Discourse salience and movement constructions

Ralph L. Rose
Northwestern University

r-rose@northwestern.edu

NPs in syntactically prominent positions are salient antecedents for pronouns.  For example, surface subjects are intuitively preferred as antecedents for pronouns as in examples (1-2) below.  Hudson-D'Zmura and Tanenhaus (1998) provide experimental evidence confirming this intuition, finding faster reading times for coherent (1) than incoherent (2) conditions.

This paper examines the previously uninvestigated question of whether underlying syntactic structure, as well as surface structural position affects saliency for pronoun resolution.  Specifically, it examines the effect of movement on the salience of NPs.  In principle, movement could either increase or decrease an NP's salience.  Comprehenders automatically re-attend to the entity associated with the gap as part of the process of gap-filling (Bever & McElree, 1988; Nicol, 1988); such re-attention might result in greater salience for the entity than when there is no gap.  Alternatively, moving an NP away from the position where it is assigned a thematic role (even if it is to a more syntactically salient position) might have the effect of decreasing an entity's salience.

To test these possibilities, subjects read short discourses with either NP-raising or Tough-movement constructions (as in (2-3) and (4-5), respectively) one sentence at a time in a self-paced reading study.  Subjects read 48 experimental items, 24 NP-raising and 24 Tough-movement constructions interspersed among 24 fillers.

Preliminary results replicate those of Hudson-D'Zmura and Tanenhaus showing a main effect of coherence: reading times were faster in the COHERENT condition than in the INCOHERENT condition.  Further, results also indicate that underlying syntactic structure is a relevant factor for discourse salience, but the effect is construction-dependent.  NP-raising constructions exhibit slower reading times (i.e., (3) slower than (2)) while Tough-movement constructions exhibit faster reading times (i.e., (5) faster than (4)).  Implications for such  models of anaphora resolution and discourse coherence as Centering Theory (Grosz and Sidner, 1986) as well as implications for syntactic  analyses of these constructions will be discussed.

Examples