“Pooling” is just aggregation methods. It could mean taking max or average values, or more exotic methods like attention pooling. It’s meant to reduce the one-per-token dimensionality to one per passage or document.