On the lingual organization of the German vowel system

Abstract

A hybrid PARAFAC and principal-component model of tongue configuration in vowel production is presented, using a corpus of German vowels in multiple consonant contexts (fleshpoint data for seven speakers at two speech rates from electromagnetic articulography). The PARAFAC approach is attractive for explicitly separating speaker-independent and speaker-dependent effects within a parsimonious linear model. However, it proved impossible to derive a PARAFAC solution of the complete dataset (estimated to require three factors) due to complexities introduced by the consonant contexts. Accordingly, the final model was derived in two stages. First, a two-factor PARAFAC model was extracted. This succeeded; the result was treated as the basic vowel model. Second, the PARAFAC model error was subjected to a separate principal-component analysis for each subject. This revealed a further articulatory component mainly involving tongue-blade activity associated with the flanking consonants. However, the subject-specific details of the mapping from raw fleshpoint coordinates to this component were too complex to be consistent with the PARAFAC framework. The final model explained over 90% of the variance and gave a succinct and physiologically plausible articulatory representation of the German vowel space.

Tongue-surface movement patterns during speech and swallowing

“…The tongue has been frequently characterized as being composed of several functionally independent articulators ͑Hardcastle, 1976; Hoole, 1999;Mermelstein, 1973;Ö hman, 1967;Perkell, 1969;Stone, 1990͒. The common use of such terms as tip, blade, body, dorsum, and root to refer to the ''parts'' of the tongue reflects the widespread acceptance of this assertion.…”

“…There have been relatively few comprehensive reports of across-speaker differences in tongue kinematics largely because the instrumentation for tracking lingual kinematic data is expensive, as are the work hours required for data reduction ͑hence the impetus for the XRMB database; see Westbury, 1994͒. Consequently, most investigations of tongue function have studied seven or fewer participants ͑e.g., Guenther et al, 1999;Harshman et al, 1977;Hoole, 1999;Kent and Moll, 1972;Lofqvist and Gracco, 1994;Perkell and Nelson, 1985;Stone, 1990͒. The few existing investigations that have studied tongue kinematics in a large number of participants have reported large differences across speakers ͑Hashi et Westbury et al, 1998͒.…”

Predicting midsagittal pharyngeal dimensions from measures of anterior tongue position in Swedish vowels: Statistical considerations

“…This could occur if a set of observations contains consonant-vowel transitions and rhoticized articulations (cf. Hoole, 1999). It will be shown that even with a small number of observations (between 7 and 13), a reduction to three PCA components is sufficient for statistically meaningful results.…”

“…Multiple regression using PCA components as independent variables is known as principal components regression. Previous studies (Shirai and Honda, 1978;Sekimoto et al, 1978;Maeda, 1978Maeda, , 1990Beaudoin and McGowan, 2000;Story, 2007) have also used PCA and combinations of PCA with other forms of factor analysis (Hoole, 1999) to study midsagittal tongue shape, particularly in vowel production. Shirai and Honda (1978) used 50 vowel tokens from ten different speakers and Sekimoto et al (1978) used over 1000 vowel tokens from two different speakers.…”

“…The other studies segregated the observations by speaker, with Maeda (1978) using 77 observations from frames of an x-ray movie, Maeda (1990) using about 1000 observations of vowel production in sentences per speaker, Beaudoin and McGowan (2000) using over 3000 observations per speaker including consonant-vowel transitions, and Story (2007) using data from the centers of 11 vowels and the frames from six vowel pairs per speaker. In his hybrid approach, Hoole (1999) appears to have used 90 observations per speaker. These studies have shown that the number of degrees of freedom for the midsagittal tongue can be reduced to two or three, accounting for between 80% and 97% of the variance in tongue fleshpoint coordinates.…”