On the joint effects of repetition and stimulus quality in lexical decision: Looking to the past for a new way forward

Abstract

Two experiments investigated the joint effects of stimulus quality and repetition in the context of lexical decision. Experiment 1 yielded an interaction between repetition and stimulus quality for words (but additive effects for nonwords) when the lag was short, replicating previous reports. Experiment 2, with a much longer lag than Experiment 1, yielded main effects of stimulus quality and repetition, but these factors no longer interact. The joint effects of stimulus quality and repetition for words as a function of lag can be understood in terms of two loci for repetition effects: one short-term and one long-term. The transient effect of repetition is on activation levels in the lexicons (and in which the input lexicon, but not beyond, is affected by stimulus quality), whereas the long-term effect is on the strength of two-way connections between lexical–lexical and lexical–semantic modules. These data and others, taken together with the account, provide a new way of thinking about a 30-year-old conundrum.

Keywords:

• Lexical decision
• Word recognition
• Repetition
• Stimulus quality

Acknowledgments

We thank Kathy Rastle and several anonymous reviewers for helpful comments. This work was supported by a postdoctoral fellowship from the Natural Sciences and Engineering Research Council (NSERC) to C.B., a doctoral fellowship from NSERC to S.O., and NSERC Grant A0998 to D.B.

Notes

¹ When sphericity was violated, the Greenhouse–Geisser correction was applied, resulting in fractional degrees of freedom.

² We are not claiming that a theory like Logan's (1988) is unable to account for the present data. However, we note that the processing episodes wind up being more and more abstract under certain conditions, suggesting that there are instances that are more frankly “episodic” and instances that seem more like prototypes (lexical representations?). Accounts with a more abstract level of description are arguably much better developed at the present time than is instance theory as currently formulated in terms of their ability to capture a variety of phenomena in both brain-damaged and intact readers (e.g., Coltheart, 2004; Coltheart et al., 2001). For example, the “direct retrieval” that characterizes a central element in instance theory currently lacks distinct levels. We find it hard to imagine that retrieval of an instance is always simply “one step” as opposed to consisting of multiple steps (e.g., how would one-step retrieval yield the various dissociations seen with brain-damaged patients; how would one-step retrieval explain the contrast between Experiments 1 and 2 for words reported here; how would one-step retrieval account for the list contexts effects reported by Besner et al., 2010; O'Malley & Besner, 2008; Yap et al., 2008, among others?).

³ How, if it all, do the present results speak to the recognition memory literature? We suspect that the answer is “very little” insofar as the present results, at least for words, relate to lexical knowledge rather than episodic memory.

⁴ Whatever orthographic coding scheme at the letter level one opts for, the effect of stimulus quality must be able to cascade through it in order to produce a Stimulus Quality × Repetition interaction for words, given the assumption that stimulus quality affects early level processing in addition to lexical processing.

⁵ Cascaded processing can also give rise to additive effects of two factors on RT provided that certain boundary conditions are respected (see Ashby, 1982; McClelland, 1979; Roberts & Sternberg, 1993). To date, however, neither of the localist computational models on the table produces systematic additivity between factors on cycles to criterion (e.g., between stimulus quality and word frequency; see O'Malley & Besner, 2008; between stimulus quality and neighbourhood density; see Reynolds & Besner, 2004; or between stimulus quality and spelling–sound regularity; see Besner et al., 2010). These computational models (Coltheart et al.'s, 2001, dual-route cascaded, DRC, and Perry et al.'s, 2007, connectionist dual-process, CDP + , models) face the additional issue that the lexical route is also engaged in interactive activation (IA; e.g., between the letter level and the orthographic input lexicon). Yet to date there is no existence proof showing that an IA model can produce systematic additivity between the effects of pairs of factor on cycles to criterion (such as stimulus quality and word frequency; see Besner, Wartek, & Robidoux, 2008). Computational modellers who prefer cascaded accounts to discrete processing accounts may therefore need to implement a model that is feed forward only and respects the boundary conditions needed to produce additive effects between factors like those discussed here and elsewhere (see Ashby, 1982; McClelland, 1979).