Given utterance, infer state, and then given utterance and state, infer goals

Predicted state given utterance

Same across face threat presence. Replicated findings for positive utterances. “not amazing” is below average. Should try running this with no threat scenario (not concerning evaluation of some kind)

Goal inferences

Replicated findings for positive utterance.

Negative utterance for: Polite: Always sloping downward. Highest likelihood for being polite when actual state is terrible/bad.

Informative: “not terrible/bad”: highest informativity when state is 3 “not okay/good”: highest informativity when state is 1 “not amazing”: highest informativity when state is 2 (definitely higher than state=1)

Compare threat vs. no threat on goal prob inferences

no diff between threat type for positive utterances. some subtle differences for neg utterances? differences in slopes for informative “terrible/bad”, polite “amazing”