When a reader encounters “The horse raced past the barn fell,” the sentence seems to break at “fell.” The initial parse — “the horse raced” as a complete clause — must be abandoned in favor of a reduced relative clause: the horse that was raced past the barn fell. This is the garden-path effect: an early commitment to one structural interpretation makes a later word unexpectedly difficult.
Maina-Kilaas and Levy (arXiv:2603.11412) show that garden-path effects are a mathematical consequence of finite computational budgets, not properties of the sentences themselves. A particle filter maintains a finite set of structural hypotheses about the sentence being read. When the filter has many particles — many parallel hypotheses — ambiguous sentences are handled smoothly, because the correct parse survives in some particle even when the dominant parse is wrong. When the filter has few particles, the correct parse gets eliminated during resampling, and recovering it at the disambiguation point becomes expensive. The garden-path effect scales inversely with particle count.
A fully parallel processor — one maintaining every possible parse simultaneously — predicts no garden-path effect at all. Every structural ambiguity is resolved without cost because the correct interpretation was always available. The processing difficulty that humans experience at “fell” is not caused by the ambiguity of the sentence. It is caused by the finiteness of the processor encountering the ambiguity.
The structural point is that reading difficulty is a property of the reader's computational budget, not of the text's information content. Surprisal theory treats difficulty as determined by the probability of the next word given context. Particle filter models add a second factor: the number of hypotheses the reader can maintain. Same sentence, same surprisal, different processing cost — depending on how many parallel interpretations fit in the processor. The bottleneck is not what the sentence says. It is how many things the reader can simultaneously believe it might be saying.