Modeling uncertainty in spontaneous speech: Lexical and acoustic features

Abstract

Previously, spoken uncertainty has been analyzed using either lexical or acoustic features, but few, if any, studies have used both feature types in combination. Therefore, it is unknown to what extent these feature types provide redundant information. Additionally, prior research has focused on the study of acoustical features of only single words, and it is unclear if those results can generalize to perceived uncertainty in spontaneous speech. The current study elicited spontaneous speech through a team dialogue task in which two people worked together to locate street-level pictures of different houses on an overhead map. The communications were recorded, transcribed, broken into utterances, and presented to 10 individuals who rated each utterance on a 5-pt Likert scale from 1 (very uncertain) to 5 (very certain). A large number of acoustic and lexical features from the literature were calculated for each utterance. Random forest classification (Breiman, 2001) was used to select features and then inves...