The garden-path effect has been observed when incremental parsing gives rise to temporary misanalyses, but recovery mechanisms are still being debated. Dominant models—serial reanalysis, constraint-satisfaction, limited-repair and good-enough processing—all typically assume retrodictive lookahead requiring retrospective revision of earlier structural commitments. We suggest that recovery involves only forward, monotonic incremental integration of new input into a single, evolving representational state and thus does not require backward projection or an event of discrete repair. This hypothesis was tested in two experiments using reduced-relative and NP/Z type ambiguities. Experiment 1 (self-paced reading, N=64) and Experiment 2 (eye-tracking, N=64) contrasted ambiguous with unambiguous controls. Measures included first-pass, go-past, and total reading times; regression probabilities/path durations; and accuracy on comprehension probes targeting final interpretations as well as initial misparses. There were no significant differences in regression rates, go-past/total times or durations of regression-paths at disambiguating regions (ps >. 50; BF₀₁ > 5). The accuracy of comprehension was similarly high across conditions, with no lingering misinterpretations. Only a brief, verb-bias-modulated cost was evident at the ambiguous verb, fully resolving with disambiguation. These results reject retrodictive mechanisms in dominant narratives and validate a parsimonious, purely incremental model. This perspective offers advantages of greater theoretical economy, computational tractability, and improved compatibility with concepts popular in predictive processing that have not yet been sufficiently reconciled with ambiguity-resolution research.

Keywords: sentence parsing, garden-path recovery, incremental integration, reanalysis, eye-tracking, self-paced reading.