Modeling language-learners' errors in understanding casual speech
Ten Bosch, L., Giezenaar, G., Boves, L., & Ernestus, M.
Modeling language-learners' errors in understanding casual speech. In G. Adda, V. Barbu Mititelu, J. Mariani, D. Tufiş, & I. Vasilescu (Eds.
), Errors by humans and machines in multimedia, multimodal, multilingual data processing. Proceedings of Errare 2015
(pp. 107-121). Bucharest: Editura Academiei Române.
In spontaneous conversations, words are often produced in reduced
form compared to formal careful speech. In English, for
instance, ’probably’ may be pronounced as ’poly’ and ’police’
as ’plice’. Reduced forms are very common, and native listeners
usually do not have any problems with interpreting these
reduced forms in context. Non-native listeners, however, have
great difficulties in comprehending reduced forms. In order
to investigate the problems in comprehension that non-native
listeners experience, a dictation experiment was conducted in
which sentences were presented auditorily to non-natives either
in full (unreduced) or reduced form. The types of errors made
by the L2 listeners reveal aspects of the cognitive processes underlying
this dictation task. In addition, we compare the errors
made by these human participants with the type of word errors
made by DIANA, a recently developed computational model of