web-sls 96.02 20.02.97

In this paper we examine whether the articulation of a consonant is affected by its prosodic position in French. Nasal and lingual articulations are observed for consonants placed in different prosodic positions. Five positions are considered in this study: Utterance initial, Intonational Phrase initial, Accentual Phrase initial, Word initial and Syllable initial positions. For the two speakers studied, prosodic position is found to influence the articulation of the consonants [t] and [n]. For [n], both lingual (as measured with EPG) and nasal (as measured with nasal flow) articulations vary depending on the position of the consonant, although the amount of linguopalatal contact seems to be affected more than the amount of nasal airflow. In general, the higher the prosodic domain the consonant is initial in, the less nasal flow [n] has and the more linguopalatal contact [n] and [t] have. These results confirm that prosody, and more particularly the position of a consonant in a prosodic domain, influences articulation. This effect appears to be a general mechanism of speech production as it affects two remote articulators, the tongue and the velum. However, the number of distinctions between prosodic positions cued by this articulatory variation may vary between speakers and articulators.

---

Dans cet article, nous examinons si l'articulation d'une consonne est influencée par sa position prosodique, en français. Nous avons observé les articulations linguale et nasale de consonnes placées à cinq positions prosodiques différentes : au début d'une Phrase, au début d'un Syntagme Intonatif, au début d'un Syntagme Accentuel, au début d'un Mot lexical, et au début d'une Syllabe. Pour les deux locuteurs étudiés, la position prosodique influence l'articulation des consonnes [t] et [n]. Pour [n], l'articulation linguale (mesurée avec l'EPG) et l'articulation nasale (mesurée indirectement avec le débit nasal) varient en fonction de la position de la consonne, et la quantité de contact linguopalatal semble être plus affectée que la quantité de débit nasal. En général, plus la position de la consonne est élevée dans la hiérarchie prosodique, plus le débit nasal de [n] est faible, et plus le contact linguopalatal est important pour [n] et [t]. Ces résultats confirment que la prosodie, et plus précisément la position d'une consonne dans un constituant prosodique, influence l'articulation. Cet effet pourrait être considéré comme un mécanisme global de la production de la parole car il affecte deux articulateurs distants : la langue et le vélum. Pourtant, le nombre de distinctions entre les différentes positions prosodiques signalées par cette variation articulatoire varie en fonction des locuteurs et des articulateurs.

1. INTRODUCTION

Fig.1 One of the possible phrasings of the sentence "La pauvre Tata, Nadia et Paul n'arriveront que demain." (Poor Auntie, Nadia and Paul won't arrive until tomorrow). Prosodic constituents are listed on the left side of the figure, from the highest to the lowest.

The best known effect of Prosodic Position, on articulation is final lengthening [Beckman, Edwards and Fletcher, 1992]. This lengthening occurs at the end of prosodic constituents like those shown in the tree of Figure 1, and is found to be hierarchical. Prosodic position, or strength of the juncture between constituents, has also been shown to influence the onset of coarticulation. In his cinefluographic study, MacClean [1973] showed that the strength of syntactic boundaries falling between the 2 vowels in a CVVN English sequence influenced the timing of velic movement. Major syntactic (prosodically marked) boundaries, such as sentence or clause boundaries, delayed the onset of velic movement for [n] relative to the preceding vowel. On the contrary, for prosodically unmarked boundaries, the velic opening was initiated close to the beginning of the articulatory movement of the first vowel. Similar results have been shown, although to a smaller extent, for tongue movement in a [kl] cluster [Hardcastle, 1985]. Prosodic Position has also been found to affect the articulatory properties of a segment compared to other segments in the speech chain (syntagmatic comparison), or in other sentences (paradigmatic comparison). For example, using the velotrace, Krakow [1989] examined the velic articulation of the consonant [m] in English as a function of its position in the Word and Syllable. She found that Word- or Syllable-initial consonants showed a higher velum position than Word- or Syllable-final consonants. Several studies have describe a weakening of articulations in Word- or Syllable-final positions compared to initial ones [among others, Benguerel, 1977; Fujimura, 1977; Vaissière, 1988; Byrd, 1994].

More recently, attention has been directed to the influence of position within higher prosodic domains, raising the question of whether particular changes in articulation may cue prosodic phrasing by marking prosodic boundaries. For example, in English [Pierrehumbert & Talkin, 1992] and in Korean [Jun, 1993], VOTs of aspirated stops have been found to be longer when the stop is initial in a large phrase (e.g. an Intonational Phrase), than if it is initial in a hierarchically lower prosodic domain. Most recently, Fougeron & Keating [1995, 1996] have shown that the strength of oral articulation depends on phrasal position. Measured with electropalatography (EPG), the amount of linguopalatal contact for [n] in English was larger in prosodic domain-initial positions (Word-, Minor Phrase- and Major Phrase-initial positions) compared to medial or final ones. This phenomenon was called "initial strengthening". Comparison of different prosodic levels showed that the degree of strengthening cued 2 or 3 levels of prosodic boundaries depending on the speaker.

In this study, our goal is to examine whether the effect of prosodic position on articulation occurs at the beginning of prosodic constituents, and whether this effect occurs at different levels in the prosodic hierarchy. We focus here on the spatial characteristics of the articulation of [n] and [t] in French depending on their prosodic position. Prosodic effects are examined for both lingual and nasal articulations for the nasal consonant (in 2 separate experiments), in order to evaluate whether prosodic conditioning of articulation may be a general phenomenon in speech or is restricted to some articulatory subsystem. EPG offers us a direct measurement of the amount of linguopalatal contact, and airflow recordings give us an indirect observation of the velopharyngeal aperture.

2. METHOD

The amount of nasal flow and of linguopalatal contact was observed for the nasal coronal consonant [n]. In the airflow experiment, the consonant was placed in 2 vocalic contexts which vary in oral impedance: small for [a_a] and strong for [i_i]. In the EPG experiment, only the open vowel [a] was used in order to minimize the occurrence of linguopalatal contact that is not directly related to consonant articulation. The amount of linguopalatal contact was also observed for the oral coronal consonant [t] in the context [õ_õ].

Table I: Stimuli designed for the sequence [ana]. Prosodic Positions are presented from highest to lowest in the examined prosodic hierarchy.

For the corpus with the consonant [n], nasal and oral flow were recorded directly into a Kay-CSL with a Rothenberg split mask. Variation in nasal airflow corresponds to variation in velum height as long as the total amount of airflow going to the nose is the same and the oral constriction remains the same. We controlled for variation in the overall airflow between the test sentences by looking for variation in the acoustic energy of the following oral vowel, but all were comparable. Therefore, we will assume throughout this study that variation in nasal flow corresponds to variation in velic articulation. In a separate session, recordings of linguopalatal contact were made with the Kay-Palatometer for [n] in the context [ana], and for [t]. The custom-made pseudopalate has 96 electrodes covering the entire hard palate and the inside surface of the molars. For these French pseudo-palates, the coverage of the electrodes extends toward the middle of the incisors, which allows the dental articulation of French [n] and [t] to be captured.

Two Parisian French subjects, one female (1F, the first author) and one male (2M), were recorded for 10 repetitions of each sentence in the 4 corpora ([ana], [ini], [ana] EPG, [tõtõ]).

Measurements were taken at the maximum of nasal flow in [n] and at the maximum of linguopalatal contact (in %) during the closure of [n] and [t]. Statistical analyses were made with unpaired t-tests (StatView) comparing the results by Prosodic Positions. A result obtained in one particular Prosodic Position is considered statistically different from the ones obtained in other positions if it shows a statistically significant difference from the other consecutive positions.

3. RESULTS

3.1. Nasal flow experiment:

Fig.2: Maxima of nasal airflow in [n] depending on its Prosodic Position. Data for speaker 1F grouping the two vocalic contexts [ana] and [ini], averaged over 10 repetitions each.

For the male speaker, the results are not so clear. When the two vocalic contexts are combined, none of the Prosodic Positions are distinguished by the amount of nasal flow. Fig.3 presents the 2 vocalic contexts separately. For the sequence [ana], only one position stands out significantly (APi position), but the others are not different. For the sequence [ini], there is a two way distinction between the phrasal levels (APi, IPi, Ui) and the word levels (Wi and Si) (p<.05 for each pairwise comparison). Again, the higher levels have less nasal flow than the lower ones.

Fig.3: Maxima of nasal airflow in [n] depending on its Prosodic Position and separated by vocalic context. Data for speaker 2M, averaged over 10 repetitions.

3.2. EPG experiment:

Fig.4: Maxima of linguopalatal contact in [n] and [t] depending on their Prosodic Positions. Data for speaker 1F (top) and speaker 2M (bottom), average over 10 repetitions.

Fig.5 gives an example of the pattern of linguopalatal contact for [n] at different Prosodic Positions for one repetition of speaker 1F. All consonants have a full closure and most of the variation in the amount of contact is behind the seal. This increase in linguopalatal contact as [n] occupies a progressively higher position in the prosodic hierarchy may be due to a modification in tongue blade orientation and/or tongue body height.

Fig.5: Patterns of linguopalatal contact for [n] depending on its Prosodic Position for one repetition of speaker 1F.

DISCUSSION and CONCLUSION

In this study, we have shown that the magnitude of oral and nasal articulation of a consonant can be influenced by its position in the prosodic structure. This influence has been found in several prosodic levels in a single study and in a language not previously studied from this point of view: French. In general, we found that consonants in higher Prosodic Positions (that is consonants preceded by a strong prosodic boundary) tend to have less nasal flow and more linguopalatal contact than consonants in lower Prosodic Positions.

The influence of Prosodic Position has been observed to apply to two remote articulatory subsystems: the tongue and the velum. However, the variation in the amount of nasal flow shows less distinction between Prosodic Positions than the amount of linguopalatal contact does. This result can be interpreted in two ways. (1) It could be that velum articulation distinguishes fewer positions than lingual articulation; a difference in degrees of freedom between the two articulators may prevent the velum from varying too much without endangering the velopharyngeal opening required to maintain nasality; for the tongue, on the contrary, greater variation in tongue height may be allowed behind the seal, as shown in the EPG profile in Fig.5. (2) The difference in the sensitivity of the techniques (airflow is only an indirect measurement of velic articulation) could also explain the difference in the number of distinctions. These two hypotheses need to be tested in an experiment using the same technique of investigation for the 2 articulators (for example with an articulograph). In either case, the fact that Prosodic Position can influence these two articulators suggests that prosodic information is a high-level component of speech production control. As a consequence, we feel the need for a prosodic tier in a model of speech production in order to account for the articulatory characteristics due not only to prosodic prominence (stress or accent), but also to particular Prosodic Positions (or phrasing). Only differences in magnitude of articulatory movements have been discussed here, but Prosodic Position may also affect the timing of articulatory gestures as suggested in the coarticulation experiments reviewed in the introduction.

The patterns of variation of lingual and velic articulation may appear at first glance to be opposite: in higher Prosodic Positions, the contact of the tongue against the palate is increased for coronal consonants, whereas the velopharyngeal opening of nasal consonants seem to be reduced (as found by Krakow, 1989). These modifications seem to be contradictory if we looked at the effect of Prosodic Position as a "strengthening" of articulation. But from an acoustic or perceptual point of view, this increased magnitude of oral articulation and decreased magnitude of nasal articulation may contribute to the same goal: the increase of the consonantality, and more generally the salience, of the consonant in "strong" Prosodic Positions. Manuel [1991] suggested that the reduction of velopharyngeal opening (therefore nasal flow) for nasal consonants in word initial positions contributes to the reduction of the sonority of the consonant. To test this hypothesis, acoustic consequences of this variation in articulation need to be further examined.

The results presented here were observed for two consonants and two subjects. The influence of Prosodic Position found agrees with previous results, but it needs to be examined with several subjects. Previous experiments involving 3 to 4 subjects have shown that prosodic influence on both nasal airflow in French [Fougeron, 1996] and lingual articulation in English [Fougeron & Keating, 1996] vary depending on the subjects. In both studies, although almost every subject showed distinctions between the more extreme prosodic levels, the number and the nature of the Prosodic Positions distinguished by articulation was a function of the subject.

The results presented here also suggest that some of the variation found in articulatory studies may be controlled, explained or understood by looking at the prosody of the speech materials. Therefore, awareness of prosodic differences between utterances can turn some apparently random variations into predictable regularities of speech production. However, these prosodic differences may not be reflected in differences in punctuation. For example, we found a variation in linguopalatal contact between the consonants initial in an Accentual Phrase and those initial in an Intonational Phrase, and these positions are both marked by a comma. On the other hand, we did not find a distinction between Utterance- and Intonational Phrase- initial positions, and these two positions are marked by different punctuation: one is marked by a period, the other by a comma.

Moreover, articulatory variation may be a valuable source of information in understanding the prosodic organization of speech. The possible acoustic and perceptual relevance of these position-dependent articulatory variations needs to be defined and tested, but from a production point of view, variation in articulation may follow the prosodic hierarchy and somehow cue prosodic phrasing.

This research was supported in part by an M.R.T. allocation to the D.E.A. de Phonétique de Paris, and by NSF grant SBR-9511118 to P. Keating. This article is an extended version of the paper presented at the 1st ESCA Tutorial and Research Workshop on Speech Production and Modeling and 4th Speech Production Seminar, held in Autrans (France) May 20-24, 1996. A shorter version of this paper (including results for only one speaker) is published in the proceedings. We would like to thank here Richard Sico from Kay Elemetrics Corp. for his assistance, the ESCA committee for the attribution of an ESCA grant to the first author to participate to the Workshop, and Laurent Girard for his cooperation as a subject in this study.

Beckman, M., Edwards, J., and Fletcher, J. (1992) Prosodic structure and tempo in a sonority model of articulatory dynamics. In Papers in Laboratory Phonology II: Gesture, Segment, Prosody, edited by G. Docherty and D. R. Ladd (Cambridge U. Press, Cambridge UK), Chap.3 68-86

Beckman, M. and Edwards, J. (1994) Articulatory evidence for differentiating stress categories. In Papers in Laboratory Phonology III: Phonological structure and phonetic form, edited by P.A. Keating (Cambridge U. Press, Cambridge UK), Chap.2 7-33

Benguerel, A.-P. (1977) Velar coarticulation in French: a fiberscopic study. Journal of Phonetics 5 149-158

Byrd, D. M. (1994) Articulatory timing in English consonant sequences. PhD dissertation, UCLA. (UCLA Working Paper 86)

DeJong, K. (1995) The supraglottal articulation of prominence in English: Linguistic stress as localized hyperarticulation. J.A.S.A. 97 491-504

Fougeron, C. (1996) Variation de débit nasal en fonction de la position prosodique de [n] et [ã] en français. In Actes des XXIè Journées d'Etude sur la Parole, Avignon, 215-218

Fougeron, C. & P. Keating (1996) Articulatory strengthening in prosodic domain-initial position. UCLA Working Paper 92 61-87

Fougeron, C. & P. Keating (1995) Demarcating prosodic groups with articulation. J.A.S.A 97-5 pt.2

Fujimura, O. (1977) Recent findings on articulatory processes. In Articulatory modeling and phonetics (Carré, Descout & Wajskop eds.). GALF 115-126

Hardcastle, W.J. (1985) Some phonetic and syntactic constraints on lingual coarticulation during /kl/ sequences. Speech Communication 4 247-263

Jun, S.-A. & C. Fougeron (1995) The Accentual Phrase and the prosodic structure of French. ICPhS XIII 2 722-725

Jun, S.-A. (1993) The phonetics and phonology of Korean prosody. PhD dissertation, The Ohio State University

Krakow, R. A. (1989) The articulatory organization of syllables: a kinematic analysis of labial and velic gestures. PhD dissertation, Yale University

Krakow, R. A., Bell-Berti, F., & Wang, Q.E. (1994) Supralaryngeal declination: evidence from the velum, in Producing speech: A Festschrift for Katherine Safford Harris, edited by F. Bell-Berti and L. Raphael (AIP Press, Woodbury NY) Chap. 23 333-353

MacClean, M. (1973) Forward coarticulation of velar movement at marked junctural boundaries. J.S.H.R. 16 286-296

Manuel, S. (1991) Some phonetic bases for the relative malleability of syllable-final vs. syllable-initial consonants. ICPhS XII 5 118-121

Pierrehumbert, J. & D. Talkin (1992) Lenition of /h/ and glottal stop. In Papers in Laboratory Phonology II: Gesture, Segment, Prosody, edited by G. Docherty and D. R. Ladd (Cambridge U. Press, Cambridge UK) Chap. 4 90-117

Vaissière, J. (1988) Prediction of velum movement from phonological specifications. Phonetica 45 122-39