Ellipsis with garden-path antecedents in French

Ellipsis with garden-path antecedents in French

Dario Paape, Shravan Vasishth (Univ. of Potsdam) & Barbara Hemforth (Univ. Paris Diderot)

paape@uni-potsdam.de

Ellipsis is typically anaphoric, and its interpretation requires access to a suitable antecedent. Adopting the framework of cue-based retrieval parsing (Lewis & Vasishth, 2005), Paape (2015) suggests that the syntactic structures of ellipsis antecedents are stored in working memory as chunks whose activation levels fluctuate over time. Cue-based retrieval parsing predicts that manipulating memory chunks increases their activation, making them easier to retrieve.

It should be possible to boost the activation of a stored antecedent through syntactic reanalysis, which changes the structure of the chunk and thereby 'refreshes' its memory trace. In a self-paced reading study on German sluicing, Paape (2015) found that antecedents that had been reanalyzed (i.e., garden-path antecedents) led to reduced reading times at the ellipsis site, indicating facilitated retrieval and thus supporting the reactivation hypothesis.

In the current study, we attempted to replicate Paape's (2015) finding using eye-tracking, which more closely resembles natural reading. We tested French sentences containing sluicing constructions with temporarily ambiguous antecedents, as shown below.

Our design used a sequence of three lexically ambiguous words. Here, the word sale is ambiguous between the third person singular form of saler, 'to salt', and the adjective 'filthy'. The word les is either a plural-marked definite article or an object pronoun. Finally, tranche could be a singular noun ('slice'), but would need an -s suffix to be compatible with the would-be article les to yield the reading 'The butcher/s salt/s the slices'. The word tranche is thus the point of disambiguation in the ambiguous conditions, where no other morphological markers signal the correct parse. In a pretest, acceptability ratings were lower for ambiguous antecedent clauses than for unambiguous ones, which we took to indicate a garden-path effect. The main experiment used a 2×2 design with the factors ambiguity (yes/no) and ellipsis (yes/no). The two control conditions used non-ellipsis anaphors (en, 'of it', in the above example). According to the reactivation hypothesis, processing at the ellipsis site (quand, 'when') should be less difficult in the ambiguous condition, where the antecedent has to be reanalyzed at tranche, 'cut'.

At the sluicing site, antecedent ambiguity led to higher total reading times (t = 3.2) and marginally higher regression path durations (t = 1.9) only for ellipsis sentences, contrary to what the reactivation hypothesis predicted. This pattern was not observed in the region right before the ellipsis, excluding differential processing spillover as an explanation. An analysis of fixations made after the ellipsis site had been fixated revealed an effect of ambiguity on re-reading times for the antecedent region (t = 2.2), as well as higher re-reading probabilities for the initial noun phrase (z = 2.6) and the adjective (z = 2.3) in the ambiguous/ellipsis condition only.

Our study yielded no evidence for a reactivation advantage due to reanalysis. The increased re-reading time on the antecedent suggests that subjects did not always resolve the ambiguity in the initial clause ‒ a type of good-enough processing (e.g. Ferreira & Patson, 2007). On such trials, no fully interpreted antecedent will be available upon encountering the ellipsis site, leading to processing difficulty and re-reading of the antecedent clause.