Are types of filled pause discrete classes

Clark and Fox Tree hypothesize that filled pauses ‘‘uh’’ and ‘‘um’’ in English serve the discourse function of alerting the interlocutor of production difficulties. They demonstrate that the mean duration of pause immediately after ‘‘um’’ is significantly longer than after ‘‘uh.’’ This account suggests that fillers are categorically differentiated from each other; however, it fails to predict a cross‐linguistic generalization. Across languages, fillers tend to have centralized vowels [Vasilescu et al. (2005)]. When there are two realizations of filled pause within a language, one tends to consist of a vowel followed by a nasal, as in ‘‘um’’ [Clark and Fox Tree (2002)]. These observations suggest filled pause is a phonetic process and, therefore, may apply gradiently. English filled pauses are revisited, showing that the discourse effect identified is not categorically distributed, but rather defined acoustically on a gradient scale. Acoustic models of fillers are trained on the identities of ‘‘uh’’ and ‘‘...