You Shall Know a Word by the Company It Keeps (quoteinvestigator.com)

Quote Investigator traced the adage "You shall know a word by the company it keeps" to linguist J. R. Firth’s 1957 essay "A Synopsis of Linguistic Theory," though the idea has deep antecedents — from the Latin legal maxim noscitur a sociis and 19th‑century jurisprudence to a loose echo in Euripides. Subsequent scholars and popular writers (A. H. Schutz, Melanie Mitchell) and outlets (The Economist) have reiterated Firth’s phrasing while documenting its intellectual lineage, so QI credits Firth with the modern formulation but situates it in a long tradition of meaning-from-context reasoning. For AI/ML the saying encapsulates the distributional hypothesis that underpins most modern NLP: a word’s meaning is derived from its co‑occurrence patterns. This principle motivated word embeddings and semantic vectors and scales up in today's foundation models via self‑supervised learning — e.g., masking words and predicting them from surrounding context across billions of tokens. Practically, that "company" approach gives models powerful statistical semantic representations but also explains limits: meaning emerges from text co‑occurrence rather than grounded, sensory experience, with implications for bias, generalization, and interpretability in downstream AI systems.