Memory Retrieval Effects on Filler-Gap Procession

Memory Retrieval Effects on Filler-Gap Processing Philip Hofmeister (philiph@stanford.edu) Department of Linguistics Stanford University Stanford, CA 94305 Abstract higher chances of successful recall. Also, various theoreti- cal works in linguistics (Pesetsky, 1987, 2000; Deane, 1991; Kluender, 1992; Frazier & Clifton, 2002) have noticed that the nature of the filler-phrase impacts acceptability in a sig- nificant way for certain extraction types. For instance, Peset- sky (1987) observes that, in multiple wh-questions when the object wh-phrase is extracted over the subject wh-phrase (also known as “Superiority-violations” (Chomsky, 1973)), accept- ability improves when informationally richer which-phrases are used (2a), as opposed to bare wh-items like who and what (2b): In this paper, I argue that the retrieval of filler-phrases in syn- tactic filler-gap dependencies is facilitated by encoding more information in the filler-phrase. Three self-paced reading ex- periments provide the evidence for this memory facilitation hypothesis: reading times decrease when more explicit wh- phrases and indefinites occur as fillers in three different syn- tactic constructions. Crucially, the decrease in reading times becomes visible only at the point of memory retrieval, i.e. the gap site. There is no effect of the amount of information en- coded in the filler-phrase prior to the retrieval site. Keywords:filler-gap dependencies; memory retrieval; sen- tence processing; syntactic islands Introduction The processing of so-called filler-gap dependencies, as in (1), is known to be affected by a variety of factors. In these constructions, a linguistic constituent (the filler) is displaced from its typical position, leaving an empty syntactic position (the gap). As Hawkins (1999) points out, there are difficulties in identifying the correct gap site, since it has no overt man- ifestation, and while trying to identify the gap site, the filler must be held in working memory while other linguistic con- stituents are processed along the path from filler to gap. The difficulty of processing these constructions further increases with distance (Gibson, 1998), interference effects (Gordon, Hendrick, & Johnson, 2004; Van Dyke & McElree, 2006), as well as processing load imposed by other referential entities along the filler-gap path (Warren & Gibson, 2002). [Which cup] did you drink from feeling ill? before you started Here, I investigate how the amount of information encoded in a filler-phrase affects filler-gap dependency processing. On the extreme view that storing more material always incurs more processing costs, a more complex and informative filler should impair dependency processing. I propose, however, a type of memory facilitation hypothesis that says that linguis- tic elements that encode more information (including syntac- tic, semantic, and lexical information) facilitate their own re- trieval later. In other words, richer linguistic descriptions of discourse entities make recall of these entities easier (all else being equal). Previous work in psychology, psycholinguistics, and prag- matics is suggestive of the effect that informativity can have on memory and discourse. For instance, Bradshaw and An- derson (1982) provide evidence that sentence recall improves when presentation of a sentence occurs along with other causally-related propositions. To explain these results, Brad- shaw and Anderson suggest that elaborating on a given topic increases the number of possible retrieval paths to a given memory trace. Having more retrieval paths, in turn, implies a. Which treatment did which patient receive? > b. What did who receive? Pesetsky explains this contrast as the result of a distinct grammatical mechanism that he terms ’D(iscourse)-linking’. 1 Similarly, Kluender (1992) discusses how increased “referen- tial specificity” facilitates “long wh-movement.” For the most part, however, these accounts are purely theoretical with- out any supporting empirical evidence obtained through con- trolled experimentation and concentrate exclusively on ac- ceptability (cf. Frazier and Clifton (2002)). More impor- tantly, this literature offers few explanations for why informa- tionally richer fillers should improve filler-gap dependencies. In this paper, I present the results of three self-paced read- ing studies that all provide positive evidence for the hypothe- sis that more informative linguistic descriptions lead to faster recall. In each experiment, the results indicate that reading times are significantly reduced at the key point of memory retrieval in a filler-gap dependency (as well as in the spillover region) 2 when a more explicit filler phrase is used. The first study considers the effect of wh-phrase informa- tivity on the processing of so-called syntactic islands. These constructions involve the extraction or dislocation of a lin- guistic element across a syntactic boundary which supposedly blocks the extraction of the element (Ross, 1967). One partic- ular constraint, the wh-island constraint, says that a wh-phrase cannot be extracted out of an embedded clause with another 1 I avoid the use of the term “D-linked” in this paper to char- acterize certain wh-phrases. No precise formulation of or motiva- tion for the D-linking analysis has ever been offered, and there is no reliable method for ascertaining whether a given phrase should be D-linked or not. Indeed, Pesetsky himself notes that the division between scope-taking and D-linked wh-expressions is not absolute, admitting that D-linking (as measured by the ability to participate in superiority-violations) can occur even with the less informative bare wh-items like who and what. 2 Spill-over results from continued processing of a particular lex- ical item or region into subsequent regions. Since most language processing tasks typically are followed by another, the completion of one task (especially complex and challenging processing tasks) often overlaps with the processing of one or more subsequent tasks.