I trained a 300‑dimensional CBOW word2vec model on 13 billion Russian fiction words (classic + modern). Lemmas only, no stop words. Load it with Gensim and explore literary semantics. Code, model, and examples included.
Can word2vec, trained on 5.4 million chess games, learn meaningful “semantics” of moves? This preprint applies distributional semantics to PGN game records, treating moves as words and games as sentences. The results are fascinating: the model groups moves by piece, distinguishes openings from endgames, detects castling, and even separates bishops by square color. But the author honestly admits: this won't help engines or humans choose better moves. Instead, it shows how vector representations capture structural patterns of the game itself – a beautiful example of “superlinguistics” in action.