Browsing by Author "Schweitzer, Katrin"

Now showing 1 - 1 of 1

Open Access
Frequency effects on pitch accents: towards an exemplar-theoretic approach to intonation
(2012) Schweitzer, Katrin; Dogil, Grzegorz (Prof. Dr.)
This thesis presents four corpus experiments which aim to bridge the gap between research on exemplar-theoretic phenomena and research on intonation. The main assumptions of these two areas are contradictory in some respects: while exemplar-theoretic, or episodic, approaches to language acquisition assume that acoustic detail is stored, and consequently fundamental frequency should be part of the stored mental representation of linguistic units, the most widespread intonation models, autosegmental-metrical theories of intonation, are based on the assumption that intonation in Germanic languages is assigned post-lexically. To investigate if and how intonation can be incorporated into existing exemplar-theoretic models, the described experiments examine tonal parameters of German and English speech data with respect to their being subject to frequency of occurrence effects. The studies target three central questions. Firstly, does frequency of occurrence affect pitch accent realisation? Secondly, how do the word and the tonal level interact in exemplar-theoretic selection? And thirdly, what are the essential features of an exemplar model accounting for intonation? To answer these questions, all four experiments explore the realisation of pitch accent tokens in context. The results demonstrate that tonal contours are subject to frequency of occurrence effects. Specifically, they are influenced by the frequency of the linguistic context. Moreover, they demonstrate that intonation can undergo entrenchment effects and they indicate that the word as well as the tonal level seem to be crucial for exemplar selection. Since the results of the corpus studies are well explainable in an exemplar-theoretic framework, an exemplar-theoretic model incorporating intonation seems desirable. At the same time, details of the outcomes have implications for such a model. In essence, the model should assume storage of F0-contours in contexts, including the possibility to store complex units with their tonal contour as one exemplar. The selection of exemplars to construct a production target should consider the word as well as tonal features (such as being prominent). Furthermore, production biases towards certain tonal characteristics such as greater accent ranges for frequent exemplars should be possible, and the model should average over several exemplars to construct a production target so that these biases can be counterbalanced and entrenchment can be modelled. To summarise, the work presented here constitutes a step towards an integrated exemplar-based model of speech production in which intonation is accounted for. The corpus studies suggest that Exemplar Theory is highly suited to model intonation production and they highlight the impact that frequency of occurrence has on tonal parameters just as it has been shown to be influential on the segmental level. Consequently, if intonation is to be modelled in other frameworks, frequency effects should be taken into account in order to attain a comprehensive picture of the production of intonation.