompared to accented; Syllable 1 in testword 10 desolately, on theother hand, was affected more strongly by Accent (49 ms effectsize) than was average (21 ms) for this word set.5 No significantby-Testword or by-Speaker contribution was found for any of theother syllables.

The lengthening pattern for presidency-type words runs counterto the Continuous Domain hypothesis according to which accentaffects a multisyllabic domain which starts with the main-stresssyllable and extends to the word-edge. Our findings of two separatelengthening sites support the view that duration is used in two waysto signal contrastive focus (i) by highlighting the prominent (pri-mary-stressed) initial syllable, and (ii) by demarcating the right edgeof the word. If indeed two lengthening mechanisms are at work(prominence-related lengthening and final lengthening), we mightalso expect different distributions of lengthening on the prominentvs. final syllables. This is because studies in the literature (Berkovits,1993a, 1993b, 1994; Cambier-Langeveld, 1997; van Santen, 1994;Turk & Shattuck-Hufnagel, 2007; Wightman et al., 1992) have foundthat final lengthening affects syllable rhymes more than onsets,while accent spreads durational effects more uniformly across thesyllable.

In order to check this prediction, we next analysed sub-syllabicconstituent durations, namely, syllable onsets and syllable rhymesin 1000 (e.g., presidency) pattern words.

4.3.2. Onset–Rhyme durations

The distribution of lengthening over the onsets and rhymes ofthe four syllables of the 1000 (e.g., presidency) pattern words isshown in Fig. 3.

5 An (impressionistic) explanation for this is that at least some of the speakers

tended to dramatise their reading of this particular word, adding extra emphasis.

None of the other words were produced in this way.

90

110

130

150

170

190

210

230

syl 1* syl 2 syl 3 syl 4*

Dur

atio

n (m

s)

contrastiveunaccented*

*

Fig. 2. Mean syllable durations (and standard error bars)—pooled data from all

6 speakers for the 1000 pattern target words (e.g., presidency). Asterisks indicate a

significant effect of Accent.

20

40

60

80

100

120

140

160

onset*syl 1*

onsetsyl 2

onsetsyl 3

onsetsyl 4*

Dur

atio

n (m

s)

contrastiveunaccented

*

*

(*)

*

rhyme(*) rhymerhyme* rhyme*

Fig. 3. Mean onset and rhyme durations (and standard error bars) of the four

syllables of presidency-pattern testwords. Asterisks indicate a significant main

effect of Accent, and brackets indicate a tendency towards significance.

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418 409

Results for Onsets and Rhymes support the view that twodifferent types of lengthening mechanisms are at work whentarget words are contrastively accented. Both the onset and therhyme of the initial primary stressed syllable were affectedsignificantly, as expected from prominence-related lengthening(Onset: 15 ms (22.6%) longer when accented as compared tounaccented, t(199)¼�3.455, po0.001; Rhyme: 6 ms (4.4%)longer, t(191)¼2.513, p¼0.017). A significant amount of length-ening was also observed on the rhyme, but not on the onset,of the final syllable, as expected from final lengthening (6 ms(5.3%), t(179)¼�2.249, p¼0.026). These results are consistentwith the view that speakers use duration in two ways to highlightaccented words: (1) to make the primary stressed syllablemore prominent, and (2) to demarcate the word. There was atendency towards significant lengthening of the rhyme of thesyllable immediately following the pitch accent (6 ms (10.4%),t(89)¼�1.862, p¼0.066, ns); this may be evidence for a spill-over effect from the main stress, as reported in earlier studies(Turk & White, 1999).

Although the general pattern involving lengthening of the onsetof the main stress syllable (onset of Syllable 1) is characteristic of all

six speakers, our analyses showed a significant contribution ofby-Speaker random slopes for Accent (w2(2)¼6.5004, N¼200,p¼0.039). An examination of the data suggests that the effect ismuch smaller for Speaker 5 than for the others (observed lengthen-ing was less than 5 ms), which may be due to the overall fastertempo characteristic of this particular speaker.

There was also a significant contribution of by-Testwordrandom slopes for Accent (w2(2)¼8.5068, N¼200, p¼0.014) inpredicting the durational behaviour of the Onset of Syllable 1;several of the target words showed considerably smaller amountof onset lengthening on their lexically stressed syllable undercontrastive accent than the average mean of 15 ms, namely, No.1—candidacies, No. 8—profitable, and No. 9—purchasable. How-ever, the general pattern of increasing the duration of the onset ofthe lexically stressed syllable when the word is contrastivelyaccented, was typical of all testwords.

4.4. Discussion

Results from Experiment 1 are consistent with the view thataccentual lengthening does not affect all parts of a phrasallyaccented word. For English four-syllable words with primarylexical stress on the initial syllable (contrastive), accent affectstwo separate, distinct sites—the primary-stressed syllable and thefinal syllable. There was also a tendency for the syllable rhymeimmediately following the primary stress to be longer in theaccented condition than in the unaccented condition. This patternis suggestive of a residual, spill-over effect.

5. Experiment 2

5.1. Introduction

The first goal of Experiment 2 was to determine whethersecondary stress also provides a potential lengthening site. There-fore, in this experiment we investigated accentual lengthening inwords which, in addition to a primary lexical stress, also havesome degree of weaker (secondary) stress. If secondary stressedsyllables serve as lengthening sites, we expect that words such assuffocating should behave differently from presidency-type words,that is, there will be accentual lengthening on the third syllable(e.g., -ca-) in addition to the lengthening sites observed inExperiment 1. If secondary stressed syllables serve as lengtheningsites, words such as democratic are expected to show lengtheningon the secondary stressed syllable (Syllable 1) in addition to thethird (primary stressed), and fourth (final) syllables.

These words might also be expected to provide a further test ofthe Multiple Site vs. Continuous Domain hypotheses. However, apositive finding of lengthening on secondary stressed syllables insuffocating-pattern words would make it difficult to distinguishbetween the two views, because suffocating-type words might beexpected to show lengthening on all syllables under both hypoth-eses. The Continuous Domain hypothesis would predict lengthen-ing on all syllables starting with the primary stressed syllable tothe end of the word. The Multiple Site view would make the sameprediction: lengthening on Syllable 1 (primary stressed syllable),Syllable 3 (secondary stressed syllable), and Syllable 4 (finalsyllable) with possible spill-over lengthening on Syllable 2. Asimilar problem exists for democratic-type words, regardless ofthe lengthening status of secondary stressed syllables. In thesewords, if the accentual lengthening domain begins with theprimary stressed syllable and extends to the end of the word,lengthening would be expected to affect Syllables 3 and 4, just asit would if Syllables 3 and 4 were separate lengthening sitesunder the Multiple Site view. If accent-related lengthening affects

Table 2Number of analysed syllables and sub-syllabic components (onsets and

rhymes)—1020 (e.g., suffocating) and 2010 (e.g., democratic) pattern words.

N

Syllables Onsets Rhymes

1020 pattern words

Syllable 1 224 226 216

Syllable 2 216 176 166

Syllable 3 194 188 186

Syllable 4 214 166 166

2010 pattern words

Syllable 1 244 234 244

Syllable 2 238 156 170

Syllable 3 236 218 214

Syllable 4 250 216 246

90

110

130

150

170

190

210

230

syl 1* syl 2* syl 3* syl 4(*)

Dur

atio

n (m

s)

*

*

* (*)

Fig. 4. Mean syllable durations (and standard error bars)—1020 pattern words

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418410

secondary stressed syllables in these words, the ContinuousDomain view would predict a domain that includes the wholeword. However, the Multiple Site view might make the sameprediction, since lengthening sites would include Syllable 1(secondary stressed syllable, e.g. de-), Syllable 3 (primary stressedsyllable, e.g. -cra-), Syllable 4 (final syllable, e.g. -tic), and possiblySyllable 2 (e.g. –mo-) if lengthening can spill over from Syllable 1(e.g. de-).

As in Experiment 1, syllable sub-component durations (forOnsets and Rhymes) will provide a test of the Edge Markinghypothesis.

5.2. Method

5.2.1. Test materials

We selected one set of eleven four-syllable English words withprimary stress on their initial syllable, followed by oneunstressed, one secondary-stressed, and a final unstressed sylla-ble, e.g.,

"suffo"cating,

"photo"copy. A second set of target words

included four-syllable words with secondary stress on their firstsyllable and primary stress on the penultimate syllable, e.g.,"demo

"cratic, "confi

"dential. We will refer to the first set as the

‘‘1020 pattern’’, or the ‘‘suffocating – pattern’’ words, and to thesecond set – as the ‘‘2010 pattern’’, or the ‘‘democratic – pattern’’words (the numbers refer to degrees of lexical stress as follows:1¼primary-stressed syllable, 2¼secondary-stressed syllable and0¼unstressed syllable).

The two sets of 11 target words each, along with the completeset of test materials, are given in Appendix B. As in Experiment 1,speakers produced two repetitions of each data set, with differentrandomisations for each repetition for each participant. Partici-pants were given the same instructions as in Experiment 1.Examples (5)–(6) using the target word suffocating illustrate thetwo experimental conditions, with the background sentencesgiven in brackets:

(e.g., suffocating).

(5)

(contrastive) (Temperatures are nearly freezing there.)

Temperatures are nearly suffocating there.

(6) (unaccented)

(Gases are nearly suffocating there.)

Temperatures are nearly suffocating there.

5.2.2. Measurements

Two hundred and sixty-four tokens of the 1020 (e.g., suffocat-

ing) pattern words and 264 tokens of the 2010 (e.g., democratic)pattern words were recorded in this experiment. For one of thespeakers, we had to exclude from the analyses two of the 1020target words, namely, persecuted and substituted, because percep-tually as well as in terms of the acoustic signal, these were judgedby both authors to have 3 instead of 4 syllables. In addition, anumber of items were excluded because of wrong pitch accentplacement. Finally, for a few syllables, onsets and rhymes had tobe excluded because of segmentation problems. The overallnumber of analysed syllables and sub-syllabic components isgiven in Table 2.

5.3. Results

5.3.1. Secondary stressed syllables

Our data suggest that secondary stressed syllables can serve aslengthening sites, but only in suffocating-pattern words. Thesecondary stressed syllable in suffocating-pattern words (Syllable 3)showed a significant main effect of Accent (9 ms (5.2%) longer,

t(193)¼�3.889, po0.001). However, for democratic-pattern words,only a tendency towards significance was found for the effect ofAccent on the syllable carrying secondary stress (Syllable 1) e.g. de-

(5 ms (4%) longer, t(243)¼�1.937, p¼0.054). In Section 5.4.1, weshow that only the onset of this syllable is lengthened by Accent. Aswe discuss there, lengthening of the onset is consistent with EdgeMarking, where word-initial lengthening is expected to preferentiallyaffect the first consonant in the word. If the secondary stressedsyllable had been a lengthening site, we would have expected to seesignificant accent-related lengthening on both the onset and rhyme.We therefore conclude that secondary stressed syllables can beaffected by Accent, but only in some word types.

Given the positive evidence for accent-related lengthening onthe secondary stressed syllable (Syllable 3, e.g. –ca- in suffocating),we do not expect our syllable results for this word-pattern todistinguish the Continuous Domain vs. Multiple Site views.Indeed, all four syllables in suffocating-pattern words werelengthened (Fig. 4), although the effect for Syllable 4 only tendedtowards significance. There was a significant main effect of Accentfor the primary stressed syllable—Syllable 1 (20 ms (12.1%),t(223)¼�3.296, p¼0.001), for Syllable 2 (unstressed) (8 ms(6.5%), t(215)¼�4.28, po0.001), and for Syllable 3 (mentionedabove). The effect of Accent on Syllable 4 (the final unstressedsyllable) was small in size and only tended towards significance(5 ms (3%), t(213)¼�1.787, p¼0.075).

For the primary stressed Syllable 1, we also found a significantby-Speaker random slopes contribution (w2(2)¼13.828, N¼224,po0.001), which reflects the fact that some speakers (e.g., Speaker2, Speaker 5) used a considerably smaller amount of lengthening

20

40

60

80

100

120

140

160

onset*syl 1*

rhyme* onsetsyl 2*

rhyme* onsetsyl 3*

rhyme* onsetsyl 4(*)

rhyme

Dur

atio

n (m

s)

**

*

*

Fig. 6. Mean onset and rhyme durations (and standard error bars) of 1020 (e.g.,

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418 411

than others (e.g., Speaker 4). In spite of the inter-speaker variation,however, all speakers preserved the overall pattern of a longerinitial primary-stressed syllable when the target word was (con-trastively) accented than when it was unaccented.

Results for syllable durations of target words from the 2010pattern (e.g., democratic, see Fig. 5) support the view that accent-related lengthening sites include the primary stressed syllable,the final syllable (and perhaps the secondary stressed syllable, asdiscussed above). Significant lengthening was found on thepenultimate, primary-stressed syllable, Syllable 3 (20 ms(10.7%), t(235)¼�3.343, p¼0.001) as well as on the final syllable,Syllable 4 (14 ms (8.9%), t(249)¼�2.498, p¼0.013). We alsoobserved a tendency towards significant lengthening on Syllable1. No significant main effect of Accent was found for Syllable 2.

For the final two syllables of the 2010 (democratic-pattern)target words there was also a significant by-Speaker randomslopes contribution for the primary stressed Syllable 3 (w2(2)¼15.347, N¼236, po0.001) and for Syllable 4 (w2(2)¼12.906,N¼250, po0.01). Although some speakers (e.g., Speaker 4)lengthened syllable 3 considerably more than others (e.g., Speaker2, for whom the amount of lengthening was non-significant), allspeakers followed the general pattern of having a longer primary-stressed syllable when the target word was contrastivelyaccented. However, for Syllable 4, the pattern of lengtheningdiffered depending on the speaker. Five of the speakers length-ened the final syllable (Syllable 4), but one speaker (Speaker 5)did not; this speaker shortened the final syllable by 3.5 ms whenaccented as compared to unaccented. This shortening occurred onthe rhyme of Syllable 4, as discussed in Section 5.3.2.2. Thispattern of results raises the possibility of optional accent-relatedlengthening on some sites. That is, whereas accentual lengtheningon the primary-stressed syllable occurs for all speakers, speakersappear to have a choice as to whether to lengthen otherlengthening sites. It may be worth noting again that this speakerhad the fastest overall speech rate.

Inter-speaker variability not withstanding, all of the speakerspreserve the distinctive word-accentual pattern of democratic-type words under contrastive accent, alternating longer (strong)and shorter (weak) syllables under both experimental conditions.

A surprising result for 1020 (suffocating-pattern) words is thatthe effect on the final syllable is smaller in magnitude comparedto the effect on final syllables for the other two word types underinvestigation, and only tended towards significance. We hypothesise

90

110

130

150

170

190

210

230

syl 1(*) syl 2 syl 3* syl 4*

Dur

atio

n (m

s)

contrastiveunaccented

(*)

*

*

Fig. 5. Mean syllable durations (and standard error bars)—2010 pattern words

(e.g., democratic).

that this smaller amount of lengthening may have something to dowith the preservation of the durational reflex of the lexical stresspattern, that is, longer lexically stressed syllables, followed byshorter unstressed syllables. A great deal of lengthening on the finalsyllable in suffocating-pattern words could make this unstressedsyllable approximately as long as the adjacent secondary stress. It ispossible that speakers constrain lengthening on the final syllable topreserve the durational correlates of the alternating stress pattern inthese words.

5.3.2. Onset–Rhyme durations

Onset–Rhyme durations were investigated as a test of the EdgeMarking hypothesis (Figs. 6 and 7). On this view, we expected tosee more lengthening on onsets than rhymes on initial syllables,and more lengthening on rhymes than onsets for final syllables.Although the effect of Accent is not significant for the finalsyllable onset and rhyme in suffocating-type words, the expectedpattern of results was observed for all word-edge syllables. Edgemarking is seen most clearly in the first syllable of democratic-pattern words, where Accent-related lengthening was observedon the onset, but not the rhyme, of the word-initial syllable,consistent with initial-lengthening.

Results for word-medial syllable sub-components showedthat the effect on the secondary stressed syllable observed in

suffocating) pattern words.

20

40

60

80

100

120

140

160

onset*syl 1(*)

rhyme(*) onsetsyl 2

rhyme onset*syl 3*

rhyme* onset*syl 4*

rhyme*

Dur

atio

n (m

s)

contrastiveunaccented

*

(*)

*

*

*

*

Fig. 7. Mean onset and rhyme durations (and standard error bars) of 2010 (e.g.,

democratic) pattern words.

0

20

40

60

80

100

120

RhymesOnsets

Dur

atio

n (m

s)

contrastiveunaccented

Fig. 8. Mean contrastive vs. unaccented onset and rhyme durations (and standard

error bars) for syllable 1 for all three word types pooled.

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418412

suffocating-pattern words, as well as the effect of spill-overlengthening on Syllable 2 were confined to the rhyme. That thepattern for spillover lengthening was similar to the pattern inExperiment 1 (affecting the rhyme rather than the onset) raisesthe possibility that spillover lengthening may preferentially affectthe syllable rhyme. Details are given below.

5.3.2.1. Word pattern 1020. The onset and the rhyme of the primarystressed Syllable 1 were affected significantly by Accent as follows:the onset was 15 ms (20.6%) longer when contrastively accented(t(225)¼�4.533, po0.001), and the rhyme was 6 ms (6.4 %) longer(t(215)¼�2.727, p¼0.007). Main effects of Accent were also foundon the rhymes (but not on the onsets) of Syllable 2 (about 5 ms(7.5%), t(165)¼�2.242, p¼0.026) and Syllable 3—the secondary-stressed syllable (7 ms (6.8%), t(185)¼�3.256, po0.01).

For the onset of Syllable 1 there was also a significant by-Speakerrandom slopes contribution (w2(2)¼6.7915, N¼226, p¼0.034).Although all six speakers significantly lengthened the onset of theprimary stressed syllable, the amount of lengthening was larger forsome speakers (e.g., 28 ms for Speaker 4) than for others (e.g., only7 ms for Speaker 5, which may be related to the faster tempo ofspeech characteristic of this particular speaker). We also found atendency towards significant by-Testword random slopes contribu-tion for the rhyme of Syllable 1 (w2(2)¼4.7266, p¼0.094). Two ofthe testwords showed rhyme durations that were virtually undis-tinguishable when accented as compared to unaccented (Testword1—criticism: 2 ms shorter when accented; and Testword 8—dedi-

cated: 1 ms shorter when accented).

5.3.2.2. Word pattern 2010. For Syllable 1, which carriedsecondary stress in the democratic-pattern words, there was asignificant main effect of Accent for the onset (9 ms (15.3%) longer(t(233)¼�4.819, po0.001), as well as a tendency towardssignificance of the main effect of Accent on the rhyme(t(243)¼1.945, p¼0.053). The rhyme, however, was on average4 ms (3.2%) shorter when the target word was accented, whichsuggests lack of accentual lengthening on the secondary-stressedsyllables of this word type. Neither the onset nor the rhyme ofSyllable 2 were affected significantly by Accent. For Syllable 3 (theprimary-stressed syllable), on the other hand, there was asignificant effect of Accent for both the onset (8 ms (15.4%)longer, t(217)¼�3.205, p¼0.002) and the rhyme (10 ms (7.2%)longer, t(213)¼�3.188, p¼0.002). Similarly, for Syllable 4 boththe onset and the rhyme were affected significantly: the onsetwas 4 ms (4.7%) longer when the word was contrastively accented(t(215)¼�2.472, p¼0.014), while the rhyme was 11 ms (12.9%)longer (t(245)¼�2.176, p¼0.03).

A significant by-Speaker random slopes contribution wasobserved for the rhyme of Syllable 4 (w2(2)¼10.174, N¼246,p¼0.006), showing that some speakers (e.g. Speaker 1—2 ms)lengthened it considerably less than other speakers (e.g., Speaker4—31 ms), while for one of the speakers (Speaker 5) the final rhymetended to be slightly shorter (by about 4 ms) rather than longerwhen the words were contrastively accented. Also, a tendencytowards significance of by-Speaker random slopes contribution wasfound for the rhyme of Syllable 3 (w2(2)¼4.8763, N¼214, p¼0.087).

5.4. Further analyses of the edge marking hypothesis

In the analyses presented above, the best evidence for initiallengthening came from democratic-type words; for these words,only the onset of the initial syllable was affected by Accent, apattern which is consistent with initial lengthening. For the otherwords, our analyses did not allow us to disentangle prominence-related from edge-related (initial) lengthening since word-initial

syllables bore primary lexical stress. In addition, our analyses didnot allow us to assess relative magnitudes of lengthening ononsets vs. rhymes (both were significantly affected by Accent inpresidency- and suffocating-type words). As for edge marking onfinal syllables, only presidency-pattern words showed unambig-uous evidence for final lengthening when these words wereaccented. In these words, only the rhyme was affected by Accent,a pattern which supports the final lengthening view. However, forthe other word types, our analyses did not provide evidence forthe relative magnitude of lengthening on final rhymes as com-pared with onsets. We therefore performed further analyses ofsub-syllabic component durations of syllables 1 and 4, for allthree of our datasets pooled (1000, 1020, and 2010). Theseanalyses were designed (1) to disentangle prominence-relatedfrom initial-lengthening effects, and (2) to compare magnitudesof lengthening on onsets vs. rhymes. The effects of Accent ononsets vs. rhymes were compared (1) for Syllable 1, where theedge marking view predicts more relative lengthening on onsets,and (2) for Syllable 4, where the edge marking view predicts morerelative lengthening on rhymes.

In this section, we report analyses using both raw and log-transformed durations. The analysis of raw durations allows us tocompare the effect of Accent on onsets vs. rhymes in absoluteterms. The analysis of log-transformed durations allows us toassess the effect of Accent on onsets vs. rhymes in proportionalterms, because the logarithmic transform converts multiplicativerelationships into additive relationships (Wooldridge, 2009). Weuse both types of analyses because it is unclear which is the mostappropriate, particularly when comparing effect sizes on differentsegment types (consonants in the case of onsets, and primarilyvowels in the case of rhymes).

5.4.1. Syllable 1

As shown in Fig. 8, the Edge Marking, initial lengtheninghypothesis prediction that Syllable 1 should show a greater effectof Accent for onsets than for rhymes (significant interaction ofAccentnOnset_vs_Rhyme) was upheld: the effect of interest here,namely, that of the fixed factor AccentnOnset_vs_Rhyme, wasfound to be significant for both absolute and log duration(proportional durational differences). The effect of Accent ononsets was 13 ms (21%), and for rhymes 3 ms (3%) (for durations,t(1249)¼5.326, po0.001, and for log durations, t(1249)¼5.57,po0.001).

0

20

40

60

80

100

120

RhymesOnsets

Dur

atio

n (m

s)

contrastiveunaccented

Fig. 10. Mean contrastive vs. unaccented onset and rhyme durations (and

standard error bars) for syllable 4 for all three word types pooled.

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418 413

A significant contribution of by-Testword random slopes to themodel was also found for both duration (w2(9)¼1814.2, N¼1252,p¼o0.001) and log duration (w2(9)¼1480.6, N¼1252, po0.001).Twenty-six out of the 32 testwords had bigger absolute effects ofAccent for Onsets compared to Rhymes, one had equal absoluteeffect size, and 5 had larger absolute effects on rhymes, comparedto onsets. In proportional terms, 30 of 32 testwords had largerproportional effects on onsets compared to rhymes, and 2 hadlarger effects on rhymes than onsets.

Because all word-initial syllables in our test items bore sometype of prominence—primary stress in the case of 1000 (pre-

sidency-pattern) and 1020 (suffocating-pattern) words, and sec-ondary stress in the case of 2010 (democratic-pattern) words, wecannot immediately conclude that the observed interactionbetween Accent and Onset_vs_Rhyme is due to an edge effectper se, since it could be due to the prominence of the word-initialsyllable. On this view, all prominent syllables would have largereffects of accent on their onsets than on their rhymes, regardlessof their position-in-word. The Edge Marking hypothesis on theother hand, predicts a greater effect of Accent on word-edge(initial) syllable onsets than on word-medial syllable onsets. Thatis, the first syllable in 1000 (e.g., presidency) and 1020 (e.g.,suffocating) words (bearing primary lexical stress) is expected tohave a greater effect of Accent on its onset than the third syllablein 2010 (e.g., democratic) words, and similarly, the first syllable in2010 words (bearing secondary lexical stress) is expected to havea greater effect of Accent on its onset than the third syllable in1020 words.

Although we make the caveat that our comparisons of word-initial vs. word-medial syllable onsets involve different segments,our results, shown in Fig. 9 largely confirm the Edge Markinghypothesis. Word-initial onsets showed an effect of Accent of13 ms (21%) as compared to 4 ms (7%) on the onsets of syllable 3.Results of a by-Subjects ANOVA with absolute duration as adependent variable showed a significant interaction betweenPosition (first vs. third syllable) and Accent (Accented vs. Unac-cented): (F(1,5)¼19.29, p¼0.007), with a non-significant three-way interaction of Position, Accent, and Stress_status (F(1,5)41,ns). For log durations, the results were similar: We found asignificant two-way interaction between Position and Accent,(F(1,5)¼18.5, p¼0.008), again with a non-significant 3-way Posi-tion X Accent X Stress_status interaction (F(1,5)¼1.071, ns).

0

10

20

30

40

50

60

70

80

Onsets - syl 1 Onsets - syl 3

Dur

atio

n (m

s)

contrastiveunaccented

Fig. 9. Mean contrastive vs. unaccented durations (and standard error bars) for

Onset 1 (the onset of syllable 1) and Onset 3 (the onset of syllable 3). Onset 1 data

came from all three word types; Onset 3 data came from 2010 and 1020 word

types only.

5.4.2. Syllable 4

According to the Edge Marking Hypothesis, Syllable 4 shouldshow a greater effect of Accent for Rhymes than for Onsets, that is,a significant interaction of AccentnOnset_vs_Rhyme.

Results are shown in Fig. 10. Although the effects of Accentwere slightly greater on rhymes than on onsets (rhymes 7 mslonger, 7% vs. onsets 3 ms longer, 5%), the interaction betweenAccent and Onset_vs_Rhyme was insignificant both in the Abso-lute duration analysis (t(1145)¼�0.808, ns), and in the logduration analysis (t(1145)¼�0.02, ns).

For absolute durations, there was a significant contribution to thelinear mixed effects regression model of by-Testword random slopes(w2(9)¼1077.8, N¼1148, po0.001) and by-Speaker random slopes(w2(9)¼18.843, N¼1148, p¼0.027). All speakers except Speaker5 had larger durational effects of Accent on Rhymes as compared toOnsets, whereas the reverse was found for Speaker 5. For allspeakers pooled, 21 out of 31 test words showed larger effects ofAccent on Rhymes as compared to Onsets, whereas 10 showed thereverse pattern. For log durations, a significant contribution of by-Testword random slopes to the model was found (w2(9)¼1066.0,N¼1148, po0.001). For these, 17 out of 31 testwords showed largerproportional effects of Accent on rhymes compared to onsets; 14showed the reverse pattern.

6. Summary and discussion

Our study builds on previous work suggesting that contrastiveaccent can affect more than a single syllable in polysyllabic words.Our study contrasted two views: (1) the Continuous Domainhypothesis, whereby contrastive accent lengthens a single, contin-uous domain vs. (2) the Multiple Site hypothesis, where contrastiveaccent lengthens several, distinct sites within the contrastivelyaccented word. Previous work had failed to distinguish these twoviews, primarily because the studied word sizes were too small toreliably separate potential lengthening sites.

Results presented here for longer, 4 syllable words supportedthe Multiple Site hypothesis: contrastive accent did not lengthena single continuous domain. Instead, lengthening sites included(1) the primary stressed syllable, (2) a ‘‘spill-over’’ site thatconsisted of the rhyme immediately following the primarystressed syllable, (3) the final syllable, and (4) the onset of theword-initial syllable (a single consonant in the data presented

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418414

here). The secondary stressed syllable was also a lengthening sitein suffocating-pattern words, where it was word-medial, but didnot appear to be a lengthening site in democratic-type words,where the secondary stress was word-initial.

Evidence for the Multiple Site hypothesis was clearest forpresidency-pattern and democratic-pattern words. In presidency-pattern words, contrastive accent affected the primary stressedsyllable and the final syllable, with ‘‘spill-over lengthening’’ onthe rhyme of the second syllable that tended towards significance.The third (unstressed) syllable was unaffected. In democratic-pattern words, contrastive accent affected the word-onset con-sonant, as well as the primary stressed and final syllables. Thesecond (unstressed) syllable was unaffected.

Suffocating-pattern words showed lengthening on all 4 sylla-bles, although lengthening on the last syllable only tendedtowards significance. We interpret the difference in lengtheningpatterns between suffocating-type words vs. presidency-typewords (i.e. lengthening on all 4 syllables vs. lengthening on only3) as due to the secondary stress on the third syllable insuffocating-pattern words. On this view, suffocating-type wordscontain all 4 possible lengthening sites, that is, the primarystressed syllable (Syllable 1), a ‘‘spill-over’’ site on the rhyme ofthe syllable adjacent to the primary stressed syllable (Syllable 2),the secondary stressed syllable (Syllable 3), and the final syllable(Syllable 4).

In summary, our evidence supports the multiple site view.Note that this outcome was observed in spite of our anti-conservative statistical analysis (we did not perform Bonferronicorrections for multiple tests), which would have made a Con-tinuous Domain outcome more likely.

6.1. Multiple sites and multiple mechanisms: prominence, edge

marking, and spill-over

Why should contrastive phrasal accent affect a number ofdistinct lengthening sites? Our results support the view thatdistinct sites reflect the operation of multiple mechanisms:Prominence-related lengthening, edge-related lengthening, and‘‘spill-over’’ lengthening. Speakers use the temporal effects ofcontrastive accent to make the stressed word more salient in twoways: (1) by making the primary stressed syllable (as well as thesecondary stressed syllable in some word types) more prominent,and (2) by Edge-Marking, via initial- and final-lengthening oncontrastively accented words. In our experiment, we had threetypes of evidence for Edge Marking. First, in presidency-typewords, the only unstressed syllable that was lengthened underaccent was the final syllable, suggestive of final lengthening, andhere the lengthening was localised to the rhyme, the patternexpected for final lengthening. Second, a comparison of lengthen-ing magnitudes on syllable onsets suggests that lengthening dueto accent was greater at word edges than word medially, whenthe stress status of these syllables was taken into account. That is,the effect of contrastive accent was greater on the onset of word-initial primary stressed Syllable 1 in suffocating- and presidency-

type words as compared to the onset of word-medial primarystressed Syllable 3 in democratic-type words. Also, the effect ofcontrastive phrasal accent was greater on the onset of word-initial secondary stressed Syllable 1 in democratic-type words ascompared to the onset of word-medial secondary stressed Sylla-ble 3 in suffocating-type words. Besides, lengthening on the onsetsof word-initial syllables was greater in magnitude than lengthen-ing on the rhyme. We had expected to see a fourth type ofevidence, that is, greater lengthening on the rhyme of the finalsyllable as compared with the onset, as is the expected pattern forfinal lengthening, but this comparison failed to reach significance inour data. In interpreting these results, we caution that word-edge vs.

word-medial comparisons were between-word comparisons, wherecompared segments were not always the same. Similarly, onset vs.rhyme comparisons were also (necessarily) made on differentsegment types (consonants vs. vowels). These results therefore askfor confirmation in future studies.

In the present study, we claim that lengthening has ‘‘spilledover’’ from the primary stressed syllable to a following unstressedsyllable. What explanation do we have for spill-over lengthening?As mentioned in the introduction, spill-over lengthening has beenobserved in previous studies (Chen, 2006; Turk & White, 1999).Evidence for spill-over lengthening comes from studies of theeffects of a preceding boundary on the magnitude of accentuallengthening on a following unstressed syllable. These studies findthat boundaries typically attenuate, but do not always block,lengthening due to accent on an unstressed syllable following themain-stress. The interpretation of these residual effects is thatlengthening has ‘‘spilled over’’ from the preceding syllable. As ofyet, we do not have an explanation for these effects, but note thatanalogous effects in the spatial domain have been observed in thenon-speech motor control literature (Jax & Rosenbaum, 2009; Vander Wel, Fleckenstein, Jax, and Rosenbaum, 2007). Van der Wel,Fleckenstein, Jax, and Rosenbaum (2007) presented experimentsin which participants moved a vertical dowel on a planar surfacefrom target to target in time to a metronome. Some of themovements involved clearing an obstacle between targets. Theyshowed that the spatial paths of hand movements following theobstacle-clearing movements were higher than those seen incontrol, non-obstacle trials, even when the successive movementswere made with the opposite hand. They interpreted their resultsas consistent with the view that movements are representedabstractly, and that planning movements involves setting para-meters (e.g. for hand, spatial characteristics, etc.). They proposethat parameters are set by changing parameter settings used inprevious movements, and that there is a cost related to the size ofparameter changes. We wonder whether the ‘‘spill-over’’ length-ening phenomenon that we observe reflects the cost associatedwith changing the durational parameter for syllables followingthe primary stress.

6.2. Variability

In terms of the order of the Accent effect (Accented longer thanUnaccented), the results reported here were remarkably consis-tent across Speakers and Testwords. Although we did observequite a few cases where the effect of Accent interacted with eitherSpeaker or Testword, an examination of effects of Accent forspecific speakers and testwords in these cases, showed that thenumber of effect reversals was rare. When they did occur, thereversals were very small in magnitude (o4 ms) and arguablynegligible.

Nevertheless, we did observe variability in the magnitude ofthe effect of Accent, particularly across Speakers. This resultsuggests that different speakers may choose to use duration tovarying degrees to signal contrastive accent. For example, it wasoften the case that Speaker 5 showed smaller effects than otherspeakers, and that Speaker 4 typically showed larger effects.Speaker 5 showed an effect reversal for the rhyme of Syllable4 in 2010 (e.g., democratic) pattern words (unaccented 4 mslonger than accented). We speculate that Speaker 5’s faster rateof speech may have had something to do with his limited use ofduration in signalling phrasal accent: On this view, this speakersaved time by limiting the size of the lengthening effect. However,we note that rate of speech can’t be the only factor in determiningthe degree of lengthening: Speaker 4, who had larger lengtheningeffects of Accent than the other speakers, was not noticeablyslower in terms of overall speech rate.

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418 415

In contrast to the by-Speaker variability that appeared to befairly widespread, systematic by-Testword variability was lim-ited. Only one of our lengthening sites showed reliable differencesin the effect of Accent across testwords; this was Syllable 1 of the1000 (e.g., presidency) pattern words, where the testword candi-

dacies showed a 2 ms lengthening reversal (unaccented longerthan accented). The rhyme of Syllable 3 in 1020 (e.g., suffocating)pattern words also showed some by-Testword differences in theeffect of Accent, but these only tended towards significance.

In terms of variability in the effect of Accent across lengtheningsites, we note that lengthening was most reliable on two of the sites:(1) on the primary stressed syllable and (2) on the onset of the word-initial syllable. Main effects of accent on these sites were statisticallyreliable for all word types. Lengthening on the final syllable rhyme,the rhyme of the secondary stressed syllable, and on the rhyme ofthe syllable following the primary lexical stress syllable (the ‘‘spill-over’’ site) depended to some extent on word-type.6 That is, Accent-related lengthening on the final syllable rhyme was significant forpresidency- and democratic-type words, but only tended towardssignificance for suffocating-type words. The secondary stressedsyllable rhyme appeared to be a lengthening target only in suffocat-

ing-type words. Lengthening on the first (secondary stress) syllable ofdemocratic-type words was confined to the onset, and is thereforesuggestive of lengthening on a word-initial onset lengthening site,rather than on a secondary-stress syllable lengthening site per se.Lengthening on the ‘‘spill-over’’ site, the rhyme of the syllablefollowing the primary lexical stress (e.g. on –i- of presidency), or one.g. –o- of suffocating, was reliable in suffocating-type words, but onlytended towards significance in presidency-type words.

Taken together, our results suggest that some lengtheningsites are preferred (i.e. the primary stressed syllable and the wordonset), but that the magnitude of lengthening on these sites, aswell as the magnitude (and likelihood) of lengthening on othersites, may depend on factors such as the speaker’s preference, thestress pattern of the word, to some extent the particular word,and, possibly, rate of speech.

7. Conclusion

In conclusion, the results presented here are consistent withthe view that several lengthening mechanisms are at work inaccented words. These include prominence-related lengthening,edge-related lengthening, and ‘‘spill-over’’ lengthening from theprimary stressed syllable onto a following unstressed syllablerhyme. These lengthening mechanisms appear to be associatedwith distinct lengthening sites, of which the primary stressedsyllable and the onset of the word-initial syllable are preferred.Lengthening magnitudes as well as the likelihood of lengtheningon other sites (i.e. the final syllable, the secondary stressedsyllable rhyme and the rhyme of the unstressed syllable followingthe primary stressed syllable) depend on a variety of factors,including speaker choice, as well as prominence relationshipsamong syllables within a word.

6 In a previous publication reporting on syllable durations of democratic-type

words for four of the speakers’ data presented here (Turk & Dimitrova, 2007), we

concluded that all of the accentual lengthening sites on this word type were

optional, including the primary stressed syllable. This conclusion was made

because the speaker-specific analyses conducted for that paper showed one

subject without significant lengthening on the primary stressed syllable (Speaker

2). The potential optionality of the primary stress syllable site is not apparent in

the analyses presented here. This type of optionality would have been supported

by a significant contribution of by-Speaker random slopes for Accent to the

regression model for Syllable 3 in this word type. Our analyses did not show such

an effect, probably because Speaker 2 did show a small effect (6 ms lengthening)

in the expected direction.

Acknowledgements

This work was supported by AHRC Grant no. B/RG/AN9288/APN14615 to A. Turk.

Appendix A. Test materials—Experiment 1

Target words—lexical stress pattern 1000:candidacies, presidency, seasonable, serviceable, creditable, market-

able, noticeable, profitable, purchasable, desolately, negligibly

word 01: candidacies10143(There’s likely to be new stories again.)There’s likely to be new candidacies again.10144(There’s likely to be good candidacies again.)There’s likely to be new candidacies again.

word 2: presidency10243(It’s been a stable opposition for him.)It’s been a stable presidency for him.10244(It’s been an awful presidency for him.)It’s been a stable presidency for him.

word 03: seasonable10343(The temperatures were unbearable this time.)The temperatures were seasonable this time.10344(The vegetables were seasonable this time.)The temperatures were seasonable this time.

word 04: serviceable10443(The old tools will be kept for us.)The old tools will be serviceable for us.10444(The old spanners will be serviceable for us.)The old tools will be serviceable for us.

word 05: creditable10543(The stories were unbelievable this time.)The stories were creditable this time.10544(The reports were creditable this time.)The stories were creditable this time.

word 06: marketable10643(The new skills could be useful today.)The new skills could be marketable today.10644(The new gadgets could be marketable today.)The new skills could be marketable today.

word 07: noticeable10743(The newer changes are obscure today.)The newer changes are noticeable today.10744(The newer differences are noticeable today.)The newer changes are noticeable today.

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418416

word 08: profitable10843(The new business is unreliable today.)The new business is profitable today.10844(The new company is profitable today.)The new business is profitable today.

word 09: purchasable10943(The new goods are serviceable today.)The new goods are purchasable today.10944(The new buildings are purchasable today.)The new goods are purchasable today.

word 10: desolately11043(She’s likely to cry loudly tonight.)She’s likely to cry desolately tonight.11044(She’s likely to react desolately tonight.)She’s likely to cry desolately tonight.

word 11: negligibly11143(They’re likely to differ extensively next time.)They’re likely to differ negligibly next time.11144(They’re likely to interact negligibly next time.)They’re likely to differ negligibly next time.

Appendix B. Test materials—Experiment 2

Target words—lexical stress pattern 1020:criticism, confiscated, photocopy, commentator, supermarket, pre-

decessor, suffocating, dedicated, duplicated, persecuted, substituted

word 01: criticism20143(It’s open to severe critique again.)It’s open to severe criticism again.20144(It’s open to general criticism again.)It’s open to severe criticism again.

word 02: confiscated20243(Artefacts have been discovered again.)Artefacts have been confiscated again.20244(Documents have been confiscated again.)Artefacts have been confiscated again.

word 03: photocopy20343(It’s luckily a good version this time.)It’s luckily a good photocopy this time.20344(It’s luckily a faithful photocopy this time.)It’s luckily a good photocopy this time.word 04: commentator20443(It’s luckily a clever operator this time.)It’s luckily a clever commentator this time.20444

(It’s luckily a good commentator this time.)It’s luckily a clever commentator this time.

word 05: supermarket20543(It’s luckily a city ice skating rink again.)It’s luckily a city supermarket again.20544(It’s luckily a renovated supermarket again.)It’s luckily a city supermarket again.

word 06: predecessor20643(He was a careful legislator for us.)He was a careful predecessor for us.20644(He was a worthy predecessor for us.)He was a careful predecessor for us.

word 07: suffocating20743(Temperatures are nearly freezing there.)Temperatures are nearly suffocating there.20744(Gases are nearly suffocating there.)Temperatures are nearly suffocating there.

word 08: dedicated20843(Big audiences are simply appreciative again.)Big audiences are simply dedicated again.20844(Big crowds are simply dedicated again.)Big audiences are simply dedicated again.

word 09: duplicated20943(Their efforts are being neglected again.)Their efforts are being duplicated again.20944(Their contributions are being duplicated again.)Their efforts are being duplicated again.

word 10: persecuted21043(Activists are being deported now.)Activists are being persecuted now.21044(Politicians are being persecuted now.)Activists are being persecuted now.

word 11: substituted21143(Clearly the facts have been checked again.)Clearly the facts have been substituted again.21144(Clearly the documents have been substituted again.)Clearly the facts have been substituted again.

Target words - lexical stress pattern 2010:democratic, condescending, confidential, Minnesota, Manitoba,

confiscation, persecution, suffocation, substitution, duplication,

dedication

Word 01: democratic30143(The newest laws are more authoritarian again.)The newest laws are more democratic again.30144

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418 417

(The newest policies are more democratic again.)The newest laws are more democratic again.

word 02: condescending30243(It was extremely disrespectful this time.)It was extremely condescending this time.30244(It was obliquely condescending this time.)It was extremely condescending this time.

word 03: confidential30343(The reports were strictly official this time.)The reports were strictly confidential this time.30344(The consultations were strictly confidential this time.)The reports were strictly confidential this time.

word 04: Minnesota30443(He’ll be flying to Massachusetts next week.)He’ll be flying to Minnesota next week.30444(He’ll be sailing to Minnesota next week.)He’ll be flying to Minnesota next week.

word 05: Manitoba30543(He’ll be flying to Newfoundland next week.)He’ll be flying to Manitoba next week.30544(He’ll be driving to Manitoba next week.)He’ll be flying to Manitoba next week.

word 06: confiscation30643(These goods are subject to approval again.)These goods are subject to confiscation again.30644(These tools are subject to confiscation again.)These goods are subject to confiscation again.

word 07: persecution30743(They escaped from political discrimination again.)They escaped from political persecution again.30744(They suffered from political persecution again.)They escaped from political persecution again.

word 08: suffocation30843(The room was crammed to the brink again.)The room was crammed to suffocation again.30844

(The barn was crammed to suffocation again.)The room was crammed to suffocation again.

word 09: substitution30943

(Avoid any further repetition this time.)Avoid any further substitution this time.30944(Neglect any further substitution this time.)Avoid any further substitution this time.word 10: duplication31043

(Avoid any further repetition this time.)Avoid any further duplication this time.31044(Ignore any further duplication this time.)Avoid any further duplication this time.

word 11: dedication31143(It begins with a simple introduction again.)It begins with a simple dedication again.31144(It begins with a short dedication again.)It begins with a simple dedication again.

References

Baayen, R. H. (2008). Analyzing linguistic data. A practical introduction to statisticsusing R. Cambridge: Cambridge University Press.

Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling withcrossed random effects for subjects and items. Journal of Memory and Language,59, 390–412.

Bates, D. M. (2005). Fitting linear mixed models in R. R News, 5, 27–30.Berkovits, R. (1993a). Progressive utterance-final lengthening in syllables with

final fricatives. Language and Speech, 36(1), 89–98.Berkovits, R. (1993b). Utterance-final lengthening and the duration of final-stop

closures. Journal of Phonetics, 21, 479–489.Berkovits, R. (1994). Durational effects in final lengthening, gapping, and contras-

tive stress. Language and Speech, 37(3), 237–250.Bland, J. M., & Altman, D. G. (1995). Multiple significance tests—The Bonferroni

method (Statistics notes series). British Medical Journal, 310, 170.Boersma, P., & Weenink, D. (2009). Praat: Doing phonetics by computer (Version 5.1.05)

[Computer program]. Retrieved May 1, 2009, from /http://www.praat.org/S.Bouchhioua, N. (2007). The acoustic correlates of stress and accent in Tunisian Arabic:

A comparative study with English. Unpublished Ph.D. Dissertation, Universite de7 Novembre, Carthage; L’Institut Superieur des Langues de Tunis.

Byrd, D., & Saltzman, E. (2003). The elastic phrase: Modeling the dynamics ofboundary-adjacent lengthening. Journal of Phonetics, 31, 149–180.

Cambier-Langeveld, T. (1997). The domain of final lengthening in the productionof Dutch. In: H. de Hoop, & J. Coerts (Eds.), Linguistics in the Netherlands (pp.13–24). Amsterdam: John Benjamins.

Cambier-Langeveld, T., & Turk, A. E. (1999). A cross-linguistic study of accentuallengthening: Dutch vs. English. Journal of Phonetics, 27, 255–280.

Chen, Y. (2006). Durational adjustment under corrective focus in StandardChinese. Journal of Phonetics, 34, 176–210.

Cho, T., & Keating, P. A. (2009). Effects of initial position versus prominence inEnglish. Journal of Phonetics, 37, 466–485.

Eefting, W. (1991). The effect of ‘‘information value’’ and ‘‘accentuation’’ on theduration of Dutch words, syllables, and segments. Journal of the AcousticalSociety of America, 89(1), 412–424.

Fougeron, C., & Keating, P. A. (1997). Articulatory strengthening at edges of prosodicdomains. Journal of the Acoustical Society of America, 101(6), 3728–3740.

Heldner, M., & Strangert, E. (2001). Temporal effects of focus in Swedish. Journal ofPhonetics, 29, 329–361.

Jax, S. A., & Rosenbaum, D. A. (2009). Hand path priming in obstacle avoidance:Rapid decay of dorsal stream information. Neuropsychologia, 47, 1573–1577.

Keating, P. (2006). Phonetic encoding of prosodic structure. In: J. Harrington, &M. Tabain (Eds.), Speech production: Models, phonetic processes, and techniques(pp. 167–186). New York and Hove: Macquarie Monographs in CognitiveScience, Psychology Press.

Nakatani, L.H., & Aston, C.H. (1978). Perceiving the stress pattern of words insentences. Journal of the Acoustical Society of America 63, Suppl. 1, S55.

Napier, J. (1614) Mirifici logarithmorum canonis descriptio.Nooteboom, S. G. (1972). Production and perception of vowel duration: A study

of durational properties in Dutch. Unpublished Ph.D. Thesis, University ofUtrecht.

R Development Core Team (2009). R: A language and environment for statisticalcomputing. R Foundation for Statistical Computing, Vienna, Austria. URL/http://www.R-project.orgS.

Sluijter, A. M. (1995). Phonetic correlates of stress and accent, Ph.D. Dissertation,Leiden.

Sluijter, A. M. C., & van Heuven, V. J. (1995). Effects of focus distribution, pitchaccent and lexical stress on the temporal organization of syllables in Dutch.Phonetica, 52, 71–89.

Suomi, K. (2007). On the tonal and temporal domains of accent in Finnish. Journalof Phonetics, 35(1), 40–55.

Turk, A., & Sawusch, J. R. (1995). The domain of the durational effects of accent.Speech communication group working papers, 10, 42–71.

Turk, A. E., & Sawusch, J. R. (1997). The domain of accentual lengthening inAmerican English. Journal of Phonetics, 25, 25–41.

S. Dimitrova, A. Turk / Journal of Phonetics 40 (2012) 403–418418

Turk, A. E., & White, L. (1999). Structural influences on accentual lengthening inEnglish. Journal of Phonetics, 27, 171–206.

Turk, A., Nakai, S., & Sugahara, M. (2006). Acoustic segment durations in prosodicresearch: A practical guide. In: S. Sudhoff, D. Lenertova, R. Meyer, S. Pappert,P. Augurzky, I. Mleinek, N. Richter, & J. Schließer (Eds.), Methods in empiricalprosody research (pp. 1–28). Berlin, New York: De Gruyter. (¼Language,Context, and Cognition, 3).

Turk, A., & Shattuck-Hufnagel, S. (2007). Phrase-final lengthening in AmericanEnglish. Journal of Phonetics, 35(4), 445–472.

Turk, A., & Dimitrova, S. (2007). English phrasal stress targets multiple, optionallengthening sites. In Proceedings of the XVIth ICPhS, Saarbruecken, 6–10 August2007, pp. 1177–1180.

Van der Wel, R. P. R. D., Fleckenstein, R. M., Jax, S. A., & Rosenbaum, D. A. (2007).Hand path priming in manual obstacle avoidance: Evidence for abstract

spatiotemporal forms in human motor control. Journal of Experimental

Psychology: Human Perception and Performance, 33(5), 1117–1126.van Santen, J. P. H. (1994). Assignment of segmental duration in text-to-speech

synthesis. Computer Speech and Language, 8, 95–128.White, L. S. (2002). English speech timing: A domain and locus approach. Ph.D.

Dissertation, University of Edinburgh.White, L., & Turk, A. (2010). English words on the Procrustean bed: Polysyllabic

shortening reconsidered. Journal of Phonetics, 38, 459–471.Wightman, C. W., Shattuck-Hufnagel, S., Ostendorf, M., & Price, P. J. (1992).

Segmental durations in the vicinity of prosodic phrase boundaries. Journal of

the Acoustical Society of America, 91(3), 1707–1717.Wooldridge, J. M. (2009). Introductory econometrics: A modern approach (4th ed.).

Cengage Learning.