Variation in non-fluencies in a corpus of simultaneous interpreting vs. non-interpreted English

The present study aims to investigate the occurrence of various non- fl uencies (mispronounced words and hesitations; self-repairs and editing terms; silent pauses; repetitions; fi llers) in interpreted and non-interpreted, spontaneously produced English. The material for the study is the English component of a parallel bidirectional corpus of Russian-English interpreting of political discourse, consisting of approx. 130,000 words from 77 speech events. The instances of non- fl uencies have been automatically extracted from the corpus, with the exception of self-repair, which was subject to manual annotation. The fi gures for the two subcorpora were compared using the Mann-Whitney U test, chi-square test, and Fisher ' s exact test, as appropriate. The results show that (1) interpreted English has more dis fl uencies overall, and serial truncations speci fi cally; (2) the number of repaired dis fl uencies is lower in interpreted English; (3) and interpreted English has fewer fi llers and dis fl uent repetitions than non-interpreted English. The results on editing terms are inconclusive. While the fi rst fi nding conforms to the predictions in the literature on SI, the other two can be ascribed to di ff erences in style among interpreters.