The Dynamics of Human and AI-Generated Language: How Semantic Content Fluctuates Across Different Timescales
Computer Speech & Language · 2027-01 · 101 · 102013
DOI: 10.1016/j.csl.2026.102013 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer Speech & Language, 2027-01. Verfasst von Han-Jen Chang und 7 weiteren Autor:innen. Die Arbeit nennt 101 Literaturangaben. Gefördert von: Wellcome Trust, Fondo de Fomento al Desarrollo Científico y Tecnológico, Fondo Nacional de Desarrollo Científico y Tecnológico.
Abstract (Originalsprache)
Spoken language, whether produced by humans or large language models (LLM), unfolds over time with varying semantic content. However, we still lack simple, interpretable time-series features that capture how generic versus specific content is distributed over time, and that can be used to compare human and AI-generated speech. We introduce a semantic-timescale analysis pipeline that turns word-level transcripts with timestamps into semantic time-series. For each spoken narrative, we compute (i) semantic specificity using WordNet-based word depth and (ii) contextual similarity using SBERT embeddings and quantify their temporal dependence using autocorrelation-window measures (ACW-0 and related metrics). We then compare original speech to multiple shuffled controls that selectively disrupt lexical identity, temporal order, and word duration. Across human-read autobiographical narratives, TTS readings, and LLM-generated texts rendered with TTS, we find that segments with longer ACW-0 in the semantic time-series tend to contain more generic vocabulary, whereas segments with shorter ACW-0 are enriched in more specific words. These associations are strongly attenuated or abolished when word order and timing are randomized, indicating that ACW-based measures capture non-trivial temporal organization of semantic content beyond static lexical distributions. Our results suggest that ACW-based semantic timescales are a useful family of features for analyzing and comparing the temporal structure of human and AI-generated speech.
Förderung
- Wellcome Trust
- Fondo de Fomento al Desarrollo Científico y Tecnológico
- Fondo Nacional de Desarrollo Científico y Tecnológico
- National Institute on Aging
- National Institutes of Health
- Fogarty International Center
- Alzheimer's Association
- Rainwater Charitable Foundation
Zitieren
APA
Chang, H., Çatal, Y., Wolman, A., Ibáñez, A., Smith, D., Su, I., Cheng, K., & Northoff, G. (2027). The Dynamics of Human and AI-Generated Language: How Semantic Content Fluctuates Across Different Timescales. Computer Speech & Language, 101, 102013. https://doi.org/10.1016/j.csl.2026.102013
Vancouver
Chang H, Çatal Y, Wolman A, Ibáñez A, Smith D, Su I, et al. The Dynamics of Human and AI-Generated Language: How Semantic Content Fluctuates Across Different Timescales. Computer Speech & Language. 2027;101:102013. doi:10.1016/j.csl.2026.102013
BibTeX
@article{Chang2027,
author = {Chang, Han-Jen and Çatal, Yasir and Wolman, Angelika and Ibáñez, Agustín and Smith, David and Su, I-Wen and Cheng, Kai-Yuan and Northoff, Georg},
title = {The Dynamics of Human and AI-Generated Language: How Semantic Content Fluctuates Across Different Timescales},
journal = {Computer Speech & Language},
year = {2027},
volume = {101},
pages = {102013},
doi = {10.1016/j.csl.2026.102013}
}
EUVIMED