Um and Uh, and the expression of stance in conversational speech

Abstract : Chapter 1 introduces the dissertation, establishes the research questions and the methodology, questions the stakes of studying the markers um and uh, and lays out the study organization. Chapter 2 defines the main types of disfluencies, clinical and naturally occurring, summarizes the state of the art on the topic, and presents the different positions on their discourse role. Chapter 3 establishes the challenges regarding the fillers um and uh and summarizes studies that support the idea of different pragmatic and functional roles, suggesting that they are markers rather than just fillers. Chapter 4 introduces the concept of ÔstanceÕ (i.e., evaluation, opinion), establishes the definition used in this study and in the ATAROS corpus, and briefly summarizes the state of the art on automatic stance recognition in spoken speech. Chapter 5 introduces the two corpora used in this dissertation, ATAROS and Switchboard (SWB), and establishes their contribution. This chapter presents the methodologies for the annotations, the two versions of SWB, as well as the methodology adopted to construct an interoperability between the corpora to analyze um and uh. Chapter 6 analyzes the distribution and the duration of the two markers in SWB and ATAROS depending on speaker and dyad gender, on the conversationÕs naturalness, and on speaker participation. This chapter shows that um and uh are different from each other, that they have different distribution and duration cues depending on the variables, and therefore indicates that they are not used randomly. Chapter 7 focuses on the production of um and uh in SWB, and on the perception of the two markers by comparing two transcription versions of the corpus. The results of this chapter show that um and uh are more often missed than other frequent words such as function words, and that SWB transcribers make more transcription errors on uh than on um, suggesting that um plays a more important role in discourse than uh. Chapter 8 investigates the relationship between stance and the presence and the position of um and uh in an utterance, and reveals that the presence and the position of the two markers is dependent with stance. Chapter 9 looks at the relationship between stance and the acoustic realization of the vowel of the markers, compared to the vowel of other monosyllabic words. The results indicate that the stance values affect the vowel realization to different extents. Chapter 10 consists of a classification experiment that incorporates the findings from previous experiments to find out which features pertinent to um and uh (lexical, position, and acoustics) improve the systemÕs performance. The experiments show that the features associated to the two markers impact the systemÕs performance and that the best results are obtained when the word unigrams um and uh are not filtered, and when their position is included. The results also indicate that different acoustic features improve the scores. Chapter 11 concludes the dissertation by summarizing the results from chapters 6 through 10, underlying the impact of this study, and addressing the future directions of this project.
Document type :
Complete list of metadatas

Cited literature [20 references]  Display  Hide  Download
Contributor : Abes Star <>
Submitted on : Friday, March 15, 2019 - 2:15:08 PM
Last modification on : Thursday, April 4, 2019 - 1:24:08 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02069026, version 1



Esther Le Grézause. Um and Uh, and the expression of stance in conversational speech. Linguistics. Université Sorbonne Paris Cité; University of Washington, 2017. English. ⟨NNT : 2017USPCC149⟩. ⟨tel-02069026⟩



Record views


Files downloads