First-pass parsing of VP-ellipsis

Lewis P. Shapiro¹ & Arild G. Hestvik ^2
1 San Diego State University, ² CUNY Graduate Center

shapiro@mail.sdsu.edu 

We argue that VP-ellipsis comprehension involves an early stage where only phrase structure, category labels, and syntactic indices of the antecedent VP are copied to the ellipsis site, but not the lexical material itself.  This explains temporally evanescent strict interpretations of reflexives in (1) (shown with priming by "policeman" at the ellipsis site [2] but not at the pre-ellipsis site [1]), both with normal transitive verbs as well as with inherently reflexive verbs (Shapiro & Hestvik,1995; Shapiro et al., 1998).

(1) The policeman defended/asserted himself, and the fireman [1] who was unhappy did ___ [2] too, according to someone who was there.

However, an alternative explanation for these findings rests on the fact that resolution of VP-ellipsis involves a search for a previously introduced antecedent VP, which itself stands in a predication relation to its own subject.  When this VP is identified, so is the predication relation including the first clause subject, and this alone "activates" the first clause "strict antecedent" subject at the ellipsis site.  A prediction differentiating this account from ours is that the distant subject should always be activated at the ellipsis site, irrespective of what is contained inside the VP.  Furthermore, since predication is symmetrical there should be priming by both subject and object when the verb is transitive.

These predictions were tested in a cross-modal lexical priming experiment using stimuli like (2).  According to the "predication" alternative, both the subject ("mailman") and the object ("tie") should be activated at the ellipsis site; according to our view, the subject should not be activated.

(2) The mailman bought a tie for Easter, and his brother, who was [1] playing volleyball, did ___ [2] too, according to the sales clerk.

Twenty normal listeners were randomly assigned to each probe position (pre-ellipsis, ellipsis) as a between-subject factor, and grammatical function of prime (subject vs. object) and probe type (control/related) were within-subject factors.  The results with mean RTs per condition are summarized in (3).  Statistical analyses (ANOVA; simple effects; protected paired comparisons) revealed that RTs to probes related to the object from the first clause (687 ms) were significantly faster than RTs to controls (727 ms) at the site of ellipsis only; e.g., F(1,38)=20.93, p<.001.  No significant effects were observed for the subject at either probe position.

In conclusion, the subject of the antecedent VP is not activated at the ellipsis site.  The priming observed with the object is due to the copied referential index of the antecedent NP, just like an anaphoric index may activate a strict reflexive antecedent.  This result allows us to maintain our account of the first-pass parsing effect of temporally ungrammatical interpretations, and also provides additional insight into the general process of VP-ellipsis interpretation.

Shapiro, L. & Hestvik, A. (1995).  On-line comprehension of VP-ellipsis: Syntactic reconstruction and semantic influence.  Journal of Psycholinguistic Research, 24.

Shapiro, L; Hestvik, A; Suzuki, E; & Garcia, R. (1998).  Verb properties and gap-filling in complex VP-ellipsis constructions.  Poster presented at the 11th Annual CUNY Conference on Human Sentence Processing, Rutgers University, New Brunswick, NJ.