Matthias Jilka University of Stuttgart Institute of Natural Language Processing Azenbergstr.12 70174 Stuttgart Room 3/7, Phone: 0049 / 711 / 121-1377

  Some biographical data

Already at the tender age of three I started correcting my parents' pronunciation. This early interest in phonetic issues has never left me since and  certainly is responsible for the direction my academic life has taken (It is also responsible for some occasional problems in my social life, but I won't go into that now).

Anyway..., I studied Linguistics, as well as English ("Anglistik") and Romance literature and languages ("Romanistik") at the University of Stuttgart and the University of Washington (Seattle).

After graduation (M.A.) I became a member of the graduate program Linguistic Foundations of Language Processing in June 1997. At the end of the three year scholarship I had managed to write my thesis and with the oral exam in July I have now manged to successfully finish the dissertation project. As of October 1, 2000 I will working at AT&T Research Labs in Florham Park, New Jersey.

Dissertation Project  

IMS PhD Students !!!

For information about the sugggested procedure and how to avoid problems when writing one's dissertation at the IMS, click here.

Summary and overview of my PhD thesis taken from its introductory chapter:

The contribution of intonation to the perception of foreign accent  

While it is generally acknowledged that prosody and especially intonation contribute to the impression of foreign accent, the number of studies and analyses explicitly dealing with the description of prosodic/intonational foreign accent is rather small, particularly in the context of models of second language acquisition.

Contrastive studies of intonation in two or more languages (for example Grabe's comparison of aspects of German and British English intonation) are however relatively frequent and do give strong indications of possible causes and manifestations of intonational foreign accent that are likely to occur, if native speakers of one of the contrasted languages were to speak the respective other language. Apart from these comparative studies of intonation there are only few studies directly combining the areas of prosody, particularly intonation, and foreign accent.

A notable exception is the 1983 dissertation "English intonation from a Dutch point of view" by Nico Willems which examines intonational deviations in the English productions of native speakers of Dutch, draws up a corresponding inventory and confirms the deviations' relevance in perceptual experiments using resynthesis. The main aim of Willems's study is to give an exact phonetic description of these relevant deviations and to use it in the creation of an effective course of (British) English intonation for native speakers of Dutch. Willems neither specifically relates his findings to an encompassing linguistic framework, nor does he analyze them within the context of second language acquisition.

Segmental aspects of foreign accent, on the other hand, have been described in detail in their various forms. Accordingly, the existing models of second language acquisition have been developed on the basis of the segmental aspects of acquisition and foreign accent. For the reasons stated above it can thus be claimed to be the overall motivation of this study to give a detailed, coherent account of intonational foreign accent (in the comparison of two specific languages) and to attempt to connect this description to the established theories of second language acquisition. In more concrete terms the major objectives of this study can be summarized as follows.

To ensure a scientifically exact identification of intonational foreign accent the use of speech technology is advantageous. In an analysis-by-synthesis procedure a foreign-accented utterance will be compared with utterances differing in only one particular tonal aspect. The respective judgments regarding the perception of foreign accent will determine if the detected and subsequently altered tonal characteristics are indeed instances of intonational foreign accent. This is also the main reason why American English and German were chosen for this analysis. For both languages the necessary intonation description and possibilty of F0 generation are available. While the aspect just described controls to a large extent how the analysis is performed, the second objective of the study must define the conditions under which a meaningful analysis of intonation and intonational foreign accent is possible. This concerns especially the recognition of the fact that the form of the detected intonational deviations, i.e. instances of intonational foreign accent depends on the chosen model of intonation description. In addition to this, elaborate assumptions about the role of context and variation in intonation must also be taken into consideration in the identification of intonational deviations. On the basis of these prerequisites the third objective of the study can address the actual identification of concrete manifestations of intonational foreign accent in the German productions of native speakers of American English, and vice versa in the American English productions of native speakers of German, as well as the subsequent systematic classification of major types of intonational foreign accent. Finally, the study aims to investigate the contribution of prosody and intonation in particular to the general perception of foreign accent, also determining the shares of intonation and the remaining aspects of prosody in relation to each other and comparing both with the contribution of segmental aspects.

Overview of the study

In the chapter following these introductory remarks a general definition of foreign accent is discussed, and previous research in the areas of foreign accent and second language acquisition is summarized. Important factors and theories concerning the neurophysiological as well as sociopsychological aspects relevant in L1 and L2 acquisition are briefly introduced. Furthermore, established models of language representation are presented along with corresponding theories of second language acquisition, namely the Speech Learning Model and the Perceptual Assimilation Model.

While this chapter introduces the "foreign accent aspect", the following chapter covers the the study's basis in intonation description and perception.A general account of how prosodic characteristics are generally perceived is given, as well as an overview of previous research directly or indirectly combining the areas of intonation and foreign accent. As already indicated, the question of the dependence of the form of the detected instances of intonational foreign accent on the applied theory of intonation description is addressed with an emphasis on models that view the intonation contour either as one continuous tonal movement or a sequence of discrete tonal events. Another necessity for the analysis of intonation, and especially the identification of tonal deviations, is also discussed in this chapter, as a number of assumptions about the general characteristics of intonation are made. These assumptions particularly concern the aspects of context-dependence and the high degree of variation in intonation, which both have to be taken into account in the identification of those tonal deviations that really do constitute instances of intonational foreign accent.

Chapter 4 then contains the actual analysis of concrete cases of intonational foreign accent in the German productions of native speakers of American English and the American English productions of native speakers of German. Recordings of foreign-accented speech are examined within the framework of the Tone Sequence Model (Pierrehumbert 1979, 1980) and the corresponding ToBI system of intonation description (Silverman et al., 1992). The analysis of intonational foreign accent should therefore interpret tonal deviations in terms of tonal categories and their phonetic realizations. The confirmation of the relevance of the detected deviations to the perception of foreign accent is, as already discussed, achieved by means of F0 generation and subsequent resynthesis. Examples of detected major types of intonational foreign accent are depicted as F0 contours both for the original and the improved, rule-generated versions. The corresponding sound examples are available at http://www.ims.uni-stuttgart/phonetik/matthias/sound4diss.html

In accordance with the prior insights concerning the dependence of the results on the chosen view of intonation, an alternative representation of the phonetic realization of tonal categories is presented in chapter 5. Using the method of Parametric Intonation Events (PaIntE), developed in Möhler (1998), the phonetic realizations of examined categories are described by means of an alternative set of parameters that are easily accessible for a statistical analysis.

In chapter 6 the results of the performed analysis of intonational foreign accent are reviewed in perception tests, the comparison of original and rule-generated/resynthesized versions of an utterance in foreign accent ratings and direct pair comparisons additionally representing the general contribution of intonation to foreign accent. In a further comparison of rule-generated stimuli (retaining segmental but not intonational foreign accent) and fully synthesized stimuli (without segmental but with intonational foreign accent), it is attempted to get an impression of the relation between the segmental and intonational aspects of foreign accent. The second major part of the perception test uses low-pass filtered stimuli to investigate the influence of prosody in general on listeners' ability to identify and distinguish languages. If the prosodic characteristics of a language are perceived as typical of it, then they will also appear and be perceived as foreign accent in the L2 productions of non-native speakers. These tests are repeated using low-pass filtered stimuli with monotonous intonation, so as to examine whether prosodic features such as rhythm or speaking rate are sufficient in the tasks of language identification and distinction. The tests also serve to evaluate the contribution of intonation, which is a source of information in the regular low-pass filtered stimuli but not in those with monotonous intonation.In the concluding chapter all results are summarized and interpreted with respect to the main objectives formulated earlier so that a comprehensive picture of intonational foreign accent in German and American English will have emerged.

If you would like to read the complete thesis go here

Publications

Matthias Jilka, 1996. Regelbasierte F0 Generierung der Intonationsmuster des Amerikanischen Englisch. Magisterarbeit. Universität Stuttgart

Matthias Jilka und Gregor Möhler, 1998. Intonational Foreign Accent: Speech Technology and Foreign Language Teaching. In: Proceedings of the ESCA Workshop on Speech Technology in Language Learning, Marholmen, pp. 115 - 118

Matthias Jilka, 1998. Bestimmung Intonatorischen Fremdsprachlichen Akzents mit Hilfe von F0 Generierung. In: Proceedings der 29. Jahrestagung der Gesellschaft für Angewandte Linguistik "Sprache und Technik" in Dresden, p.81 (Abstract)

Matthias Jilka, 1999a. Intonational Foreign Accent in the Speech of American Speakers of German. JASA Vol. 105, No. 2, Pt. 2, p. 1094 (Proceedings of 137th Regular Meeting of the Acoustical Society of America on CD Rom)

Matthias Jilka, Gregor Möhler und Grzegorz Dogil, 1999b. Rules for the Generation of ToBI-based American English Intonation. Speech Communication 28, pp. 83 - 108.

Matthias Jilka, 1999c. Identifying Intonational Foreign Accent with the Help of Different Methods of F0 Generation. In: Proceedings of ICPhS 1999 Vol. 2, pp. 1447 - 1450 (on CD Rom)

Matthias Jilka, 2000. The Contribution of Intonation to the Perception of Foreign Accent. Doctoral Dissertation, Arbeiten des Instituts für Maschinelle Sprachverarbeitung (AIMS) Vol. 6(3), University of Stuttgart

Former member of the graduate program Linguistic Foundations of Language Processing at the University of Stuttgart.

jilka@ims.uni-stuttgart.de