Dependency distance is a key measure of syntactic complexity and processing constraints, but as a scalar cannot capture how information is distributed between a head and dependent. We analyze dependency spans—the endpoints and intervening words—as position-aligned units. Interveners lie outside the binary dependency but constitute its sequential processing context. Using Universal Dependencies treebanks and XGLM-2.9B, we estimated word-level surprisal for distances 4–10 across 22 languages, yielding 154 mean curves. Dynamic time warping and clustering identified an approximately monotonic decline and a nonmonotonic contour with an initial decline, stable middle, and final rise. Membership was stable across distances in 20 languages; Russian had three Type 1 and four Type 2 curves, whereas German had six Type 1 and one Type 2 curve. Segmented models favored three stages for both types, differing mainly in the final stage. Principal component analyses revealed a more dispersed latent structure in the middle stage and concentration around fewer variables at the edges. The final-stage contrast was associated with dependency direction and verb roles. Dependency-span analysis thus complements dependency syntax with a position-sensitive account of linear realization, revealing cross-linguistic information patterns that dependency length alone cannot recover.