Stress at higher levels (i.e. the phrase, sentence and discourse level) is usually considered to be predetermined by the syntactic structure of phrases and/or the semantic representation of the sentence (Dogil, 1987; Cinque, 1993). The phonetic realizations of the outputs of the various stress levels are also different. The higher levels of stress are realized phonetically by the distribution of pauses and intonational features across the utterance, in contrast to the phonetic realization of lexical (word) stress, which is much more variable and language-dependent. Intonation is used for marking grammatical boundaries (phrases or clauses) and for signalling sentence modality (statement or question) as well as in the paralinguistic layer of communication to indicate the speaker's attitude, emotions, etc.. Fery (1993) and Uhmann (1991) developed models of German intonation that focus on the phonological representation of the tonal aspects of intonation.
German intonation is primarily described by changes in the fundamental frequency. The characterization of sentence melody provided by a state-of-the-art linguistic analysis leads to an inventory of pitch accents which have a distinct semantic or pragmatic discourse meaning (Beckman, 1986; Hirschberg, 1995). The pitch accent inventory for German (Fery, 1993; Kohler, 1995; Grice et al., 1995; Mayer, 1995), usually based on the standard ToBI pitch contour description (Silverman et al., 1992), includes different pitch accents aligned with stressed syllables, each of them specifying a certain level in the overall pitch range or a movement between two levels (low and high). The Stuttgart System of transcription of German intonation (Mayer, 1995) uses an inventory of labels and linking rules which constitutes a minimal system of symbols that are assigned to specific functions in the domain of discourse representation theory (Kamp & Reyle, 1993). Within the Stuttgart System, to which we shall be referring in the experiment to be described below, the following pitch accents are used: H*L , L*H, L*HL (late peak), HH*L (early peak) (Mayer, 1995). Prosodic phrasing of utterances is described on two levels, intermediate and intonational phrases. Intonational phrases are marked with initial and final boundary tones (%T, T%). By default, pitch at the beginning of an intonational phrase is expected to start in the lower part of the speaker's pitch range. The final boundary tone is either high (H%) or low (L%) and is aligned with the utterance's final syllable.
The intonational pitch movements are the result of laryngeal activity. The pitch accent affects not only the changes in pitch but also other parameters of the glottal airflow. This is partially caused by the fact that the accent is carried by the lexically stressed syllable, which, as we will show in the experiment described below, influences voice quality. It is our goal to find the physiological correlates of pitch accent.
An example of labelled pitch accents is given in the sentence "Peter ging zum Postamt, das in der Königstraße liegt" (Peter goes to the post office which is at King's Street) in Fig.32.
| Figure 32. Example of a sentence with intonation labelled according to the Stuttgart System (Mayer, 1995) |
|
As noted above, German intonation is described as a distribution of pitch accents and boundary tones. The two basic types of labels, in terms of the Stuttgart System, which we refer to in this experiment are: H*L and L*H (Mayer, 1995) for pitch accents and H% and L% for boundary tones, i.e. we will concentrate on rising and falling pitch accents which were not addressed in the word stress experiment.
The changes in the F0 contour are not the only correlates of the changing pitch level. Pierrehumbert (1989) argues for changes in the glottal waveform due to the pronunciation of various pitch accents. These differences are caused by a higher voice effort (subglottal pressure) and additional adjustments of the vocal folds depending on whether pitch is raised or lowered. Using the inverse filtered speech signal and the L-F (Liljencrants-Fant) voice source model (Fant, 1995), Pierrehumbert (ibid.) found a significant difference in the excitation of accented English vowels. The H-type tones have a greater Open Quotient than the L-type tones, louder voice has a smaller OQ than softer voice. The asymmetry of the glottal waveform increases with the level of sound intensity. It was our task to verify Pierrehumbert's findings for German, but using the robust parametrization of the EGG signals instead of method of inverse filtering method or the indirect acoustic method. The word stress experiment had shown the correlations between these methods, and the question was then whether German pitch accents correlate with differences in glottal activity.
