- Perception: audition (vision for sign language) to phonology, meaning to all other perception
- Action: phonology to articulation, all other action to meaning
- Affect
- Brain states
- Events in the world
- Interaction with other agents
- Practical reasons: without grounding, no robot can talk to or understand us.
- Theoretical reasons
- Without grounding, our models depend on us for their interpretation.
- Without grounding, we have no account of how meaning gets into the system.
- Visual preprocessing + supervised learning of spatial maps with words
- Unsupervised learning of visual categories + unsupervised learning of "acoustic" categories + supervised learning of association of visual and acoustic categories
- Relate semantics to natural states and trajectories that emerge in certain types of dynamical systems
- Onomatopoeia: linguistic form resembles its meaning acoustically
- Articulatory iconicity: production of a linguistic form looks or feels like some aspect of its meaning