Linear versus hierarchical agreement feature processing in comprehension

Two experiments examined whether syntactic number features are tracked during comprehension with a linear slot-based memory system or with a hierarchical feature-passing system. In a construction such as The pond near the trail(s) for the horse(s) was ..., a linear account of subject-number tracking predicts greater interference from horses (N3), whereas a hierarchical account predicts greater interference from trails (N2). Experiment 1 used singular-head subject noun phrases (e.g., pond) and showed equal interference from N2 and N3, failing to differentiate between linear and hierarchical accounts. Experiment 2 used plural-head subjects and revealed more interference from N2 than N3. The pattern across the experiments accords with the ideas that (1) feature-tracking is hierarchical (e.g., Vigliocco & Nicol, 1997), (2) plurals are marked (e.g., Eberhard, 1997), and (3) subject-number information decays across intervening number-marked elements.