Sentence processing research examines how the parser — the cognitive system responsible for syntactic analysis — operates in real time. Since sentences unfold word by word, the parser must build structural interpretations incrementally, often before disambiguating information is available. How it handles this ambiguity, and what happens when its initial analysis proves wrong, are central questions in psycholinguistics.
Garden-Path Sentences
Garden-path sentences (e.g., "The horse raced past the barn fell") initially lead the parser down an incorrect analysis that must be revised when contradicting information arrives. The difficulty readers experience at the disambiguating word ("fell") reveals the parser's initial commitment to a particular structure. Frazier's garden-path model proposed that the parser initially pursues the simplest structure (minimal attachment, late closure) and reanalyzes only when this fails.
Constraint-Based Models
Constraint-based models (MacDonald, Pearlmutter, Seidenberg) propose that the parser simultaneously considers multiple analyses weighted by all available constraints: word frequency, semantic plausibility, discourse context, and subcategorization preferences. Rather than committing to a single analysis, the system maintains a ranked set of alternatives. Ambiguity resolution involves the convergence of multiple probabilistic constraints rather than serial commitment and reanalysis.
Eye tracking has become the primary method for studying real-time sentence processing. Readers fixate longer on words that are syntactically ambiguous, semantically unexpected, or structurally complex. Regression eye movements (looking back to earlier words) indicate processing difficulty and reanalysis. The millisecond resolution of eye tracking reveals that syntactic, semantic, and discourse information all influence processing within the first fixation on a word.