In this study, we investigate quantitatively the use of Filled Pauses in Hebrew and compare them to the use of silence-based attributes and fluency-based attributes in task-oriented dialogues. Our aim is to explore whether the use of Filled Pauses in Hebrew supports either the filler-as-word hypothesis or the filler-as-pause hypothesis. We computed seven measures as independent variables, among them: Normalized rate of words, Filled Pauses rate, silent pauses rate, and articulation rate. Findings show that Filled Pauses occur twice as many as the most frequent words and on the other hand, Filled Pauses are far less frequent than the overall number of silent pauses. We did find that Filled Pauses are correlated with a sub-category of silent pauses. Our analysis supports that Filled Pauses used by Hebrew speakers are closer to the filler-as-pause hypothesis. Further, speakers would use FPs differently depending on their role in the Map Task dialogues.
|Title of host publication||Speech and Computer - 23rd International Conference, SPECOM 2021, Proceedings|
|Editors||Alexey Karpov, Rodmonga Potapova|
|Publisher||Springer Science and Business Media Deutschland GmbH|
|Number of pages||11|
|State||Published - 2021|
|Event||23rd International Conference on Speech and Computer, SPECOM 2021 - Virtual, Online|
Duration: 27 Sep 2021 → 30 Sep 2021
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Conference||23rd International Conference on Speech and Computer, SPECOM 2021|
|Period||27/09/21 → 30/09/21|
Bibliographical noteFunding Information:
Acknowledgements. This work was supported by the Open Media and Information Lab at The Open University of Israel [Grant Number 20184]. The authors wish to thank Mr. Eyal Rabin for his valuable assistance with the statistical tests.
© 2021, Springer Nature Switzerland AG.
- Filled pauses
- Map task corpus
- Silent pauses
- Speaker’s role