In a stable theory, if p, q ∈ S(A) are two types over the parameters A, then we say that p and q are almost orthogonal, writting p ⊥^a q if whenever a realizes p and b realizes q, then a ↓_A b. We say that the type p ∈ S(A) is orthogonal to the type q ∈ S(B) (written p ⊥ q) if for any superset C of A ∪ B and nonforking extensions p' ∈ S(C) of p and q' ∈ S(C) of q, one has p' ⊥^a q'

While the difference between ⊥ and ⊥^a may seem technical, it is real. For example, if G is the additive group of C one can form a two sorted structure in which each sort is interpreted by G, on the first sort one has the full group structure, but the only other nonlogical symbol is function expressing the regular action of G on the second copy. The generic types the two sorts are almost orthogonal, but not orthogonal.

Hrushovski showed that the above example is actually general. That is, any instance of almost orthogonality, but nonorthogonality is explained by the action of a type-definable group on a type-definable set.