kehaosakest

kehaosakest is a metric used in computational linguistics and digital humanities to quantify the degree of semantic divergence between two corpora over a specified time period. The term was coined in the early 2020s by researchers at the Institute for Language Technology in Zurich as an extension of the well‑known KL‑divergence framework, with a particular focus on tracking the evolution of word senses in genre‑specific texts. The name combines the Greek root *kehou*, meaning “to divide,” with the suffix *sakest*, derived from an abbreviation of “semantic aspect cross‑entropy.”

Technically, a kehaosakest score is computed by first extracting frequency vectors of lemmas from two corpora, smoothing them with a Laplace estimator, and then calculating the Jensen-Shannon divergence between the resulting probability distributions. The resulting value is normalized to fall between 0 and 1, where values closer to 0 indicate highly similar semantic profiles and values near 1 reflect substantial divergence. Unlike traditional measures that focus on overall lexical similarity, kehaosakest emphasizes shifts in meaning rather than changes in usage volume alone.

Key applications include diachronic studies of literary styles, monitoring policy language in governmental documents, and assessing the semantic impact of translation in multilingual corpora. The metric has also been incorporated into several open‑source linguistic analysis libraries, allowing researchers to perform large‑scale comparative studies.

Related concepts are semantic drift, lexical change detection, and the use of word embeddings to capture contextual shifts. While the framework remains a relative novelty, early adopters report that kehaosakest provides a more fine‑grained understanding of meaning evolution compared to traditional divergence measures.