Disfluencies in simultaneous interpreting: a corpus-based analysis

This paper presents an analysis of disfluencies in the European Parliament Interpreting Corpus (EPIC), an electronic corpus of speeches in English, Italian and Spanish and their simultaneous interpretations into the same languages, in all possible combinations and directions. The focus of the paper is on two specific types of disfluencies, i.e. truncated words and mispronounced words, produced by both speakers and interpreters. After automatically extracting the relevant occurrences, their incidence in the nine sub-corpora making up EPIC was calculated. The main features of these disfluencies are described and their possible causes are discussed. The specific nature of EPIC enabled us to carry out both comparable and parallel analyses of the material, i.e. to compare disfluencies occurring in the sub-corpora of speeches originally delivered in Italian, English and Spanish with those in the interpreted speeches in the same language (comparable analysis); and to compare disfluencies in each sub-corpus of original speeches with those occurring in the corresponding two interpreted versions available in EPIC (parallel analysis). Several insightful results were obtained, including language pair-related trends.