Semantic Change Detection

Semantic change detection studies how word meanings, concepts, discourse, or observed scenes shift between time periods. In language, researchers compare period-specific word and contextual embeddings, usage and sense distributions, semantic neighborhoods, and networks, using historical corpora and benchmarks to evaluate change. In images and remote sensing, models compare paired or multi-temporal observations, often combining visual features with language guidance to identify what changed and what remained stable. A central challenge is distinguishing meaningful change from noise such as shifting data distributions, sparse historical evidence, or visual differences caused by viewpoint and registration; metrics and benchmarks do not yet capture every kind of change reliably.