Running Head: PHONOGRAPHIC SUBLEXICAL UNITS Phonographic sublexical units in visual word recognition

Recent models of visual word recognition assume that sublexical orthographic-phonological information is organized according to an onset-nucleus-coda scheme (Plaut, McClelland, Seidenberg, & Patterson, 1996; Jacobs, Rey, Ziegler, & Grainger, 1998, but see Coltheart, Curtis, Atkins & Haller, 1993 for an alternative view). In this study we test the hypothesis that onsetnucleus-coda subsyllabic components are sublexical reading units mediating visual word recognition. We present a sublexical measure that is based on cumulated frequency of these phonology-dependent, orthographic subsyllabic components: Subcomponent Frequency (SCF). In a lexical decision task we found that SCF facilitates visual word recognition for low-frequency words, but not for high-frequency words. For nonwords, we observed that those with high-SCF are harder to distinguish from words than nonwords with low-SCF. In contrast, a standard and purely orthographic measure of sublexical structure, Bigram Frequency, did not produce an effect in our study, when SCF was controlled. Thus, we conclude that SCF is a promising first step towards indexing phonographic sublexical processing. Phonographic Sublexical Units 2 Introduction Orthographic and phonological information has been shown to be strongly interdependent in visual word recognition (Stone, Vanhoy & Van Orden, 1997; Van Orden & Goldinger, 1994; Ziegler, Montant & Jacobs, 1997; Ziegler, Van Orden & Jacobs, 1997). Evidence for bi-directional orthographic-phonological influences comes from a variety of empirical phenomena, such as the pseudohomophone effect (e.g. Coltheart, Davelaar, Jonasson, & Besner; 1978; Rubenstein, Lewis, & Rubinstein, 1971; Seidenberg, Peterson, McDonald, & Plaut, 1996; Ziegler & Van Orden, & Jacobs, 1997), the bi-directional consistency effect (Jared, 1997; Jared, McRae & Seidenberg, 1990; Stone et al., 1997; Treiman, Mullenix, Bijeljac-Bibac, & Richmond-Welty, 1995; Ziegler, Montant, & Jacobs, 1997) or the multi-letter grapheme effect (Rastle & Coltheart, 1998; Rey, Jacobs, Schmidt-Weigand, & Ziegler, 1998). In recent years, when the interdependency of orthography and phonology became practically undisputed, different research groups have built connectionist word recognition models that include both, orthographic and phonological coding, such as the DRC (Coltheart, Curtis, Atkins & Haller, 1993; Coltheart & Rastle, 1994), the PDP-model of Plaut et al. (1996), and the MROM-p (Jacobs et al., 1998). These models postulate different levels of orthography and phonology and in their connection weights they quantify the functional interdependence of orthography and phonology. Inspired by these models, we propose and test a new hypothetical answer to an old question in the domain of visual word recognition. The question is: Which are the sublexical orthographic units that mediate visual word recognition and what measure indexes those units best? A number of such sublexical units have been proposed, such as bigrams, trigrams, or the letters themselves, and measures have been derived from those assumptions such as bigram frequency, trigram frequency, or positional letter frequency (e.g. Biederman, 1966; Gernsbacher, 1984; Grainger & Jacobs, 1993; Humphreys, Evett, & Quinlan, 1990; Massaro & Cohen, 1994; Seidenberg & McClelland, 1989). However, we believe that these measures and units fall short in one important aspect of current theories and models of visual word recognition: they are purely orthographic measures and, thus, neglect the demonstrated role of phonological processes in visual word recognition. As an answer to the question about how to index sublexical units best we propose a measure that relies on units that play a functional role in the mapping from orthography to phonology in at least two of the recent connectionist models of visual word recognition. Both, the MROM-p (Jacobs et al., 1998) and the PDP-model of Plaut et al. (1996) Phonographic Sublexical Units 3 use the subsyllabic onset-nucleus-coda scheme to organize the mapping from orthography to phonology in monosyllabic words. Based on these three functional subsyllabic components we propose and test a new measure that indexes a sublexical structure that mediates visual word recognition: Subcomponent Frequency (SCF). Definition: Subcomponent Frequency is defined as the mean of the logarithms of orthographic onset, nucleus, and coda frequency. Orthographic onset frequency of a target word is calculated as the summed frequency of all monosyllabic words that share the same onset. Orthographic nucleus and coda frequency of a target word are calculated as the summed frequencies of all monosyllabic words that share the same orthographic nucleus or coda, respectively. Thus, SCF represents a phonographic measure (Peereman & Content, 1997). It is phonologybased, as the segmentation of the syllable into onset, nucleus, coda relies on phonological properties of the language. It is orthographic, as its calculation is based on the orthographic units that correspond to phonological onset, nucleus and coda. Like for bigram frequency, cumulated frequencies of words that share the same sublexical orthographic units, are computed. However, the difference to bigram frequency is that these orthographic units are not constructed independent of phonology but rather defined by the structure of phonological subsyllabic units. Predicted Subcomponent Frequency Effects: This study investigates the question whether SCF indexes sublexical processing during visual word recognition. If subsyllabic components are critical reading units by which word recognition is mediated, then high-SCF should facilitate word recognition. However, sublexical phonologicalorthographic effects such as consistency effects (Andrews, 1982; Backman, Bruck, Hebert, & Seidenberg, 1984; Seidenberg, Waters, Barnes, & Tanenhaus, 1984), homophone errors in semantic decision tasks (Jared & Seidenberg, 1991), multi-letter grapheme effects (Rey et al., 1998) or body neighbor effects (Ziegler & Perry, 1998) are mostly observed for low-frequency words only (but see Jared, 1997). Recognition of high-frequency words may be mainly and rapidly performed on the basis of a lexical procedure and may be less affected by a slower sublexical processing involving the conversion of sublexical orthographic units into sublexical phonological units (e.g. Jacobs et al., 1998; Coltheart et al., 1993). Therefore, we expect a facilitating SCF effect for low-frequency words but no such effect or at least a minor effect for high-frequency words. Phonographic Sublexical Units 4 Testing the effects of one sublexical measure and claiming that the observed effects are due to this particular sublexical measure is problematic since, by definition, most sublexical measures are highly correlated. While controlling for other established measures of visual word recognition like frequency, neighborhood and consistency, it is virtually impossible to separate SCF and bigram frequency (BF) for words. However, the measures can be disentangled for nonwords, and then be tested against each other. As BF is probably the most prominent, yet purely orthographic, sublexical measure we tested SCF while holding BF constant for nonwords. Only if SCF effects still prevail when BF is controlled, we can consider the SCF measure as a promising index for sublexical processing in visual word recognition. For SCF, the hypotheses for nonwords are straightforward. If SCF is a relevant index for sublexical processing, then high-SCF nonwords should be more similar to words in the lexicon as they more frequently share sublexical components. Thus, high-SCF nonwords should be harder to distinguish from words in a lexical decision task than low-SCF nonwords. Similarly, if BF is a relevant index for sublexical processing then, high-BF nonwords should be harder to distinguish from words than lowBF nonwords.