Potsdam at SRG on Thursday

Jason Overfelt writes:

Please join us this Thursday, April 11th, for a meeting of the Syntax/Semantics Reading Group.  The current syntax guru, Eric Potsdam, has agreed to talk to us about "The processing of Russian numerical constructions".  An abstract for the talk can be found below.  We will be meeting at 6:30p at Rajesh's house at 11 Myrtle in Northampton.  Please bring $5 for pizza  or your dinner. 

Abstract:  

Natural language has numerous ways to encode anaphoric dependencies, including filler-gap (movement) constructions, antecedent-anaphor relations, control, variable binding, and coreference. Such relations can be created in the syntax (e.g. movement constructions), in the semantics (e.g. variable binding), or in the discourse (e.g. coreference). Reuland 2011, building on Reinhart 1983 and others, proposes the following hierarchy in the economy of the encoding of anaphoric dependencies.

(1) syntax  <  semantics  <  discourse

The hierarchy translates into processing preferences; the processing of dependencies farther to the left should be easier than the processing of those to the right. A specific prediction is that syntactic dependencies require less processing effort than discourse-derived dependencies (Koornneef 2008). In other words, movement is, perhaps surprisingly, less burdensome for processing than pronominalization. To test this prediction, this paper investigates two constructions from Russian which have not been previously fully analyzed. They minimally differ on the surface but we show that they involve distinct kinds of anaphoric dependencies. An experimental study confirms that the syntactic dependency requires less effort than the discourse dependency.

In the Russian examples in (2), a nominal can be fronted out of a numerical expression, stranding a modifying numeral. When the stranded numeral is a so-called paucal number (1.5, 2, 3, 4, and the expression ‘both’), the fronted nominal can appear in a form that matches in number with the numeral, (2a), or in a non-matching plural form, (2b).

(2) a. A’ movement dependency: matching morphology
         Sobor-a                   v gorodke bylo tri        sobor-a
          cathedral-paucal in town       was three.paucal

        b. pronominal coreference dependency: non-matching morphology
           Sobor-ov        v gorodke bylo tri        pro
            cathedral-pl in town       was three.paucal
‘As for cathedrals, there were three in that town.’

We argue that the construction with matching between the fronted element and the numeral (2a) involves A'-movement of the fronted element but (2b) without matching involves co-indexation between the fronted element and a null pronoun, as shown. Evidence comes from island sensitivity, number connectivity, binding reconstruction, parasitic gaps, word order, and resumption with a pronoun or epithet.

The hierarchy in (1) predicts that (2a) should be processed more easily than (2b). We test this prediction in a reading time experiment. The results show a strong effect of the number difference (p=0.0085), with a statistically significant slowdown in reading time in the region after the numeral in the non-matching case (discourse dependency) compared to the matching case (syntactic dependency). The result supports Reuland’s hierarchy in (1).

We reject an alternative explanation of the data in which the reading time slow down is due to a simple morphological mismatch, which has been noted by several studies (e.g Fanselow & Frisch 2006, Molinaro et al. 2011). We also reject an alternative in which the slow down is due to the agreeing paucal form providing  fewer structural or lexical expectations, which are also known to facilitate processing (Vasishth 2003, Yoshida 2006, Lau et al. 2006, and others).

In conclusion, our analysis of a syntactic minimal pair in Russian forms the basis of a processing study comparing a syntactic dependency to a discourse dependency. Our experimental investigation of the two constructions shows that syntactic dependencies are processed more quickly than discourse dependencies, providing novel support for the hierarchy in (1). From a processing perspective, movement is less burdensome than pronominalization.