Memory for linguistic features and the focus of attention: evidence from the dynamics of agreement inside DP

ABSTRACT

The amount of information that can be concurrently maintained in the focus of attention is strongly restricted. The goal of this study was to test whether this restriction was functionally significant for language comprehension. We examined the time-course dynamics of processing determiner-head agreement in English demonstrative phrases. We found evidence that agreement processing was slowed when determiner and head were no longer adjacent, but separated by modifiers. We argue that some information is shunted nearly immediately from the focus of attention, necessitating its later retrieval. Plural, the marked feature value for number, exhibits better preservation in the focus of attention, however, than the unmarked value, singular.

KEYWORDS:

• Language comprehension
• grammatical agreement
• number
• working memory
• focus of attention
• speed–accuracy trade-off (SAT)

Acknowledgments

The authors acknowledge the support of NIH grant #HD-056200. Thanks to Kathy Akey for assistance in collecting data. The authors are grateful to the audience of the 22nd Annual CUNY Sentence Processing Conference (University of California, Davis) for their helpful comments; to two anonymous reviewers for their insight and critical commentary; to Pranav Anand for his help parsing the Gigaword corpus; and to the following individuals for important contributions to the development of the project: Julie Van Dyke, Ellen Lau, Colin Phillips, Michael Shvartsman & Clare Stroud.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 As an anonymous reviewer points out, it is conceivable that comprehenders could (learn to) preactivate different kinds of information in the number-marked demonstratives compared to the simple the-conditions. Agreement is never at stake between the and the NP it combines with, which means that other effects, like plausibility, could be more pronounced. In other words, the cost of modification might vary depending on what others relations have to be computed. An alternative design that only used demonstratives could be valuable here. But we agree generally with the reviewer that there is a complex interaction between prediction and integration at all levels, one which is likely to be affected by language-general properties and but also by experiment-specific statistics.

2 It is conceivable that the one- versus two-modifier rate difference in plausibility controls (e.g., “the risk-taking burglars/*jewels”; “the clever, risk-taking burglars/*jewels”) could also be attributed to retrieving the first modifier from memory. This is possible and we cannot definitively rule it out. But there are two considerations that push us away from that interpretation. The first is the fact that, in the plausibility controls, the two modifiers were always mutually consistent such that the last modifier was always sufficient to determine (un)acceptability when combined with the head noun. We didn’t have any mixed cases like, “the shiny, risk-taking jewels”, where one modifier was compatible with the head noun, and the other wasn’t. The second is a commitment to early and incremental semantic composition (Brennan & Pylkkänen, 2012), which predicts a partial meaning to be elaborated by the time the head noun is reached. For this reason, we attribute the one versus two-modifier cost to the complexity of the meaning. But further work is required here to arrive at a better conclusion.

3 Whether plurals are actually semantically more complex than singulars is a matter of debate (de Swart & Farkas, 2010; Farkas, 2006; Sauerland et al., 2005). The resolution of this debate is independent of the morphosyntactic facts discussed. If plurals do turn out to be semantically “basic”, then they would represent the interesting, less frequent case in which formal morphosyntactic markedness does not align with semantic markedness.

4 This is only an approximation, since the entire conditional probability of an expression is relevant to the speed of processing on a particular word. For this analysis we assume that the external distribution of the demonstrative DP does not affect how its length varies with number. That may turn out to be false (for example, if plural DPs are shorter in subject position but not in object position).

5 That the similarity of these estimated values is not in some way an artifact of the technique can be appreciated by surveying results from other constructions that cannot be plausibly characterised as a simple difference between whether a feature is available in focal attention or must be retrieved from memory. For example, the difference between single- and double-gap dependency resolution examined in McElree et al. (2003) leads to rate differences on the order of 400 ms. Here, the SAT procedure is tracking differences in the time to retrieve one versus two representations, and, crucially, the respective position (or order) in syntax.

Additional information

Funding

This work was supported by the National Institutes of Health: [Grant Number HD-056200].