Consonant clusters are subject to language-specific restrictions, not only on their composition (Russian allows /mr/ but English does not), but also on their temporal organisation. With Electromagnetic Articulography (EMA), one can see the co-ordination of movements of articulators such as tongue and lips. I illustrate the use of EMA with a case study of transfer of temporal organisation of clusers. Previous literature suggests that high overlap (the articulation of the second consonant starts during the first consonant; e.g., as in English or German) is not attained by speakers of low overlap languages (e.g., Georgian, Russian), who transfer the low overlap pattern. The transfer does not take place from high overlap to low overlap languages (Zsiga, 2000, 2003). We studied Georgian, a language with many complex consonant clusters and varying amounts of articulatory overlap between first and second consonant (Chitoran, Goldstein, and Byrd, 2002) and German. Native speakers of both languages produced their own (cluster-specific) timing patterns and were also elicited to produce the pattern of the other language.
We expected speakers to increase overlap after a few repetitions, but found that Georgian speakers settle on a level of overlap between their native pattern and the German pattern. German speakers, however, seem to imitate the low overlap pattern of Georgian but not the fine-grained differences. I will discuss the opportunities of EMA as a paradigm to study phonetic knowledge, the language-specific nature of consonant co-articulation patterns and implications for second language production and perception. I will also discuss a machine-learning model (Support Vector Machines) that shows that both groups are approaching each others’ productions when producing the non-native pattern.