Articulatory Coordination in Coronal Stops:

Articulatory data obtained from X-ray films were analyzed to investigate articulatory coordination in Swedish vowels and dental and retroflex stops. Several quantitative frameworks for predicting observed tongue contours were evaluated. Performance was assessed on a set of articulations traced at the V1:C boundary in V1:CV2 sequences. The best model, as determined by an rms-criterion, had 3 parameters: length of blade (dental/retroflex), degree of coarticulation, pharynx width. According to this account the coarticulatory (vowel-dependent) variations in [F] and [Ç] tongue contours are seen as arising from the overlap of a gradual deactivation of the V1: gesture and the movement towards a vowel-independent consonant target, a process successfully captured by interpolation in the model. This result is discussed in terms of co-production (Öhman, [1]) and the overlapping innervation wave theory (Joos [2]). 1. COARTICULATION AND ’LOCUS EQUATIONS’ There are probably few phonetic topics that have generated so much experimentation and theoretical speculation as coarticulation. Nonetheless, it is fair to say that, despite massive research efforts, our understanding of the physiological origins and perceptual motivation for this phenomenon still remains rather incomplete. One approach to coarticulation is the use of ’locus equations’ as promoted by Sussman and colleagues [3]. This format arises from the fact that the F2 onset of a consonant tends to be a linear function of the F2 of the following vowel. Numerous studies have shown that LE slopes & intercepts vary systematically and robustly with consonant place. Descriptively LE’s can be seen as restating the fact that adjacent phonemes interact physically. In capturing that interaction, LE’s offer a way of quantifying ‘coarticulation’. What are constraints that determine possible ‘locus’ patterns? Are LE’s always linear? When linear, what is the origin of their slopes and intercepts? According to Sussman et al [4] LE’s have a strong perceptual motivation, whereas Fowler [5] maintains that LE properties (linearity, slopes, intercepts) are inevitable consequences of production constraints. On this view a consonant produced at a certain place is supposed to offer a fixed degree of articulatory ‘resistance’ to adjacent vowels. Therefore, within a given stop place category, coarticulation tends to be uniform across vowel contexts. LE’s simply reflect uniform coarticulation. We shall return to this issue at the end of the paper. 2. AN X-RAY DATABASE Recent collaboration with Danderyd Hospital in Stockholm on an X-ray project provides a unique opportunity to re-examine some of the traditional issues raised by coarticulation. So far the database consists of X-ray films of 11 subjects (Swedish, Tamil, Hindi) recorded for 20 seconds at 50 images per second. Speech samples were selected to elucidate how movements are coordinated in the coarticulation of vowels and labial, dental, retroflex and velar stops. For background and methodology of this research see Stark et al, this conference [6]. 3. QUESTION AND SPEECH SAMPLE Here we summarize the results of an in-depth analysis of the dental-retroflex contrast for a single Swedish subject [7]. The relevant speech samples are listed in Table 1 below. Table 1. The speech samples. The first vowel carries main stress and accent 2. The second vowel is unstressed. Consonant Vowel (V1) place anterior posterior