Revisiting frequency and storage in morphological processing*

1 Introduction The balance between storage and computation of complex words is a major point of departure both for theories of lexical representation (e. The atoms of lexical memory that are implicated in lexical processing experiments—be they whole words, roots and affixes, or some combination thereof—must ultimately coincide with the units of morphological theory if the latter are to be theories of the mental lexicon. One clue into the architecture of lexical memory comes from the presence of robust frequency effects in lexical decision tasks, in which subjects judge whether a written or spoken stimulus is a real word and processing complexity is measured by reaction time. The recognition of complex words is facilitated both by whole word frequency (as measured from some representative corpus) as well as base (also called cluster or root) frequency, the summed frequency of all words sharing the same root. Frequency effects have long been known to account for a large amount of variance in lexical decision latencies (e.g., Howes & Solomon 1951). The existence of base frequency effects suggest that the roots posited by mor-phologists are stored in memory and are integral to lexical access. Many models of lexical processing, however, downplay the role of roots and of base frequency. Proponents of these dual-route models (e.g., Baayen et al. 1997, Baayen & Schreuder 1999, Caramazza et al. 1988) argue that complex words are also stored in memory and emphasize the role of whole word frequency in word recognition. Some studies (e.g., Sereno & Jongman 1997) deny base frequency effects altogether. In this study, we investigate whole word frequency effects for English regularly inflected words reported by Alegre & Gordon (1999), Baayen et al. (2007), New et al. (2004), and Sereno & Jongman (1997). We model a very large database of visual lexical decision latencies with mixed effects regression, using residualization to control for confounding predictors of reaction time. The result is a more nuanced view of base and whole word frequency effects and the storage-computation debate. After reviewing previous experimental work on complex word recognition (§2), we describe the data set and construct a statistical model of lexical decision laten-cies for regularly inflected complex words in English (§3). We use this model to investigate word and base frequency effects (§4) and conclude that the whole word frequency effects do not argue in favor of whole word storage; in fact, they are inconsistent with the …