Pauses as an Indicator of Psycholinguistically Valid Multi-Word Expressions (MWEs)?

In this paper we investigate the role of the placement of pauses in automatically extracted multi-word expression (MWE) candidates from a learner corpus. The aim is to explore whether the analysis of pauses might be useful in the validation of these candidates as MWEs. The study is based on the assumption advanced in the area of psycholinguistics that MWEs are stored holistically in the mental lexicon and are therefore produced without pauses in naturally occurring discourse. Automatic MWE extraction methods are unable to capture the criterion of holistic storage and instead rely on statistics and raw frequency in the identification of MWE candidates. In this study we explore the possibility of a combination of the two approaches. We report on a study in which we analyse the placement of pauses in various instances of two very frequent automatically extracted MWE candidates from a learner corpus, i.e. the n-grams I don’t know and I think I. Intuitively, they are judged differently in terms of holistic storage. Our study explores whether pause analysis can be used as an objective empirical criterion to support this intuition. A corpus of interview data of language learners of English forms the basis of this study.