Learning pronunciation variants for words in a foreign language: Towards an ecologically valid theory based on experimental research and computational modeling
07 / 2012 - 07 / 2017
Learning a foreign language implies learning pronunciation variants of words in that language. This includes the words' reduced pronunciation variants, which contain fewer and weaker sounds than the words' canonical variants (e.g. cpute for English computer, and Neland for Dutch Nederland), and which are highly frequent in casual conversations. The learner has to build mental representations for these variants. Importantly, late learners will build representations that differ significantly from native listeners' representations, since reduction patterns in their native language will shape their interpretation of reduction patterns in the foreign language. Moreover, their representations will also be weaker and less well-specified: they encounter reduced variants only after having learned the word's canonical pronunciation and spelling, and only after having lost sensitivity to characteristics of the speech signal that are not functional in their native language. This project will develop the first, fully specified, theory of how late learners of a foreign language build mental representations for pronunciation variants in that language. This theory will crucially differ from existing theories of foreign language processing. First, it will go far beyond the traditional focus on clearly articulated words. Second, it will incorporate recent insights from native speech processing on representation types in the mental lexicon. Third, the theory will be based on new evidence from behavioral experiments, brain imaging experiments, and computer modeling. For the brain imaging experiments, novel techniques will be developed that simultaneously reveal the location and time course of brain activation during speech processing. The computational model will, unlike nearly all existing models, take real speech as its input. All experiments will investigate Dutch and Spanish learners of English and French with low proficiencies. The project will also provide information that is crucial for theories of native speech processing. Moreover, it will show how to improve late learners' listening skills.
In informele gesprekken spreken moedertaalsprekers veel woorden maar half uit (bijvoorbeeld 'tuuk' voor 'natuurlijk'). Vreemde-taalleerders hebben daar grote moeite mee. Hoe leren zij gereduceerde uitspraakvarianten? Ik ga dit onderzoeken met Nederlanders en Spanjaarden die Frans of Engels leren.