Mechanisms behind accelerated adoption of ChatGPT-associated vocabulary in spoken academic communication

Determine the mechanisms that drive the accelerated adoption, in English-language academic YouTube speech transcripts, of vocabulary that is highly distinctive of ChatGPT-edited abstracts (e.g., “delve,” “realm,” “meticulous,” “adept”) following the release of ChatGPT.

Background

The study analyzes approximately 280,000 English-language YouTube video transcripts from academic institutions and finds significant shifts in the usage trends of words that are distinctive of ChatGPT-edited texts following ChatGPT’s release in November 2022. The acceleration is most pronounced for the top 20 words most associated with ChatGPT, indicating a specific influence rather than a uniform shift across all vocabulary.

The authors observe that adoption occurs in both spontaneous speech and readings of prepared manuscripts but that the trend is not universal (e.g., words like “groundbreaking” and “underscore” did not accelerate). This suggests that multiple factors may govern the adoption process, motivating the need to identify the causal mechanisms underlying the observed acceleration in usage of particular ChatGPT-associated words.