PWorkshop Archives: Spring Term 2003
Aoju Chen (University of Nijmegen)
On the universality of paralinguistic meanings of intonation
Intonation, in particular, pitch variation, is argued to signal similar meanings across languages at the level of phonetic implementation of phonological pitch contours. The meanings, paralinguistic by nature, form the body of universal intonational meanings and can be derived from three biologically determined codes: Ohala's Frequency Code, the Effort Code and the Production Code (Gussenhoven forthcoming). In our cross-linguistic investigations, we have found that although by and large languages employ the biological codes in phonetic implementation, they differ in the extent to which they use these codes. Two types of language-dependence have been identified: (1) a difference in degree of the use of these codes; (2) reversed interpretations of a given prosodic variable. We argue that these differences can be accounted for by linguistic factors regarding which languages differ from each other. Three such factors have been proposed: (1) intonational grammar regarding how a given meaning should be signalled; (2) the mean pitch range of a language; (3) the primary meaning a given prosodic variable is used to signal. In this talk, in the light of our findings on the use of the Frequency Code and the Effort Code, I will discuss the two types of language-dependence in the use of universal intonational meanings and try to account for these differences by means of the three factors, in addition to presenting empirical evidence for the universality of intonational meanings.
Hannele Nicholson & Andreas Hilmo Teig
How to tell beans from farmers: Cues to the Perception of pitch
accent in whispered Norwegian
East Norwegian employs pitch accent contours in order to make lexical distinctions. This paper researches listeners' ability to make lexical distinctions in the absence of f0 (ie. whispered speech) as the listener attempts to determine which pitch accent word token best fits into a whispered ambiguous utterance in spoken Norwegian. The results confirm that syntactic context alone is not a reliable cue to assist in lexical selection and concur with Fintoft (1970) in suggesting that listeners utilise a separate prosodic cue, possibly syllable duration or intensity, to make the pitch accent distinction in whispered speech.
|Some phonetic observations on the concept of prosodic invariance: Evidence from a corpus of spoken Italian|
Jim Scobbie (QMUC)
|An articulatory study of English /l/ sandhi and vocalisation I use Alan Wrench's MOCHA multispeaker articulatory speech database to explore /l/ in English. EPG analysis reveals that eight ramdomly selected speakers, drawn from three nations with distinct phonological systems (Scotland, England and USA) all display pervasive and systematic /l/ vocalisation (lack of alveolar contact). Vocalisation of word-final /l/ is very strongly context-dependent for seven subjects. These English speakers have a post-lexical external sandhi alternation of consonantal vs. vocalic /l/. Onset vs. coda syllabification of /l/ is not sufficient to condition the distribution of vocalisation, though prosodic factors are necessary, as well as segmental and phrasal ones. I will present the general tendencies and the systematic linguistic differences between the speakers, which are orthogonal to national dialect.|
Quantitative Tests of an (Almost) Autosegmental-Metrical Model of
Stem-ML can be used to build quantititative models of intonation, and one can test those models by comparing them to f0 tracks. One can build linguistically suggestive models of Mandarin, Cantonese, and a small domain of English intonation, that describe the data well, and are almost (but not quite) Autosegmental-Metrical models. The models compute tonal coarticulation based on a simple model of muscle dynamics and the brain's planning process. The models yield a measurement that can be interpreted as the importance of a word or syllable. We show that these measurements are generally consistent with linguistic assumptions.
Articulatory-acoustic features in ASR - beyond the
In current state-of-the-art automatic speech recognition systems, the acoustic signal is usually described in terms of phones, and words are simply seen as concatenations of phone sequences. However, as many before me have pointed out, the notion that a word is composed of a sequence of phone segments, i.e., the "beads-on-a-string" paradigm, is questionable. In this talk, I will present my views on how articulatory-acoustic features in combination with syllables can be used to circumvent the beads-on-a-string issue and at the same time can be used to improve automatic speech recognition.