Suzanne Belanger & Ron Smyth
University of Toronto
We report on a replication/extension experiment investigating whether deep and surface anaphors access different levels of representation during resolution. Using items such as in (1), Tanenhaus, Carlson & Seidenberg (1985) reasoned that if an anaphor accessed a surface representation, then response times to the verification sentence should increase in the Mismatch condition, because the original order would be kept active, causing interference. They found that only surface anaphors resulted in this increase, and concluded that deep anaphors access a conceptual level. We suggest that a full interpretation of their results is not possible without adequate control conditions. Specifically, it is not clear whether surface anaphors affect the linguistic representation by keeping it active or whether deep anaphors affect the linguistic representation through suppression (or both). The idea of suppression during anaphor resolution has been put forth by Lucas, Tanenhaus & Carlson (1990) who argue that resolution of NP-anaphors involves inhibition of inappropriate antecedents as opposed to reactivation of appropriate ones. We propose that inhibition might also apply to levels of representation during anaphor resolution. We ran an experiment intended to shed light on the activation/suppression issue. We devised two additional conditions, as illustrated in (2) and (3). The Null Condition was designed to test the level of activation of linguistic form immediately following Line 1 --- if Surface anaphors induce a reactivation of the antecedent to initial levels, then we expect these two conditions to pattern alike. The Neutral Condition was designed to test the level of activation in the context of normal decay of surface form. If the Deep condition is less affected, then we have some evidence for suppression; if the Surface condition is more affected, then we have some evidence of reactivation beyond what it would otherwise be. Preliminary results indicate a replication of Deep-Surface difference in the subjects analysis. The results also suggest that suppression is involved with deep anaphors, as this was the only condition not significantly affected by mismatch. Finally, reactivation appears to be involved with Surface anaphors, which were more affected than Neutral and just as affected as Null. We are currently running additional subjects to further clarify the data. We discuss verbal anaphors in the context of the literature on anaphor resolution, specifically the issues of reactivation and antecedent form. We suggest that more levels than simply 'linguistic' versus 'conceptual' need to be considered in order to account for the range of findings.
Examples
| (1) | Line 1 | Jenny asked Ann's boyfriend out. | [match] |
| or | Jenny asked out Ann's boyfriend. | [mismatch] | |
| Line 2 | Ann was furious that she did (it). | [surface (ellipsis), or deep (do it)] | |
| Verification: | Jenny asked Ann's boyfriend out. | ||
| (2) | Null Condition --- verification immediately followed Line 1 | ||
| (3) | Neutral Condition --- no verbal anaphor in Line 2 (no reactivation of original VP), e.g., | ||
|
Line 2 |
He suggested dinner and a movie. | ||
References
Lucas, M., Tanenhaus, M., & Carlson, G. (1990). Levels of representation in the interpretation of anaphoric reference and instrument reference. Memory & Cognition, 18:6, 611-631.
Tanenhaus, M., Carlson, g., & Seidenberg, M. (1985). Do listeners compute linguistic representations? In D. R. Dowty, L. Karttunen, & A. M. Zwicky (Eds)., Natural Language Parsing: Theoretical, Psychological and Computational Perspectives. Cambridge University Press