Abstract :
[en] Nasal vowels (NV) illustrate both universal and language-specific patterns. A well-documented universal is that the number of NV never exceeds and is often less than the number of oral vowels (OV) in a language; mid vowels are most likely to be missing. This loss of height contrasts can be related to the frequency shifts that are often reported in the F1 region due to nasal coupling: high vowels are lower and low vowels are higher when nasalized.
Less studied is the very low frequency of non-peripheral NV, which are much less frequent than non-peripheral OV. Mergers of NV along the front-back dimension all occur [1]. These gaps and mergers are expected, along with those in the height dimension because nasal coupling changes the entire spectrum of a NV compared to its oral counterpart, not just the F1 region.
This paper is about the phonetic implementation of the [nasal] contrast for French vowels, in which the F2-related front-back dimension plays a leading role. The first part of this study describes the production of French NV, relating articulatory (MRI) and acoustic data for four Belgian speakers. The second part manipulates the resulting articulatory-acoustic properties in a three-part speech perception experiment.
The corpus consisted in 28 sequences including (1) CV, VC, [mV], [Vm] and [mVm] items, where C is [p] or [s], and V can be every nasal vowel and its oral counterpart, (2) VV~V items, where the middle vowel is either a phonemic nasal vowel, or a 'nasalized' vowel. In the latter case, the speakers were asked to hold all oral articulators in the same position while pronouncing the sequences, in order to study the acoustic effects of nasal coupling alone. The acoustic study consisted in the analysis of cepstrally-smoothed spectra computed in the middle of the vowels. The MRI data were acquired at the Brussels Erasmus hospital using the real-time technique [2].
The main results are the following:
(1) The lowering of the velum has many acoustic effects : the general level of energy is notably inferior for nasalized and nasal vowels. The reduction in energy affects in particular the region of F1, where the spectral prominence is broadened. F3 is also affected, often to the point of being cancelled.
(2) French NV not only differ in soft palate height from their oral counterparts but also in the positions of tongue and lips, to an extent that their acoustics are quite different. The phonetic implementation of the [nasal] phonological contrast involves many covarying articulations, including tongue retraction, tongue lowering and lip rounding: /e~/= [{~], /9~/= [&~], /a~/ = [Q_r], /o~/ = [o~].
The consistency of the observations leads us to investigate the role of these articulatory-acoustic properties for the detection of the phonological contrast between nasal and oral vowels in French. Three perceptual experiments are planned using 7-by-7 arrays of synthetic stimuli. The stimuli are CVC items, where C is [t], and the spectral properties of V [E/,A/] are varied along two dimensions per array, from an oral to a nasal(ized) endpoint.
The production results just described were used to select acoustic dimensions for manipulation. Two candidates are considered for the first dimension, which is related to the acoustic effects of the lowering of the velum, these are 'intensity', which consists in an increase of all formant bandwidths, and 'compactness', which involves F1 broadening, F2 sharpening and F3 cancelling. The second dimension, 'gravity', corresponds to the general acoustic effect of lip rounding and tongue retraction: the lowering of F2.
First, a trading-relations experiment will be carried out, in order to test the hypothesis that both acoustic properties are taken into account by French listeners in deciding whether a vowel is oral or nasal. If they do, then the differences in the oral articulations are as much a part of the contrast as the soft palate lowering.
Second, the integration of each pair of dimensions will be measured using a fixed classification paradigm on a subset of the stimuli. This experiment measures the extent to which the covariation of nasal coupling and oral articulations increases the perceptual distance between vowels [3].
Finally, a category-goodness experiment will be run, in order to find which one of the acoustic dimensions is prior when a listener must classify a sound as a nasal vowel and judge how good it is as an exemplar of this category.
The perception of these stimuli by American English listeners will also be tested in each of these tasks to determine the extent to which the absence of a [nasal] vowel contrast in that language influences the perception of these stimuli.
References
[1] Vallée, 1994. Systèmes vocaliques: de la typologie aux prédictions. Doctoral dissertation, ICP Grenoble.
[2] Metens, Soquet, Demolin, 2000. Real Time MRI and Coordination of Speech Movements, Model Driven Acquisition, Grenoble.
[3] Macmillan et al. 1999. Integrality of nasalization and F1. II. JASA 106,5, 2913-2932.
