Expression constraints in multimodal human-computer interaction