Search Abstracts | Symposia | Slide Sessions | Poster Sessions
How continuous is continuous speech?
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Jannika Merle Alena Hollmann1, Cosimo Iaia1,2, Alessandro Tavano1,2; 1Goethe University Frankfurt, 2Cooperative Brain Imaging Center - CoBIC
Research on speech and language often assumes that speech unfolds as a continuous signal, yet there is limited empirical evidence on how continuous speech actually is at the word-to-word level. Here, we examined the temporal structure of speech discontinuities in two openly available sets of audiobook excerpts of Alice in Wonderland by Lewis Carroll, one in English and one in German. We defined discontinuous speech as any silent interval between the acoustic offset of one word and the onset of the following word with a duration greater than 0.001 s. We analyzed these discontinuities with respect to their linguistic position, focusing on pauses occurring within and between sentences, as well as between words, phrases, and sentences. For the English audiobook excerpt, speech discontinuities showed a trimodal distribution, with peaks around 0.1 s, 0.35 s, and 0.75 s. The peaks at 0.1 s and 0.35 s were present for breaks between both words and phrases, and were more pronounced at phrase boundaries than at word boundaries. In contrast, the peak at 0.75 s was observed only for breaks between words. These findings suggest that “continuous” speech contains systematic and concurrent discontinuities at multiple temporal scales, and that these discontinuities are partly structured within linguistic boundaries. The literature proposes a boundary at 0.25 s of breaks, with breaks < 0.25 s maybe representing phono-articulation purposes and breaks > 0.25 s maybe representing more complex cognitive processes, like e.g word finding processes and selective retrieval. This is reflected in our data as the breaks of 0.75 s were only pronounced between words. However, the peak at 0.1 s was also pronounced for breaks between phrases, suggesting that directly clustering long and short pauses in cognitive vs. articulatory processes may miss the mark. Our results challenge the assumption, common in the neural speech tracking literature, that speech can be treated as a uniformly continuous signal. Measurements built on that heuristic, such as speech rate, can be misleading if speech is indeed discontinuous. Even read aloud speech - which typically presents with fewer disfluencies than conversational speech - is organized by recurrent pause durations that reflect concurrent levels of linguistic and prosodic organization. Future work will examine whether similar temporal patterns are observed in conversational speech, where planning demands, interactional turn-taking, and spontaneous production may shape speech discontinuities differently.
Topic Areas: Language Production,