EFFECT OF UTTERANCE LENGTH ON F0 SCALING
Maria del Mar Vanrell Bosch
Universitat Autònoma de Barcelona [email protected]
ABSTRACT utterance on scaling of nuclear accents, the research suggests that speakers can make a This study examines the effect of utterance length meaningful use of the presence or absence of on utterance-initial F0 values and H and L scaling downstep. The present study tests whether of nuclear accent in Majorcan Catalan. Research utterance length affects the scaling of utterance- on the correlation between utterance length and initial F0 values as well as H leading tone and L initial F0 values has thus far yielded contradictory tone F0 values. answers to the question of whether utterance length is a determining factor for initial pitch height. Regarding the impact of utterance length on 2. METHODS scaling of nuclear accents (known as downstep), it Yes-no questions and what-questions in has been shown that downstep may be under the Majorcan Catalan have been described as conscious control of the speaker and be governed containing a difference in pitch height in the by a clearly communicative function. Firstly, the pretonic nuclear syllable. Thus, although both results reveal that there exists a correlation types of questions are characterized by a falling between sentence length and initial pitch height, nuclear accent H+L*, that is, an H leading tone but this correlation is not constant across speakers aligned with the pretonic nuclear syllable and a L* and sentence-types, suggesting that this is an tone associated with the last stressed syllable, in instance of soft preplanning. Secondly, our results the case of yes-no questions, the H leading tone show that downstep or, more precisely, the failure has been shown to be systematically realized with of downstep may be grammaticalized in a a higher tone than that found in what-questions particular phonological context. ([15]). Moreover, yes-no questions are headed by Keywords: utterance-length, F0 scaling, the unaccented interrogative particle que (‘that’) preplanning, downstep. while what-questions are headed by the accented what-particle què (‘what’). Previous work ([5]) has 1. INTRODUCTION also reported that analogous prosodic features are able to disambiguate between different question The effect of utterance length on the scaling of types. utterance-initial F0 peaks is a controversial issue in intonational studies. A correlation between utterance-initial F0 peaks and length of utterance 2.1. Material and experimental procedure would mean that speakers are able to plan F0 The data used in this study consisted of yes-no contours at a phrase level by beginning higher questions and what-questions of lengths ranging when the utterance is longer and lower when the from 1 to 3 pitch accents and divided into blocks utterance is shorter (what is known as hard according to the number of pitch accents in each preplanning or the global hypothesis). utterance. Each block was made up of paired yes- Alternatively, speakers may proceed at a more no and what-questions that were identical or very local level, that is, accent-by-accent (i.e. the local similar at the segmental level (see Table I). The hypothesis). Work in this field has produced last word always contained the stress on the conflicting results. Thus, some findings have penultimate syllable to provide more room for reported that speakers begin higher or lower nuclear accent and boundary tone realization. depending on utterance length ([3], [4], [10], [11]), Table 1: Examples of the utterances used in this while other researchers have reported no utterance experiment. Stressed syllables are marked in boldface. length effect on peak values ([1], [2], [6], [7], [8], [12], [13], [14]). As regards the impact of length of Yes-no questions What-questions Que en va vendre? Què en va vendre? (‘Did s/he sell any of it?’) (‘What did s/he sell of it?’) Speaker FF Speaker MC 220,0 Number of 270,0 Number of pitch pitch 260,0 [k va vnd] [k va vnd] accents accents 195 1 200,0 1 250,0 Que n’esperava vendre? Què n’esperava vendre? 186 2 2 3 240,0 3 (‘Did s/he expect to sell any (‘What did s/he expect to sell 180,0 230,0 of it?’) of it?’) 238 187 220,0 160,0 197 198 210,0 [k nspeav vnd] [k nspeav vnd] 310
200,0 140,0 v0 (Hz) Frequency Fundamental Que hi volia tornar a Què hi volia tornar a (Hz) v0 Frequency Fundamental vendre? (‘Did s/he want to vendre? (‘What did s/he Yes-no questions What-questions Yes-no questions What questions Sentence type Sentence type sell there again?’) want to sell there again?’) Speaker MV [kj vuli tona vnd] [kj vuli tona vnd] 408 Number of pitch 260,0 accents Five native speakers of Majorcan Catalan (two 1 2 240,0 females and three males) read three repetitions of 3 the three blocks of utterances. Thus a total of 450 220,0 utterances (5 sentences x 2 types x 3 lengths x 3 200,0 361 repetitions x 5 speakers) were obtained. In order to (Hz) v0 Frequency Fundamental Yes-no questions What-questions facilitate interpretation of the utterances by the Sentence type speakers, each test question was answered verbally Results of Friedman tests (Table 2) revealed that by the author of this study. speakers AB, AC and MC showed a significant Pitch extraction and measurements were tendency in what-questions to anticipate the length performed using Praat software. As the aim of this of the utterance by beginning higher if the work was to examine the effect of utterance length utterance was longer (presumably to avoid the on utterance-initial F0 values and H and L scaling pitch falling much lower) and lower if the of nuclear accent, no measurements related to utterance was shorter. This anticipation also prenuclear pitch accents were taken. Non- occurred in yes-no questions, though for only two parametric statistical tests were used to test for speakers (AB and MC). differences in F0 values. Table 2: Results of Friedman tests with significance level set at 0.05. 3. RESULTS Subject Yes-no questions What-questions AB χ2(2)=12.4 χ2(2)=6.40 3.1. Utterance-initial F0 values. AC χ2(2)=1.00, ns χ2(2)=8.22 2 2 Figure 1 shows the box plots of F0 for the vowel FF χ (2)=3.60, ns χ (2)=1.85, ns MC χ2(2)=12.13 χ2(2)=12.13 [´] of the particle que/què in yes-no questions (left- MV χ2(2)=.93, ns χ2(2)=.40, ns hand boxes) and what-questions (right-hand boxes) for each of the five speakers, broken down into 3.2. H leading tone F0 values different utterance lengths (1-3 pitch accents). Figure 2 shows the box plots of F0 at the start of Figure 1: Box plots of F0 for the vowel [´] of the the fall within the falling nuclear accent H+L* in interrogative particles que (yes-no questions) and què utterances containing from 1 to 3 pitch accents for (what-questions) in utterances of different lengths for yes-no questions (left-hand boxes) and what- each speaker. questions (right-hand boxes). Speaker AB Speaker AC 200,0 Number of 250,0 Number of 85 pitch pitch Figure 2: Box plots of H-F0 within the nuclear accent 190,0 accents 225,0 accents 71 1 1 H+L* for yes-no questions and what-questions in 180,0 150 2 200,0 2 utterances of different lengths for each speaker. 170,0 3 3
2 175,0 160,0 75 Speaker AB Speaker AC 150,0 250,0 220,0 150,0 Number of Number of pitch pitch 125,0 103 140,0 92 225,0 accents 200,0 accents
8 24 1 1 Fundamental Frequency v0 (Hz) v0 Frequency Fundamental Fundamental Frequency v0 (Hz) v0 Frequency Fundamental 130,0 100,0 200,0 2 180,0 2 21 40 3 3 Yes-no questions What-questions Yes-no questions What-questions 37 175,0 160,0 Sentence type Sentence type 60 79 126 150,0 140,0 86
125,0 120,0 33
Fundamental Frequency H (Hz) Frequency Fundamental 100,0 (Hz) H Frequency Fundamental 100,0
Yes-no questions What-questions Yes-no questions What-questions Sentence type Sentence type Speaker FF Speaker MC Speaker FF Speaker MC 220,0 Number of 300,0 Number of 150,00 Number of 220,00 Number of pitch pitch pitch pitch accents 280,0 accents accents accents 314 200,0 1 1 140,00 1 200,00 1 314 2 260,0 2 2 2
3 329 3 3 3 249 180,0 240,0 130,00 180,00
302 236 220,0 325 302 160,0 120,00 160,00 200,0 335
243 Fundamental Frequency L (Hz) L Frequency Fundamental (Hz) L Frequency Fundamental Fundamental Frequency H (Hz) Frequency Fundamental Fundamental Frequency H (Hz) H Frequency Fundamental 140,0 180,0 110,00 140,00
Yes-no questions What-questions Yes-no questions What-questions Yes-no questions What-questions Yes-no questions What-questions Sentence type Sentence type Sentence type Sentence type Speaker MV Speaker MV 320,0 180,00 Number of 391 Number of pitch pitch 300,0 accents 378 accents 1 1 280,0 170,00 2 384 2
260,0 3 3
240,0 160,00
220,0 376 400
415 380 200,0 150,00 Fundamental Frequency L (Hz) Frequency Fundamental Fundamental Frequency H (Hz) H Frequency Fundamental 180,0
Yes-no questions What-questions Yes-no questions What-questions Sentence type Sentence type
A constant utterance-length effect is not present Note that although there is evidence that in the L-F0 value. Nevertheless, it seems that utterance length affects the H pitch height value, utterance length affects the L-F0 of what-questions this utterance length effect is greater for what- more that that of yes-no questions: the longer the questions (right-hand boxes) than for yes-no utterance, the lower the L-F0—with the exception questions (left-hand boxes), since it is statistically of speaker MC, who does not follow this pattern. significant for all speakers in what-questions but Statistical analysis (Table 4) confirms that the only for one speaker (FF) in yes-no questions utterance length effect on the L-F0 value is (Table 3). significant only for speaker AC in yes-no questions Table 3: Results of Friedman tests with significance and for speakers AB, MC and MV in what- level set at 0.05. questions. Subject Yes-no questions What-questions Table 4: Results of Friedman tests with significance 2 2 AB χ (2)=5.20, ns χ (2)=28.13 level set at 0.05. AC χ2(2)=2.71, ns χ2(2)=10.66 FF χ2(2)=16.93 χ2(2)=23.28 Subject Yes-no questions What-questions 2 2 MC χ2(2)=4.80, ns χ2(2)=6.53 AB χ (2)=4.13, ns χ (2)=23.33 2 2 MV χ2(2)=5.73, ns χ2(2)=19.60 AC χ (2)=8.14 χ (2)=1.55, ns FF χ2(2)=3.60, ns χ2(2)=2.71, ns 3.3. L tone F0 values MC χ2(2)=1.73, ns χ2(2)=6.93 MV χ2(2)=1.20, ns χ2(2)=10.80 Figure 3 shows the box plots of F0 for the end of the fall within the nuclear accent H+L* in 4. DISCUSSION utterances of different lengths (1 to 3 pitch accents) As regards utterance-initial F0 values, evidence for yes-no questions (left-hand boxes) and what- was found for significant anticipatory raising for questions (right-hand boxes). three of the five speakers in what-questions and for Figure 3: Box plots of L-F0 within the nuclear accent two of the five speakers in yes-no questions. Thus, H+L* for yes-no questions and what-questions in speakers may vary the height of the initial F0 utterances of different lengths for each speaker. values in what-questions and yes-no questions Speaker AB Speaker AC Number of Number of depending on the utterance length as measured by pitch pitch 150,00 110,00 accents accents the number of pitch accents in the utterance. In 1 1 2 153 2 140,00 100,00 3 3 other words, our results are consistent with those previous findings which report that speakers begin 130,00 90,00 15 higher or lower (with a higher or lower pitch) 120,00 80,00
Fundamental Frequency L (Hz) L Frequency Fundamental (Hz) L Frequency Fundamental depending on utterance length ([3], [4], [10], [11]). Yes-no questions What-questions Yes-no questions What-questions By contrast, other researchers have reported no Sentence type Sentence type utterance length effect on F0 values ([1], [2], [6], [7], [8], [12], [13], [14]). However, the fact that the interaction between utterance length and initial F0 values is not statistically significant for all 6. REFERENCES speakers suggests that the observed global [1] Connell, B. 2003. Pitch realization and the four tones of preplanning must be interpreted as soft Mambila. In: Kaji, S. (ed), Cross-Linguistic Studies of preplanning, that is, a strategy that speakers may Tonal Phenomena. Tokyo: Research Institute for the opt to use but that is not essential as a production Languages and Cultures of Asia and Africa. mechanism ([6], [9]). [2] Connell, B. 2004. Tone, Utterance Length and F0 Scaling. In: International Symposium on Tonal Aspects of The present results also confirm the interaction Languages: With Emphasis on Tone Languages. Beijing, between utterance length and the H value of the China, March 28-31. nuclear accent, with the F0-H value falling much [3] Cooper, W.J., Sorensen, J. 1981. Fundamental lower as the number of pitch accents increases. Frequency in Sentence Production. Heidelberg: Springer. [4] 'Hart, J. 't 1979. Explorations in automatic stylization of However, though this interaction between length F0 curves, IPO Annual Progress Report 14, 61-65. and H is high in what-questions, it is nearly non- [5] Jun, S.A., Mira, O. H. 1996. A Prosodic Analysis of existent in yes-no questions. Considering the Three Types of Wh-Phrases in Korean, Language and difference in the height of the leading tone that Speech 39 (1), 37-61. characterizes yes-no questions as opposed to what- [6] Liberman, M., Pierrehumbert, J. 1984. Intonational Invariance under Changes in Pitch Range and Length. In: questions in Majorcan Catalan (described in [15]), Aronoff, M., Oehrle, R.T. (eds), Language Sound this sort of blocking of the downtrend tendency in Structure. Cambridge: MIT Press, 157-234. yes-no questions could be interpreted as contrast [7] Lindau, M. 1986. Testing a model of intonation in a tone enhancement between yes-no questions and what- language, J. Acoust. Soc. Am. 80, 757-764. [8] Prieto, P., Shih, C., Nibert, H. 1996. Pitch downtrend in questions in this language. This strategy would Spanish, Journal of Phonetics 24, 445-473. facilitate the perception of the L tone in yes-no [9] Prieto, P., D'Imperio, M., Elordieta, G., Frota, S., & questions, since otherwise the L tone might be Vigário, M. 2006. Evidence for soft preplanning in tonal misinterpreted from a downstepped or plain H. production: Initial scaling in Romance. In: Hoffmann, R., By contrast, the L point is less affected by Mixdorff, H. (eds), Proceedings of Speech Prosody. Dresden: TUDpress Verlag der Wissenschaften GmbH, utterance length than the H point. The effect of 803-806. utterance length on the L-F0 value is still nearly [10] Rialland, A. 2001. Anticipatory raising in downstep null in yes-no questions and lower than that of the realization: Evidence for preplanning in tone production. H-F0 value in what-questions. In: Kaji, S. (ed), Proceedings of Symposium Cross- Linguistic Studies of Tonal Phenomena: Tonogenesis, 5. CONCLUSIONS Japanese Accentology, and Other Topics. Tokyo: University of Foreign Studies, 301-321. The results reported in this article confirm that [11] Snider, K. 1998. Tone and utterance length in Majorcan Catalan speakers plan the height of the Chumburung: An instrumental study. Paper presented at the 28th Colloquium on African Languages and initial F0 values in what-questions and in yes-no Linguistics, Leiden. questions, that is, they anticipate the length of the [12] Thorsen, N. 1982. Intonation patterns and stress group utterance by beginning higher when the utterance patterns in declarative sentences of varying length in is longer and lower if the utterance is shorter. ASC Danish, Annual Report of the Institute of Phonetics 14, 1-29. However, the fact that there is variation within [13] Toledo, G.A., Martínez Celdràn, E. 1997. speakers in the magnitude of the interaction Preplanificación psicolingüística y entonación en el between utterance length and the height of the español mediterráneo, Estudios de Fonética Experimental initial peak indicates that any global preplanning VIII, 185-206. that occurs in this instance must be interpreted as [14] van dern Berg, R., Gussenhoven, C., Rietveld, T. 1992. Downstep in Dutch: Implications for a model. In: soft preplanning ([6], [9]). Docherty, G., Ladd, D.R. (eds), Papers in Laboratory Utterance length is also shown to have an Phonology II: Gestures, Segment, Prosody. Cambridge: important effect on the H scaling of nuclear accent Cambridge University Press, 335-359. as long as this effect does not conflict with the [15] Vanrell, M.M. 2006.The phonological role of tonal scaling in Majorcan Catalan interrogatives. Unpublished effect that the sentence-type factor exerts on the Master's thesis, Universitat Autònoma de Barcelona pitch height of the leading tone H of the nuclear (http://seneca.uab.es/ggt/Tesis/MVanrell.pdf). accent H+L*. This failure of downstep could be interpreted as contrast enhancement between yes- no questions and what-questions in Majorcan Catalan.