Identifying fluency parameters for a machine-learning-based automated interpreting assessment system

Fluency is an important yet difficult-to-measure criterion in interpreting assessment. This empirical study of English-Chinese consecutive interpreting aims to identify fluency parameters for a machine-learning-based automated assessment system. The main findings include: (a) empirical evidence supports the choice of the median values as the cut-offs for unfilled pauses and articulation rate; (b) it informs the selection of outliers as particularly long unfilled pauses, relatively long unfilled pauses, particularly slow articulation and relatively slow articulation; (c) number of filled pauses, number of unfilled pauses, number of relatively slow articulation, mean length of unfilled pauses, mean length of filled pauses can be chosen to build machine-learning models to predict interpreting fluency in future studies as they can explain the variance of established temporal measures and show stronger explanatory power than dependent variables when predicting scores. The study identifies assessment rubrics on an empirical basis and provides a methodological solution to automate the labour-intensive tasks in interpreting assessments.