Fabian Tomaschek

Investigating dialectal differences using articulography

The present study introduces articulography, the measurement of the position of tongue and lips during speech, as a promising method to the study of dialect variation. By using generalized additive modeling to analyze articulatory trajectories, we are able to reliably detect aggregate group differences, while simultaneously taking into account the individual variation across dozens of speakers. Our results on the basis of Dutch dialect data show clear differences between the southern and the northern dialect with respect to tongue position, with a more frontal tongue position in the dialect from Ubbergen (in the southern half of the Netherlands) than in the dialect of Ter Apel (in the northern half of the Netherlands). Thus articulography appears to be a suitable tool to investigate structural differences in pronunciation at the dialect level.

Words from spontaneous conversational speech can be recognized with human-like accuracy by an error-driven learning algorithm that discriminates between meanings straight from smart acoustic features, bypassing the phoneme as recognition unit

Sound units play a pivotal role in cognitive models of auditory comprehension. The general consensus is that during perception listeners break down speech into auditory words and subsequently phones. Indeed, cognitive speech recognition is typically taken to be computationally intractable without phones. Here we present a computational model trained on 20 hours of conversational speech that recognizes word meanings within the range of human performance (model 25%, native speakers 20–44%), without making use of phone or word form representations. Our model also generates successfully predictions about the speed and accuracy of human auditory comprehension. At the heart of the model is a ‘wide’ yet sparse two-layer artificial neural network with some hundred thousand input units representing summaries of changes in acoustic frequency bands, and proxies for lexical meanings as output units. We believe that our model holds promise for resolving longstanding theoretical problems surrounding the notion of the phone in linguistic theory.

Practice makes perfect: the consequences of lexical proficiency for articulation

Abstract Many studies report shorter acoustic durations, more coarticulation and reduced articulatory targets for frequent words. This study investigates a factor ignored in discussions on the relation between frequency and phonetic detail, namely, that motor skills improve with experience. Since frequency is a measure of experience, it follows that frequent words should show increased articulatory proficiency. We used EMA to test this prediction on German inflected verbs with [a] as stem vowels. Modeling median vertical tongue positions with quantile regression, we observed significant modulation by frequency of the U-shaped trajectory characterizing the articulation of the [a:]. These modulations reflect two constraints, one favoring smooth trajectories through anticipatory coarticulation, and one favoring clear articulation by realizing lower minima. The predominant pattern across sensors, exponents, and speech rate suggests that the constraint of clarity dominates for lower-frequency words. For medium-frequency words, the smoothness constraint leads to a raising of the trajectory. For the higher-frequency words, both constraints are met simultaneously, resulting in low minima and stronger coarticulation. These consequences of motor practice for articulation challenge both the common view that a higher-frequency of use comes with more articulatory reduction, and cognitive models of speech production positing that articulation is post-lexical.

Lexical frequency co-determines the speed-curvature relation in articulation

Abstract The relation between speed and curvature provides a characterization of the spatio-temporal orchestration of kinematic movements. For hand movements, this relation has been reported to follow a power law with exponent - 1 / 3 . The same power law has been claimed to govern articulatory movements. We studied the functional form of speed as predicted by curvature using electromagnetic articulography, focusing on three sensors: the tongue tip, the tongue body, and the lower lip. Of specific interest to us was the question of whether the speed-curvature relation is modified by articulatory practice, gauged with words’ frequencies of occurrence. Although analyses imposing linearity a priori indeed supported a power law, relaxation of this linearity assumption revealed that the effect of curvature on speed levels off substantially for lower values of curvature. A modification of the power law is proposed that takes this curvature into account. Furthermore, controlling statistically for number of phones and word duration, we observed that the speed-curvature function was further modulated by an interaction of lexical frequency by curvature, such that for increasing frequency, speed decreased slightly for low curvatures while it increased slightly for high curvatures. The modulation of the balance between speed and curvature by lexical frequency provides further evidence that the skill of articulation improves with practice on a word-to-word basis, and challenges theories of speech production.

Neural Processing of Acoustic Duration and Phonological German Vowel Length: Time Courses of Evoked Fields in Response to Speech and Nonspeech Signals.

Recent experiments showed that the perception of vowel length by German listeners exhibits the characteristics of categorical perception. The present study sought to find the neural activity reflecting categorical vowel length and the short-long boundary by examining the processing of non-contrastive durations and categorical length using MEG. Using disyllabic words with varying /a/-durations and temporally-matched nonspeech stimuli, we found that each syllable elicited an M50/M100-complex. The M50-amplitude to the second syllable varied along the durational continuum, possibly reflecting the mapping of duration onto a rhythm representation. Categorical length was reflected by an additional response elicited when vowel duration exceeded the short-long boundary. This was interpreted to reflect the integration of an additional timing unit for long in contrast to short vowels. Unlike to speech, responses to short nonspeech durations lacked a M100 to the first and M50 to the second syllable, indicating different integration windows for speech and nonspeech signals.

Strategies for addressing collinearity in multivariate linguistic data

Abstract When multiple correlated predictors are considered jointly in regression modeling, estimated coefficients may assume counterintuitive and theoretically uninterpretable values. We survey several statistical methods that implement strategies for the analysis of collinear data: regression with regularization (the elastic net), supervised component generalized linear regression, and random forests. Methods are illustrated for a data set with a wide range of predictors for segment duration in a German speech corpus. Results broadly converge, but each method has its own strengths and weaknesses. Jointly, they provide the analyst with somewhat different but complementary perspectives on the structure of collinear data.

Categorical processing of German vowel quantity.

The present study investigates the behavioral and neural processing of duration in the German vowel /a/ using synthesized disyllabic nonsense words. The vowel /a/ was chosen since its long and short cognates are differentiated merely on the basis of vowel duration, unlike high and mid vowels which have additional differences in formant structure. By means of an identification test a sharp category border was found at 105.9 ms in theduration continuum from short to long category associated with a local peak in reaction time. Furthermore, an adaptive discrimination test applied to the entire continuum showed a minimum of the just‐noticeable‐difference for vowel duration at the category border. In an MEG study, the continuumof the disyllabic words was presented in randomized order. Besides two P50‐M100 complexes in response to the first and the second syllable, the ERP data showed a secondary M100‐like peak in case vowel duration exceeded the duration of the short category, which was not observed in a durati...