CNL Lunch

Nina Kazanina & Colin Phillips

ERP Evidence on the Time Course of Processing Demands in Wh-Dependencies

Thursday March 1st, 12:30pm, 3416 Marie Mount Hall

This study uses ERP measures to investigate the time-course of processing demands in sentence comprehension. The study focuses on the processing of sentences involving wh-dependencies, spanning one or two clauses. It is known that longer structural dependencies between words and phrases entail increased processing demands, but the source of this length effect is unclear. On the one hand, the length effect may be due to increased difficulty in storing long incomplete structural dependencies (e.g. holding a wh-phrase in memory, before its thematic role assigner has been identified). On the other hand, the length effect may be due to increased difficulty in the integration of material at the completion of long structural dependencies (e.g. creating the structural relationship between a wh-phase and a verb). Most existing behavioral evidence for the length effect can be explained using either of these mechanisms (Gibson, 1998).

These competing accounts of the length effect make different predictions about the time course of resource demands, and can be distinguished using ERP measures. Our experiment tested 4 conditions: a short wh-dependency condition (1-clause extraction, (1b)), a long wh-dependency condition (2-clause extraction, (1d)), and control conditions which lacked wh-dependencies (1a,c).

1.a. The patients thought that the nurse knew that the suspicious visitor had stolen the drugs from the storeroom.
b. The patients thought that the nurse knew which drugs the suspicious visitor had stolen from the storeroom.
c. The nurse knew that the patients thought that the suspicious visitor had stolen the charts drugs from the storeroom.
d. The nurse knew which drugs the patients thought that the suspicious visitor had stolen from the storeroom.

28 participants read 40 sentences from each of the 4 conditions, in a Latin Square design, interspersed with 320 filler sentences, spread across two recording sessions. Sentences were presented in a RSVP paradigm, with a 500ms/word presentation rate. All sentences were followed by a comprehension question. Grand average ERPs were collected for each word, in order to track the temporal evolution of the dependency-length effect.

The study investigated the contribution to the length effect of both (i) the Left Anterior Negativity (LAN), an ERP component associated with storage of structural dependencies (e.g. King & Kutas, 1995; Fiebach et al. 2000), and (ii) the P600, an ERP component that has been associated with the integration of words at the completion of syntactic dependencies (e.g. Kaan et al. 2000).

In the interval while the incomplete wh-dependency was held in memory a slow wave LAN response was observed, which was significantly different from the responses in the control conditions (p<0.05). Although the contribution of each word to the slow LAN was equivalent in the short-wh and long-wh conditions, the greater length of the long wh-dependencies resulted in a higher amplitude overall LAN response. Although no P600 response was observed at the verb at the completion of the wh-dependency, the widely distributed posterior positivity characteristic of the P600 was observed at the immediately following word (e.g. 'from' in (1a-d)). This effect was stronger in the long-wh than in the short-wh conditions. We conclude from these results that, at least for the case of English wh-dependencies, the demands of both storage and integration processes contribute to the dependency-length effect.