How good a cue is a resumptive pronoun? Processing relative clauses in Slovenian

We explore the time course of processing Slovenian subject and object relative clauses (SRC and ORC, respectively) and the role of resumptive pronouns (RP). Participants (adult native speakers of Slovenian, Exp1: N=37Exp2: N=33, Exp3: N=35) read the sentences in the self-paced mode, followed by a comprehension question after each sentence. In Exp.1 we ask whether the basic SRC/ORC processing asymmetry obtains in Slovenian, despite the presence of an RP. Results: The RC verb was read longer in ORCs compared to SRCs, and postverbal NPs were read longer than preverbal NPs (Figure 1). Both observations are likely to reflect integration effects, suggesting that the presence of RP does not cancel the stand-ard SRC/ORC processing asymmetry. In Exp. 2, we ask whether this asymmetry depends on the structural position of the RC within the sentence. We manipulated RC type and structural position (center-embedded, right-peripheral), across four conditions. Results: similarly to Exp.1, a SRC/ORC asymmetry was observed at the RC verb as well as between the postverbal vs. preverbal NPs, independently of the structural position of RC. The main clause predicate was read slower in sentences with center-embedded RCs compared to those with right-peripheral RCs, in accordance of predictions of metric-based theories of integration cost. Questions following ORCs took somewhat longer to answer than those following SRCs. At the same time, all RCs were read slower in the right-peripheral position than in the center-embedded position, and questions following right-peripheral RCs were answered significantly less accurately than those following center-embedded RCs. We attribute the greater complexity associated with the right-peripheral position to availability of a competitive parse based on a pseudo-relative structures. In Exp.3, we investigate how the feature structure of an RC head and its corresponding RP may affect retrieval of the RC head with the ORC subject as a featural intervenor. Design: by crossing values for number (sg., pl.) and gender (masc.,fem.) between the RC head, RP and the ORC subject we created a continuum of feature matching across four conditions. Results: the integration effect at the RC verb was significantly greater in the conditions with non-matching gender, but not those with non-matching number, suggesting that an RP does not cancel the intervention effect caused by featural similarity, while supporting the conjecture that different patterns of processing nominal features may correlate with their grammatical status (e.g. semantic vs. morphological)