When people speak, they constantly speed up and slow down, i.e. they modulate their articulation speed. At what particular points in discourse do these modulations occur and why? This DoBeS project, funded by the Volkswagen Foundation, investigates variation in the articulation speed of words in context and investigates the complex interaction of factors that determines it. We have good indications, for instance, that people pronounce relatively uninformative words relatively quickly and that they tend to compress long words. But how and why exactly? And does this hold for all languages?

Previous studies on these issues have only been able to give partial answers for two reasons: They are often based on experimental rather than naturalistic data and, more importantly, they are almost exclusively based on data from individual languages, mostly English (which is particularly unfortunate since English has relatively simple morphology, i.e. few means to build complex words). The project proposed here promises to take our understanding of articulation speed variation a considerable step further by carrying out a set of integrated studies using as a unique resource a set of corpora from ten languages spoken all around the world, from the Amazonian rainforest (Bora and Baure), to Mexico (Texistepec Popoluca), the North American great plains (Hoocąk), Siberia (Sakha and Even), the Himalaya (Chintang), the Kalahari desert (Nǀuu), and to urban Western Europe (Dutch) and the USA (English).

This diversity will allow us to identify, on the one hand, the universal features of articulation speed variation, which we hypothesize are connected to general properties of human cognition, and, on the other hand, languagespecific features, which are probably related to properties of the grammar of individual languages or are culture-specific, e.g. rooted in local narrative traditions and conversational strategies.

This project investigated the relative frequencies of core parts of speech, such as nouns, verbs, and pronouns, in spoken language corpora of seven languages that represent a wide range of areal and typological diversity.

The AUVIS project is a collaboration between different institutions from IT and the humanities aiming at developing audio/video software to manage large quantities of data in the social sciences. Technological progress in AUVIS is paired with scientific research on the interaction of grammar, prosody and gesture in event segmentation.

Find more information on the AUVIS project at the Language Archive website.

Contact: Meytal Sandler M.A.