ChronoLens: Measuring Language Change Across Time, Languages, and Linguistic Levels

ChronoLens introduces a shared analytical framework for comparing historical language change across morphology, syntax, semantics, and pragmatics. It combines frozen multilingual language models, feature-aligned crosscoders, and post-hoc linguistic interventions, and is applied to 44.98 million documents, totaling approximately 17.2 billion tokens, from five parliamentary traditions between 1803 and 2026. The resulting sparse representations correlate more strongly with linguistic statistics than dense embeddings or a pooled sparse autoencoder, with reported Spearman correlations of 0.72 versus 0.29 and 0.28. The study finds broadly comparable change magnitudes within languages, but substantial differences in timing, extent, and direction across languages.