International Research Training Group
Language Technology
&
Cognitive Systems
Saarland University University of Edinburgh
 

The filler may be a gentleman: local coherence effect in long-distance dependency processing

Speaker: Zhenguang Cai

Institution: University of Edinburgh

Abstract:

In the literature of long-distance dependency processing, it is relatively uncontroversial that the parser integrates a filler into the string as soon as possible1, 2. Also, it has been demonstrated that the parser is reluctant to reanalyze3. However, a recent study4 shows that the parser tends to reanalyze its initial analysis when there is a locally coherent, though globally incompatible, structure, e.g., reanalyze the object the player in "The coach smiled at the player tossed a frisbee" as the subject of tossed. The present study presents a similar local coherence effect in long-distance dependency processing. We examined sentences like (1a), where the filler John is the object of the verb threatened, and the NP the neighbour is the subject of the main clause. According to current theories, the filler John would be integrated as the object as soon as the verb threatened is encountered, and the neighbour would be treated as the subject of a new clause and would not be considered as the object of the preceding verb. We conducted a self-paced reading experiment (48 subjects, 24 items) with a 2 (construction: the ambiguous object cleft (1a,b) vs the unambiguous subject cleft (1c,d)) * 2 (comma) design. We predicted that if the parser considers the neighour as the object of threatened, there will be a filled-gap effect at the region the neighbour in (1a) but not in the other conditions. ANOVAs show that at the neighbour there is an interaction between construction and comma (p1=.077, p2=.009). The interaction is driven by longer RTs in (1a) (849 ms) than in (1c) (755 ms) (p1=.04, p2=.013); there is no significant difference between (1b) (709 ms) and (1d) ) (725 ms) (Fs<1). This indicates that the parser considers the post-verb NP as the object, resulting in a conflict between the post-verb NP and the filler. We are currently investigating whether the parser adopts the filler-object analysis (initial analysis) or the local-object analysis (reanalysis), and what effects filler-gap distance and plausibility have.

  1. Because it was John that Ralph threatened the neighbour recorded their conversation.
  2. Because it was John that Ralph threatened, the neighbour recorded their conversation.
  3. Because it was Ralph that threatened John the neighbour recorded their conversation.
  4. Because it was Ralph that threatened John, the neighbour recorded their conversation.

References

  1. Frazier, L, & Flores d'Arcais, GB. (1989). JML, 28:331-344.
  2. Pickering, MJ. (1994). JPR, 23:323-352.
  3. Sturt, P, Pickering, MJ, Scheepers, C, & Crocker, MW. (2001). JML, 45:283-307.
  4. Tabor, W, Galantucci, B, & Richardson, D. (2004). JML, 50:355-370.

Last modified: Sat, Aug 09, 2008 01:48:20 by

Valid HTML 4.01 Transitional Valid CSS!