PDF hosted at the Radboud Repository of the Radboud University Nijmegen

Transcription

1 PDF hoted at the Radboud Repoitory of the Radboud Univerity Nijmegen The folloing full text i a publiher' verion. For additional information about thi publication click thi link. Pleae be advied that thi information a generated on and may be ubject to change.

2 Alignment of the Second Lo Target in Dutch Falling-Riing Pitch Contour Jörg Peter, Judith Hanen, Carlo Guenhoven Centre for Language Studie, Radboud Univerity Nijmegen, The Netherland Abtract To production experiment ere conducted to etablih the anchor point for the beginning of the final rie in Dutch falling-riing pitch contour. We ytematically varied the proodic tructure of the pot-nuclear ord by including the tre level (primary or econdary) of the penultimate yllable and the ditance of the lat treed yllable to the utterance end a factor. None of the yllable type provided an anchor point for the timing of the beginning of the rie, hich appeared to be mot contant relative to the utterance end or to the end of the rie. Our finding i not conitent ith earlier experiment hich found a tendency for the beginning of the rie to be attracted to the lat treed yllable. Additionally, e found that unaccented primary treed yllable are omehat longer than un-accented econdary treed yllable confirming earlier finding for Dutch obtained on the bai of reiterant peech. Index Term: intonation, pitch timing, tonal alignment, phrae accent, Dutch 1. Introduction Phonetic tudie have provided evidence that turning point in F0 contour are aligned relative to anchor point in the egmental tring ([1], [2], [3], [4]). A contour hich ha recently attracted ome attention i the nuclear fall-rie, hich conit of a falling pitch movement from the nuclear yllable and a riing movement at the end of the utterance ([5], [6]). The contour ha been expected to provide evidence for the phrae accent, introduced by [7] and further developed by [8], [9]. For [9], phrae accent are tone hich come ith the final boundary of ome larger phrae and have a econdary aociation to a treed or ome other deignated yllable. The lo turning-point in Englih and Dutch fall-rie ould be expected to behave in thi ay, a reflected in the repreentation of the contour a H*L-H% in [9]. A tudy by [5] of Dutch fall-rie in quetion in hich the tre pattern of the final nuclear ord a varied (S, S, S, and S, here S i the nuclear yllable and a pot-nuclear treed yllable) did not provide direct evidence for a cloe link beteen the lo turning-point of the fall-rie and the pot-nuclear tre, but the variation a apparently in conflict ith alternative anchor point. In particular, they found a variable ditance of the firt turning-point (L1) to the nuclear peak and of the econd turning-point (L2) to the voel onet of the lat yllable a ell a to the utterance end, and imilar variation a found for the ingle F0 minimum beteen the nuclear fall and the final rie (L). Earlier report on the timing of the final rie include t Hart [10] informal tatement that it come very late in the lat yllable. Hi obervation failed to ditinguih beteen yllable containing cha, hich in [5] clearly fail to behave a he predict, and other yllable. Yet, it ugget that L2 may occur at a fixed ditance from the end of the utterance, or from ome point ithin the lat (non-cha) yllable. An indication that the end of the utterance determine the location of L2 i provided by a tudy that examined the effect of econd occurrence focu (SOF) [11]. Sentence like Maar je hebt toch al een mooi lam? But don t you already have a beautiful lamb? poken in repone to Ik il graag een mooi lam I ould like to have a beautiful lamb contrated ith Maar je hebt toch al een mooi lam? (SOF underlined, capital for nuclear accent). Varying the tre pattern of the final and pre-final pot-nuclear ord, [11] found that L2 i tably aligned ith reference to the end of the utterance, ithout the lat treed yllable or the SOF-bearing yllable having an effect. The quetion arie hy L2 occurred at a variable ditance to the end of the utterance in [5], but not in [11]. One factor hich might have affected the alignment of L1, L and L2 in [5] i the variation in the ditance beteen the nuclear yllable and the utterance end, combined ith the proximity of thee to event. Their tet entence contained a maximum of three and a minimum of to pot-nuclear yllable, and L2, if realied at all a a eparate event from L1, may have been affected by the timing of the preceding turningpoint. In [11], the nuclear yllable a further removed from the utterance end, and peaker ere thu under no time preure hen realizing to eparate lo turning point. While the pot-nuclear treed yllable in [5] had econdary ord tre, thoe in [11] ere deaccented ord, and thu repreented primary treed yllable. If L2 i attracted to pot-nuclear tre, another quetion i therefore hether it alignment i affected by the level of tre. If a place name like Oudeater, ued by [5] in nuclear poition to repreent the tre pattern S, ere to be ued in potnuclear poition, hould e then expect the rie to be aligned relative to the lat tre (a) or the lat primary tre (Ou)? Accordingly, e intended to addre to quetion. Firt, to hat extent i L2 in Dutch determined by the end of the utterance and/or by the location of the lat (treed) yllable? And econd, hat i the role played by primary ord tre? If there i no effect of tre level, are pot-nuclear primary and econdary tree ditinguihed at all? To aner thee quetion, e carried out to production experiment. We firt had peaker produce fall-rie on utterance in hich the ditance from the lat tre to the end of the utterance a varied (,, and ). We ued entence ith longer pot-nuclear tretche, alloing our peaker to realie the econd lo turning-point (L2) of the lo plateau independently of the firt one. More particularly, e attempted to aner the folloing quetion, inpired by [5]: (i) Doe L2 align relative to the lat pot-nuclear tre? (ii) Doe L2 align relative to the final yllable? (iii) Doe L2 align relative to the end of the utterance (or the end of the final rie, hich ill occur cloe to the end of the utterance)? In the econd experiment, e examined hether the order of primary and econdary tre in final pot-nuclear ord affect the timing of L2, and, if not, ho primary and econdary tre are ditinguihed in thi poition.

3 2.1. Method Material 2. Experiment 1 Carrier entence ere declarative quetion containing the modal adverb toch ( after all ), hich typically trigger a falling-riing melody. We varied the ditance from the lat tre to the end of the utterance by chooing three et of target ord ith the tre pattern,, and, a hon in Table 1. All target ord occurred in pot-nuclear final poition and ere eparated form the nuclear yllable by four yllable to prevent time preure effect originating from a firt lo target. We alo tried to keep the rhythmical pattern of the intervening yllable contant, but ord boundarie ere not alay in the ame poition. Table 1. Target ord ued in Experiment 1. S Mol Molber Molberen Lam Lamber Lamberen Wel Welder Welderen Bul [Y] Bulder [Y] Bulderen [Y] Doen [u] Doender [u] Doenderen [u] The carrier entence ere preceded by a context entence, hich a read by a econd peaker. (1) illutrate a context entence and a carrier entence repreenting the third tre pattern (target ord underlined). (1) Ik moet nog chaaten kopen voor al e met tante Molder naar de ijbaan gaan. Jullie gingen toch zemmen met tante Molder? S I mut buy kate before going the ice rink ith aunt Molder. You ere going imming ith aunt Molder, right? Procedure The mini-dialogue of the type given in (1) ere preented on card to ubject in peudo-randomied order, hich a revered for half of the ubject. They ere interpered ith tet entence from experiment 2 a ell a ith 77 filler entence, hich ere equally preented a part of minidialogue. Recording ere made in a ound-treated booth at the Radboud Univerity Nijmegen ith a portable DAT recorder (TASCAM DA-P1) and digitally tranferred to a computer Subject We recorded ix male and eleven female peaker, aged beteen 18 and 30. Their regional background varied, but on the bai of the econd author auditory judgment, all ubject poke Standard Dutch. The data from to female peaker ere excluded, a a number of target ord had irregular pitch pattern involving harmonic break Acoutical analyi Segment boundarie ere placed at negative-to-poitive zerocroing during viual inpection of the ave form and a broadband pectrogram. The end of the final rie (H2) a identified at the F0 maximum near the final IP boundary. The poition of L2 a emi-automatically determined by adapting to croing regreion line to the F0 trace uing the elbo cript by Mary Beckman and Pauline Welby [12]. They ere applied to a 150 m interval that had a likely location of L2 at it centre, a identified by viual inpection of the F0 trace. In 31% of the utterance, L2 had to be fully determined by viual inpection, a microproodic F0 variation ithin the analyi indo made the output of the regreion analyi unreliable. 7.5% of the utterance ere excluded from analyi becaue of unexpected choice of phraing, contour, or accent location. All meaurement ere done by the firt author. To etimate the reliability of our meaurement method, the data of one male and one female peaker ere independently proceed by the econd author. The meaurement proved to be highly reliable: mean difference beteen meaurement acro both experiment ere 1.85 m for the ditance from L2 to the end of the IP; 0.26 m for that from L2 to H2; 2.66 m for that from L2 to the voel onet of the lat yllable; 0.3 m for that from L2 to the voel onet of the lat yllable; 1.59 m for that from H2 to the end of the IP; and 2.81 m for the poition of the voel onet of the lat treed yllable. The mean difference beteen the meaurement of F0 ere 0.69 Hz for L2 and 1.68 Hz for H2. All meaurement ere done ith the help of Praat [13]. The elbo cript ere carried out ith Praat and R [14] Reult Figure 1 ho mean yllable duration for,, and and mean alignment of L2 and H Duration (m) Figure 1: Mean yllable duration, poition of L2 (left arro), and poition of H2 (right arro). White boxe: onorant rhyme of trong () and eak () yllable. Grey boxe: onet. Final IP boundary et to 0 m. L2 appear to lie at a fairly contant ditance from both the end of the utterance and H2. We report the reult of four one-ay repeated-meaure ANOVA. Firt, e examined the null hypothei that L2 i contantly aligned ith the lat tre of the utterance. The ditance from L2 to the voel onet in the lat pot-nuclear treed yllable a the independent variable, and STRESSPATTERN (,, ) a fixed ithin-ubject factor (Table 2-1). We found a ignificant effect of STRESSPATTERN, F (2, 130) = , p <.001. Contrat revealed that the ditance in ord a longer than in ord and in ord longer than in ord (p <.001). An alternative analyi uing the centre of the onorant rhyme of the treed yllable a a point of reference provided comparable reult. We may thu reject the null hypothei and conclude that the poition of the lat treed yllable ha no effect on the timing of L2 in the three potnuclear tre pattern.

4 Table 2. Mean ditance from L2 to (1) voel onet of lat pot-nuclear tre, (2) voel onet of lat yllable, (3) end of utterance, and (4) end of final rie (H2). Stre Pattern N Mean (m) SD In a econd analyi, e examined the null hypothei that L2 i contantly aligned ith the lat yllable of the utterance. The ditance from L2 to the voel onet in the lat potnuclear yllable a the independent variable, and STRESSPATTERN a fixed ithin-ubject factor (Table 2-2). Mauchly tet indicated that the aumption of phericity had been violated ( 2 (2) = 9.70, p <.01), and degree of freedom ere corrected uing Greenhoue-Geier etimate of phericity ( =.88). STRESSPATTERN had a ignificant effect on the ditance from L2 to the voel onet in the lat yllable, F(1.75, ) = , p <.001. Contrat revealed that the ditance in ord differed ignificantly from the ditance in and ord (p <.001), hile the ditance in and ord did not differ ignificantly. We obtained comparable reult hen uing the beginning of the lat yllable or the centre of it onorant rhyme a point of reference. Thee finding ugget that L2 doe not conitently align relative to the lat yllable. In a third analyi, e examined the hypothei that L2 i contantly aligned ith the end of the utterance. The ditance from L2 to the end of the onorant rhyme of the final yllable a the independent variable, and STRESSPATTERN a fixed ithin-ubject factor (Table 2-3). We found no ignificant effect of STRESSPATTERN, F (2,130) =.35, p =.709. Since H2 did not alay coincide ith the end of the onorant rhyme of the final yllable, e carried out an alternative analyi uing the ditance from L2 to H2 a dependent variable (Table 2-4). Again, e found no ignificant effect of STRESSPATTERN, F (2, 130) =.14, p =.873). Thee finding ugget that L2 align neither relative to the lat tre nor to the lat yllable, leaving open the poibility that it align at a fixed ditance to either the end of the utterance or the end of the final rie. The data trongly ugget that thi concluion i correct. Mean ditance from L2 to the voel onet of lat pot-nuclear tre vary from 58 m to 243 m, and thoe beteen L2 and the voel onet of the lat yllable vary from 58 to -17 m. By contrat, mean ditance from L2 to the utterance end or thoe from L2 to H2 differ by maximally 4 m. 3. Experiment 2 A L2 i not bound to the lat tre, it cannot be ued a a cue to the poition of the lat tre. Would thi independence perit if tre i ditinctive, a in Hogelonen (primary tre on ho-, econdary tre on -lo-; proper name) v. hoge lonen (econdary tre on ho-, primary tre on -lo-; high alarie )? The experiment in [5] did not allo for the detection of any influence of tre level on the timing of L2, a all primary tree occurred in nuclear poition, thu confounding any effect of primary ord tre and focal accent Method We ued the ame peaker and the ame procedure of data elicitation and analyi a reported for Experiment 1. The tet entence contained (near-)minimal pair of ord varying in the poition of primary and econdary tre. The firt half of the target ord conited of compound, the econd of a hort noun phrae (adjective-noun). Compound are pronounced ith primary tre on the firt element in Dutch, herea in adjective-noun combination primary tre fall on the econd element. Accordingly, e ued the pair of compound and phrae hon in Table 3 in three pair of tet entence in pot-nuclear final poition. They ere eparated by three yllable from the nuclear yllable. In the example in (2), target ord are underlined and the nuclear yllable i in bold type. (2) a. Maar jullie gingen toch naar het trand met meeter Hogelonen? ' But you ent to the beach ith Mr. Hogelonen, didn t you? ' ' b. Maar ze heffen toch ook veel meer belating op die hoge lonen? ' But they apply much higher taxe to higher alarie, don t they? Table 3. Target phrae ued in Experiment 2. ('=primary, =econdary, =untreed) ' Hogelonen (proper name) Oudebomen (proper name) Grotehuizen (proper name) 3.2. Reult ' ' hoge lonen high alarie oude bomen old tree grote huizen large houe Figure 2 ho mean yllable duration for the tre pattern ' and ' in pot-nuclear final poition and the timing of the beginning and the end of the final rie. ' Duration (m) Figure 2: Mean yllable duration, poition of L2 (left arro), and poition of H2 (right arro). White boxe mark onorant rhyme of the firt, third, and fourth yllable. The final rie tart near the beginning of the lat yllable in both tre pattern. A dependent-t tet (to-tailed) ho that the ditance of L2 from the beginning of the primary tre i ignificant (t = , p <.001) (Table 4). We may

5 thu reject the null hypothei that the timing of L2 depend on the poition of the primary tre in the tre pattern teted. Mean ditance from the voel onet to L2 differ by no le than 252 m. We may alo conclude that the pair of target ord are not ditinguihable by the timing of the final rie. Table 4. Mean ditance from L2 to voel onet of primary tre. Stre Pattern N Mean (m) SD ' ' In order to etablih any other phonetic difference beteen unaccented primary and econdary tre, e compared duration and F0 ratio beteen the firt and econd treed yllable in the to tre pattern. They ho that (i) the econd treed yllable i longer than the firt one in both tre pattern (ratio are 1.22 for ' and 1.38 for ', N = 35); (ii) the econd yllable i lengthened more hen bearing primary tre (t = -3.05, p < 0.01); (iii) the econd treed yllable i loer than the firt one in both tre pattern and loered more in ' ord; but thi difference did not reach tatitical ignificance (t = 1.88, p = 0.068). To ummarize, the ' ord differ from the ' ord in pot-nuclear final poition by an additional amount of penultimate yllable lengthening. There ould appear to be a trend to loer the penultimate yllable more in ' ord than in ' ord. 4. Dicuion The reult of Experiment 1 and 2 ugget that the beginning of the final rie in Dutch (L2) i not aligned ith reference to the lat primary tre, the lat tre, or the lat yllable, but i located at a highly table ditance from the end of the utterance and the end of the final rie. For thee ditance, e found a maximum difference of 4 m beteen the mean poition in ord and the mean poition in and ord. Thee reult are in line ith the finding in [11]. They appear to be in conflict ith the finding of [5]. A poible explanation for the apparent tendency of L2 to occur near treed yllable in that tudy may lie in the hort ditance beteen the nuclear yllable and the end of the utterance. In our tet entence, there ere minimally four yllable beteen the nuclear yllable and the lat yllable, hile in [5] there a a minimum of one. A a conequence, there a little opportunity to realie L2 outide a treed yllable. In many cae, peaker did not realie to eparate lo pitch target, uch that L1 coincided ith L2. In our experiment, a in the experiment reported by [11], the effect of the timing of L1 on the timing of L2 a greatly reduced. Indeed, the longer pot-nuclear tretche ued in [6] alloed L2 conitently to fall after the lat treed yllable in comparable contour in American Englih. Experiment 2 a conducted to ee if a durational or F0 difference beteen unaccented primary and unaccented econdary treed yllable exit in natural, read peech. It a found that primary treed yllable ere longer than econdary treed yllable in penultimate poition in the pot-nuclear tretch. Thi confirm reult by [15] ith reiterant peech. The alignment of L2 in Dutch i thu very different form the alignment of the lat high target in Roermond Dutch quetion, hich i contant relative to the lat treed yllable, but variable relative to the utterance end [16]. 5. Acknoledgement We thank Joop Kerkhoff for hi aitance ith the cript and our peaker for participating in the experiment. A part of the project Intonation in Varietie of Dutch, thi reearch i upported by the Dutch Science Foundation (NWO). 6. Reference [1] Arvaniti, A., Ladd, D. R., and Mennen, I. Stability of tonal alignment: The cae of Greek prenuclear accent, J. of Phonetic, Vol. 26, pp. 3 25, [2] Ladd, D. R., Faulkner, D., Faulkner, H., and Schepman, A. Contant egmental anchoring of F0 movement under change in peech rate, J. Acout. Soc. Amer., Vol. 106, pp , [3] Ladd, D. R., Mennen, I., and Schepman, A. Phonological conditioning of peak alignment in riing pitch accent in Dutch, J. Acout. Soc. Amer., Vol. 107, pp , [4] Atterer, M. and Ladd, D. R. On the phonetic and phonology of egmental anchoring of F0: Evidence from German, J. of Phonetic, Vol. 32, pp , [5] Lickley, R. J., Schepman, A., and Ladd, D. R., Alignment of phrae accent lo in Dutch falling riing quetion: Theoretical and methodological implication, Language and Speech, Vol. 8, pp , [6] Barne, J., Shattuck-Hufnagel, S., Brugo, A., and Veilleux, N. The domain of realization of the L- phrae tone in American Englih, in Proc. 3rd Int. Conf. on Speech Proody, Hoffmann, R. and Mixdorff, H., Ed. Dreden: TUD Pre, [7] Pierrehumbert, J. B. The phonetic and phonology of Englih intonation. Ph.D. diertation, MIT, Ditributed by Indiana Univ. Linguitic Club, [8] Beckman, M. E. and Pierrehumbert, J. B. Intonational tructure in Englih and Japanee, Phonology Yearbook, Vol. 3, pp , [9] Grice, M., Ladd, D. R., and Arvaniti, A. On the place of phrae accent in intonational phonology, Phonology, Vol. 17, pp , [10] t Hart, J. Intonation in Dutch, in Intonation ytem: A urvey of tenty language. Hirt, D. and Di Crito, A., Ed. Cambridge: Cambridge Univerity Pre, pp , [11] van de Ven, M. The alignment of lo target in Dutch fall-rie quetion. MA thei, Dept. Englih, Radboud Univ. Nijmegen, [12] Beckman, M. E and Welby, P. (202). Elbo cript. Conulted at praat.html [13] Boerma, P. and Weenink, D. Praat: Doing phonetic by computer (ver ), Comp. program, [14] R (verion 2.4.1). A Language and Environment for Statitical Analyi. The R Foundation for Statitical Computing [15] Guenhoven, C. The boundary tone are coming: On the non-peripheral realization of boundary tone, in Paper in Laboratory Phonology V. Broe, M.R., and Pierrehumbert, J.B., Ed. Cambridge: Cambridge Univerity Pre, pp , [16] Rietveld, T., Kerkhoff, J., and Guenhoven, C. Word proodic tructure and voel duration in Dutch., J. of Phonetic, Vol. 32, pp , 2004