Zipf in conversation

Frequency/rank distributions of tokenized items (‘words’) and recurring turn formats in conversational corpora with at least 20 such turn formats, representing 22 languages (8 phyla). Tokenized items (blue) show a linear frequency/rank relation in log/log space. Recurring turn formats (whether one-word ○ or multi-word +) appear to obey a similar frequency/rank distribution for the 20% of turns that occur >20 times (purple), tapering off towards lower frequencies and unique turns (grey). Fit fluctuates with corpus size and the parallelism of distributions is most apparent in larger corpora.

Dingemanse, M., & Liesenfeld, A. (2022). From text to talk: Harnessing conversational corpora for humane and diversity-aware language technology. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 5614–5633. doi: 10.18653/v1/2022.acl-long.385 PDF