Lauserõhu akustilised korrelaadid eesti keeles
Acoustic correlates of sentence stress in Estonian

Summary/Abstract: The study examines the acoustic correlates of sentence stress in Estonian. The data consists of 18 four-word sentences read aloud by nine speakers in answer to three questions eliciting different information structures. The test words include six sets of triplets of words differing minimally in quantity degree. The words occur in three different stress conditions: (i) after narrow focus (the unstressed condition); (ii) as the nuclear accented word in a broad focus sentence (the stressed condition); and (iii) as an emphatically accented narrow focus (used for the purposes of normalisation). The test word is always the third word in the four-word sentence in order to avoid interferences from boundary signals. The potential correlates under study are the following: (i) the F0 range and peak height of the test word, measured in the test words with a H*L pitch accent respectively as the difference between the F0 maximum of the stressed syllable and the F0 minimum of the unstressed syllable, and as the difference from the mean F0 maximum of the test word in the three stress conditions produced by a speaker; (ii) the duration of the test word in the different stress conditions, measured as the difference from the mean duration of the test word in all three conditions; (iii) the intensity range and the maximal intensity level of the test word; (iv) the values of the F1 and F2 formants as indicators of vowel quality; and (v) the spectral emphasis of the stressed syllable of the test words. The above values were analysed separately for the three quantity degrees in order to determine a potential effect of the prosodic structure of the word on the acoustic realisation of sentence stress. The data was statistically analysed with the Systat software package. The classification power of the different parameters was determined, using linear discriminant analysis. The strongest correlate of sentence stress turned out to be F0, as was expected, given that sentence stress is phonologically realised as a pitch accent. From the two F0 values, the peak height was the stronger one, with a classification power of 89%, F0 range permitting to classify correctly 76% of the data. Almost equally strong correlates as the F0 range were the duration and intensity level, classifying correctly respectively 75% and 73% of the data. Vowel quality and spectral emphasis did not correlate significantly with sentence stress. The results also revealed an effect of lexical prosody on the acoustic realisation of sentence stress: the lengthening and the rise of the intensity level were the largest in the stressed words of the third (overlong) quantity degree.

• Details
• Contents