Questions: Garden-Path Sentences and Syntactic Parsing

Question 1

A reader slows down sharply at the word 'fell' in 'The horse raced past the barn fell.' What best explains this reading difficulty?

• A. 'Fell' is a low-frequency word that requires extra time to retrieve from the mental lexicon
• B. The parser committed to an active-voice parse ('a horse was racing') before seeing 'fell,' and must now reanalyze the sentence as a reduced relative clause
• C. The sentence violates English grammar rules, and the parser flags it as unacceptable
• D. The parser suspended judgment about structure until 'fell' arrived, then had to build the parse from scratch

Answer: (B)

The garden-path effect is purely structural. Active constructions are statistically far more common than reduced passive relative clauses, so the incremental parser committed to an active reading. 'Fell' cannot be the second main verb of an already-complete sentence, forcing costly reanalysis. Option A confuses lexical difficulty with structural difficulty; option C is wrong because the sentence is grammatically well-formed; option D gets the architecture backwards — the parser commits early, not late.

Question 2

Which evidence most directly supports the claim that the parser is 'incremental' rather than waiting for the full sentence before assigning structure?

• A. People find garden-path sentences harder to understand than semantically unusual but structurally simple sentences
• B. EEG studies show a P600 response — a neural signature of syntactic reanalysis — arising several hundred milliseconds after the disambiguating word arrives
• C. Garden-path sentences are rarely produced by native speakers
• D. Re-reading a garden-path sentence eliminates all comprehension difficulty

Answer: (B)

The P600 is a neural event-related potential that occurs around 600ms after the disambiguating word — precisely when reanalysis is triggered. Its timing shows that the parser had already committed to a structure before the disambiguating word arrived, then detected a conflict requiring repair. This is direct neural evidence for early, incremental commitment. Option A could be explained by many factors; option D is false — re-reading helps but doesn't fully eliminate difficulty.

Question 3

Garden-path sentences are initially difficult because the parser commits to a parse based on statistical frequency, not just grammatical rules.

• T. True
• F. False

Answer: True

This is the core insight. Both constraint-based and serial models of parsing agree that frequency-based statistics shape early parsing commitments. Active voice is more frequent than reduced relative clauses; transitive uses of verbs are more frequent than some intransitive constructions. When the statistically favored parse is wrong, reanalysis is required. Grammar rules alone do not determine which parse is chosen first — frequency does.

Question 4

Garden-path sentences are ungrammatical — they violate English syntax — which is why comprehension fails.

• T. True
• F. False

Answer: False

Garden-path sentences are fully grammatical. 'The horse raced past the barn fell' is syntactically correct — it is a reduced relative clause ('the horse [that was] raced past the barn fell'). The difficulty is not grammatical unacceptability but the need to abandon an initially committed parse and assign a new one. If garden-path sentences were ungrammatical, re-reading would not make them comprehensible.

Question 5

Why do garden-path sentences 'work' — why does comprehension fail initially and succeed upon re-reading? What does this reveal about the parser's normal operation?

Model answer: On first reading, the parser incrementally commits to the most statistically common parse at each word, without waiting for the full sentence. When a later word is incompatible with the committed parse, reanalysis is required — which is cognitively costly. On re-reading, you already know the ending, so you approach the ambiguous region with the correct commitment and parse smoothly. This reveals that the parser normally bets early and confidently on likely structures rather than maintaining multiple interpretations in parallel.

The re-reading asymmetry is key evidence: if parsing were simply about decoding fixed grammar, re-reading wouldn't matter. The fact that re-reading dramatically helps shows that the initial difficulty was about structural commitment, not linguistic knowledge. The parser's incremental, frequency-guided design is normally efficient — most sentences confirm the initial parse — but garden-path sentences expose the cost of this efficiency when the initial bet is wrong.