Abstract
The degree of oral fluency of a non-native (L2) speaker is an important measure in assessing language proficiency. Previous studies have analysed listeners' subjective ratings and have attempted to relate these ratings to objective acoustic measurements of the stimuli. Across these studies, however, there is much diversity in the instructions given to raters, even though it is unknown what role these instructions play. For example, instructions to rate fluency by listening for pauses may influence raters to such an extent, that they only attend to the pauses in the speech while disregarding other cues of oral fluency. In this manner, research aiming to relate perceived fluency to measurable speech phenomena runs the risk of circularity. In our experiment, we explicitly manipulated the instructions provided to raters in order to answer three research questions:
a) To what extent are listeners capable of rating breakdown fluency, speed fluency, and repair fluency separately?
b) Which acoustic correlates contribute to each type of fluency rating?
c) Which acoustic correlates contribute to ratings of overall fluency?
Four groups of non-expert raters (n = 20 in each group) assessed the same set of L2 Dutch speech materials. One group received instructions to rate overall fluency as the sum of silent and filled pauses (the acoustic correlates of breakdown fluency), speech rate (the acoustic correlates of speed fluency), and corrections and hesitations (the acoustic correlates of repair fluency). Each of the other groups was instructed to attend to only one type of these acoustic correlates (i.e. to pauses, to speech rate, or to corrections and hesitations). The various fluency ratings are related to each other and to the objective acoustic measurements of the speech stimuli. The findings of this correlation study will be relevant for fluency perception studies, and for (second) language testing in general.
