Annotating disfluency in speech corpora

angleški

2026

letnik 14, številka 1

govorjeni jezik   ponavljanje   prozodija   samopopravljanje   zapolnjeni govor

Univerza v Ljubljani, Filozofska fakulteta
(Obvezni izvod spletne publikacije)

The paper investigates disfluencies in speech, defined as phenomena that typi-cally disrupt the flow of speech and do not contribute to its primary proposi-tional content. These phenomena typically include filled and silent pauses, self-repairs, repetitions, and related features. Rather than being treated as errors, disfluencies are conceptualized as an integral component of spoken interac-tion. A review of the most influential annotation schemes reveals that no single framework has been established as a generally accepted standard. Combining an inductive, data-driven approach with a critical analysis of existing catego-rizations, the study proposes a tailored, multi-layer annotation scheme. This scheme covers a broad range of phenomena previously identified as disfluen-cies in earlier frameworks. The evaluation demonstrates that the scheme is suf-ficiently clear and unambiguous, offers well-defined and appropriate catego-ries, and is comparable to other contemporary schemes. The study contributes to the development of spoken-language resources for Slovenian and provides a foundation for future linguistic, psycholinguistic, and technological research.

19.10.2026