WordPiece
Domain: NLP Fundamentals
🎯 Motivation
Like BPE but builds vocabulary for language understanding (BERT). Uses prefix for non-starting pieces.
📋 Concrete Example
Playing splits to [Play, ##ing]. The double-hash marks that ing attaches to the previous piece.
🔗 Analogy
Like labeling puzzle pieces - the piece with is a connector that must follow another piece.