User Manual Privacy Policy Disclaimer Contact us
  Advanced SearchBrowse





When is speech fluent? The relationship between acoustic speech properties and subjective fluency ratings

There are no MPG-Authors available
There are no locators available
Fulltext (public)
There are no public fulltexts available
Supplementary Material (public)
There is no public supplementary material available

Bosker, H. R. (2011). When is speech fluent? The relationship between acoustic speech properties and subjective fluency ratings. Talk presented at The Language Acquisition Group, Max Planck Institute for Psycholinguistics. Nijmegen, The Netherlands.

Cite as: http://hdl.handle.net/11858/00-001M-0000-0023-D7BB-5
The oral fluency level of an L2 speaker is often used as an important measure in assessing language proficiency. In order to improve the objectivity of such language tests, previous studies have attempted to determine the acoustic correlates of fluency (e.g., Cucchiarini et al. 2002). The results of such studies are difficult to interpret since many of these studies have used multifaceted and intercorrelated measures of speech. An example of such a measure is speech rate which is related to both the speed of articulation and the use of pauses. If we want to discern the separate contributions of speed and pausing to fluency judgments, more precise measures are necessary to reveal more subtleties in perceived fluency ratings. Also, we wanted to see to what extent the relationship between acoustic measures and fluency ratings is dependent on the sensitivity of listeners to such speech phenomena. Our experiment investigated fluency perception by first establishing what speech properties listeners are most sensitive to. Three groups of listeners rated the same set of L2 Dutch speech stimuli on either the use of (silent and filled) pauses, speed of delivery or the use of repairs (corrections and repetitions). Stimuli were 20-seconds excerpts from turns in a simulated discussion. Using linear mixed models the subjective ratings were modelled using non-confounded acoustic measures which only measured one of three aspects of fluency: pausing, speed or repair. Very explicit test instructions resulted in high interrater reliability. Of the three rater groups the ratings from the ‘pause group’ were best predicted by our linear mixed models as evidenced by high explained speaker variance. It is concluded that raters are most sensitive to the use of pauses in speech. A fourth group of listeners rated the same stimuli on overall fluency. Modelling these ratings using only pause measures as predictors already resulted in high explained speaker variance. It is concluded that pause measures are the best acoustic correlates of fluency. Our results will be related to previous literature and a recent follow-up experiment further investigating the relationship between silent pauses and fluency ratings in both L1 and L2 speech will be introduced.