Orthographic and Associative 1 PSYCHOPHYSIOLOGY, in press Running head: Orthographic and Associative Neighborhood Density Orthographic and Associative Neighborhood Density Effects: What is Shared, What is Different?

Words with many orthographic neighbors elicit a larger N400 than words with few orthographic neighbors. This has been interpreted as stronger overall semantic activation due to orthographic neighbors activating their semantic representations. To investigate this claim, we manipulated the number of associates of words (NoA), a variable directly affecting overall semantic activation, and compared this to the ERP effect of the number of orthographic neighbors (N) in a lexical decision task. Words with high NoA and with high N produced a very similar increase of the N400. In addition, a higher N increased the amplitude of the Late Positive Complex. The common N400 effect suggests that N affects semantic activation, like NoA does. The late positive effect specific to N could occur because words with few orthographic neighbors initially elicit little activity in the orthographic system, thereby resembling nonwords, which leads to distinct processing. Descriptors: visual word recognition, semantic representation, association, orthographic neighbors, language, ERP Orthographic and Associative 3 Orthographic and Associative Neighborhood Density Effects: What is Shared, What is Different? The recognition of written words entails the activation of various levels of representation, from a visual-perceptual level, via orthographic and phonological levels, to the semantic level. A wide-spread assumption in theories of visual word recognition is that the presentation of a given word does not activate only the representations of that particular word, but also those of similar words (e.g., Bowers, Davis, & Hanley, 2005). For the orthographic level this is borne out in the idea of an orthographic neighborhood. Coltheart, Davelaar, Jonasson, and Besner (1977) defined an orthographic neighbor as a word that can be derived from another word by changing one letter in a specific position, keeping the number of letters constant – in other words, orthographic neighbors are words that are identical except for one letter. The word cat, for example, has the orthographic neighbors bat, rat, cut, cap, car, etc. When reading the word cat, its orthographic neighbors are assumed to be activated also to some degree because of their extensive overlap in letters in each position 1 . Coltheart‟s N, also referred to as orthographic neighborhood density, indicates for a given word how many orthographic neighbors it has. A wide range of studies has shown that orthographic neighborhood density influences visual word recognition (for a review see Andrews, 1997). The majority of studies using lexical decision has found a facilitatory effect of orthographic neighborhood density: responses to words with a high N (i.e., many orthographic neighbors) are faster and more accurate than to words with a low N (e.g., Andrews, 1989, 1992, 1997; Carreiras, Perea & Grainger, 1997; Pollatsek, Perea & Binder, 1999; see Siakaluk, Sears & Lupker, 2002, for a review). This is a well-established effect that has been replicated in different languages and with various populations, including Orthographic and Associative 4 novel readers and dyslexic readers (e.g., Duñabeitia & Vidal-Abarca, 2008; Lavidor, Johnston & Snowling, 2006; Laxon, Coltheart & Keating, 1988). The effect of neighborhood density is often explained in terms of an interactive activation model where a written word activates matching letter units and those letter units in turn activate word units containing these letters, with activation also flowing back from the word to the letter units (McClelland & Rumelhart, 1981). In such a framework, the presentation of a particular word would not only activate the word unit corresponding to that word but also the word units of its orthographic neighbors, due to the extensive orthographic overlap. As a consequence, words with many orthographic neighbors would lead to the activation of more word units than words with few orthographic neighbors. Furthermore, activation from orthographic word units would flow back to letter units and back again to word units, leading to a further build-up in overall lexical activation. Grainger and Jacobs (1996) proposed that this might actually be the basis of the facilitatory effect in lexical decision, by way of a task-specific strategy. In their Multiple Read-Out Model (MROM) two mechanisms are available that can lead to a “word” response. The first is unique identification and corresponds to a specific word unit reaching a certain activation threshold. Once a particular word is identified the participant gives a “word” response. The second mechanism is a fast familiarity-based guess as to whether a stimulus is an actual word, as suggested by Balota and Chumbley (1984). Within the framework of the interactive activation model, this is operationalized through overall lexical activation: If a stimulus generates a large amount of (unspecific) lexical activation then a good guess would be that the stimulus is word-like, granting a “word” response. As words with many orthographic neighbors would activate many word units and thus generate high overall lexical activation, this could be the basis for high-N words showing faster RTs than low-N words in lexical decision. Orthographic and Associative 5 An interesting question is whether orthographic neighbors might also activate their corresponding semantic representations. This could be expected in the spirit of an interactive activation or feed-forward model of visual word recognition. A study with event-related potentials (ERPs) by Holcomb, Grainger, and O‟Rourke (2002) actually seems to indicate that orthographic neighbors activate their semantic representations. Holcomb et al. presented highand low-N words to the participants in their study and found that the ERP showed a bigger N400 for highthan for low-N words. The N400 is a component that responds to a series of semantic manipulations, like semantically incongruent words in sentences, cloze probability, and semantic priming (for a review see Kutas & Van Petten, 1994). Based on such findings, researchers have argued that the N400 reflects the activation of semantic representations in long-term memory (Kutas & Federmeier, 2000) or, alternatively, the integration of semantic information in a post-lexical stage (Brown & Hagoort, 1993; Holcomb, 1993). Interestingly, the N400 also reacts to manipulations that are not per se semantic and might be assumed to rather affect pre-semantic lexical stages, like lexical frequency and phonological priming. Thus, the effects of both lower-level and semantic factors seem to converge on the process underlying the N400. This is consistent with a semantic locus of the N400 as in the proposals cited above, given the assumption that activation can spread from lexical to semantic representations in a feed-forward manner. Given this view of the N400, Holcomb et al. proposed that the N400 effect of orthographic neighborhood density reflected overall semantic activation, even though the underlying manipulation is orthographic in nature: The activation of orthographic neighbors on the lexical-orthographic level would lead to the activation of the respective semantic representations and consequently produce an N400 effect proportional to the number of orthographic neighbors. They argued that the general semantic interpretation of the N400 outlined above would be suggestive of this proposal, Orthographic and Associative 6 although they admittedly could not exclude that the effect might be entirely orthographic in nature. Several behavioral studies actually found evidence that orthographic neighbors activate their respective semantic representations (Boot & Pecher, 2008; Duñabeitia, Carreiras, & Perea, 2008; Pecher, Zeelenberg, & Wagenmakers, 2005; Rodd, 2004). Duñabeitia, Carreiras et al. (2008), for example, found evidence that words activate their ortho-phonological neighbors up to the semantic level in a study of ortho-phonologically mediated associative priming. They used Spanish prime-target pairs like oveja – MIEL (sheep HONEY), with the prime being preceded by a forward mask and presented for only 50 msec, being immediately replaced by the target. Oveja is an ortho-phonological neighbor of abeja (bee) in Spanish, which in turn is an associate of MIEL. In a lexical decision task, participants responded significantly more rapidly when such ortho-phonological neighbors of associates preceded target words than when unrelated control words did so. It is noteworthy that this ortho-phonologically mediated associative priming effect was very similar in size to the direct associative priming effect, that is, the effect for associated prime-target pairs such as abeja – MIEL, as obtained in the same study. Behavioral studies providing evidence for the semantic activation of orthographic neighbors show that a semantic origin of the N400 effect of orthographic neighborhood density is a definite possibility. A more direct test of the claim that this N400 effect is semantic in nature would be to compare it with a manipulation assumed to involve the semantic level of representation, such as a manipulation involving the number of semantic associates. This is one of the aims of the current study. In the following we first give some background on semantic activation in visual word recognition and then detail the semantic manipulation which we will compare to the orthographic neighborhood density effect. Orthographic and Associative 7 Various studies have investigated the potential influence of semantics on visual word recognition, manipulating variables such as the number of semantic features (Pexman, Holyk, & Monfils, 2003; Pexman, Lupker, & Hino, 2002; Pexman, Hargreaves, Siakaluk, Bodner, & Pope, 2008), the diversity of contexts in which a certain word occurs (Adelman, Brown, & Quesada, 2006; McDonald & Shillcock, 2001; Pexman et al., 2008), and the number of semantic neighbors in a high-dimensional semantic space derived from co-occurrences in large text corpora (Buchanan, Westbury, & Burgess, 2001; Pexman et al., 2008). A very promising measure proposed to reflect semantic richness is the number of associates (NoA) a word has – which is the measure we will focus on in this study. NoA is defined as the number of different first associates produced in a free association norming study, where participants are presented with a particular word and asked to note down the first word that comes to their mind. This variable has a long tradition in memory research and it has been shown that words with few associates lead to better performance in cued recall than words with many associates (e.g., Nelson, Schreiber, & McEvoy, 1992). It has recently also found its way into research on visual word recognition: Buchanan et al. (2001) found that words with a high NoA produced faster lexical decision times than words with a low NoA, although this effect depended on the interaction with another semantic variable derived from the Hyperspace Analog to Language (HAL) model (Lund & Burgess, 1996). Subsequent studies provided further evidence that high-NoA words result in faster lexical decision times than low-NoA words (Balota, Cortese, Sergent-Marshall, Spieler, & Yap, 2004; Duñabeitia, Avilés, & Carreiras, 2008; Locker, Simpson, & Yates, 2003; Mirman & Magnuson, 2008; Yates, Locker, & Simpson, 2003). For instance, Mirman and Magnuson (2008) investigated the influence of various potential measures of semantic richness on lexical decision. Partial correlations indicated that NoA was as good a predictor of lexical decision performance as measures based on semantic feature norms and co-occurrence based measures. Orthographic and Associative 8 The findings from memory and psycholinguistic research regarding NoA have been interpreted as evidence that the presentation of a particular word leads to the activation of its associates. Thus, NoA is conceived as associative neighborhood density, with words having many associates as inhabiting a dense associative neighborhood and words having few associates as inhabiting a sparse associative neighborhood. Regarding the different direction of the effects of NoA on cued recall and lexical decision, we suggest that it probably reflects different task demands. In cued recall, participants receive a cue word and have to retrieve a (associated) target word presented during an earlier study phase. Producing a correct answer, thus, implies the activation of a specific word in its absence. Under such circumstances, the activation of associates might be harmful. According to Nelson and colleagues (Nelson et al. 1992; Nelson, McKinney, Gee, & Janczura, 1998), who have done the bulk of work on the influence of associates in cued recall, target words implicitly activate their associates during the study phase. Later in the test phase, the retrieval of a target word depends partially on its reactivation during an implicit search process. During this process the previously activated associates serve as competitors of the target word, resulting in an inhibitory influence of NoA (Nelson et al. 1992; Nelson et al., 1998). In contrast, in lexical decision the target word is present during task execution and the orthographic input will primarily drive the response and not the search for a previously presented word among memory traces. Furthermore, as mentioned above, lexical decisions might rely to some extent on the familiarity of a stimulus instead of unique identification (Balota & Chumbley, 1984), and the activation of associates might increase familiarity and thus lead to a facilitatory effect. In general, the facilitatory effect of NoA in lexical decision can be explained by models assuming that semantic activation – at least partially – determines lexical decisions (e.g., Masson, 1995; Plaut, 1997; Rodd, Gaskell, & Marslen-Wilson, 2004) or that feedback occurs from semantic to orthographic representations, whose activation then would determine lexical decisions Orthographic and Associative 9 (Balota, Ferraro, & Connor, 1991). Although we interpret NoA as a measure reflecting semantic representation, it is important to note that other proposals situate associative relationships at the lexical-orthographic level and explain them by mere co-occurrence rather than semantic relatedness (Lupker, 1984; Moss, Hare, Day, & Tyler, 1994; Shelton & Martin, 1992). This position primarily relies on studies of priming that have found automatic priming for so-called pure associate pairs like spider web, supposedly lacking a semantic relationship, whereas prime-target pairs with a semantic relationship but no measurable association showed no or reduced automatic priming. However, Hutchison (2003) noted that associated word pairs without any semantic relation are very rare. He also mentions that one can doubt that phrasal associates such as spider web, often chosen as examples for pure associates, have no semantic relation at all, as the two words converge on a common concept. In line with Hutchison's observations, the associates included in the NoA count for our material show an additional semantic relationship. The associates for giraffe found in the Spanish database used by us (Fernández, Díez, & Alonso, 2006), for example, include among others its superordinate animal, the category coordinate elephant, the feature tall, and zoo, which might be interpreted as containing a script relationship (Moss, Ostrin, Tyler, & Marslen-Wilson, 1995; Schank & Abelson, 1977). Thus, it seems that the associates included in the NoA count would reflect semantic relations and that NoA therefore provides information about the semantic representation of a word. What is more, there is an extensive and continuing line of research using semantic associates to explore how meaning is accessed and how it is represented (some recent examples are: Holcomb & Grainger, 2009; Hutchison, Balota, Cortese, & Watson, 2008; Perea, Duñabeitia, & Carreiras, 2008; Rolke, Heil, Streb, & Henninghausen, 2001). In general, one can say that associates gained from the free association procedure are thought to reflect the semantic field of the cue word. Thus, NoA seems to be a suitable measure of overall semantic activation and therefore makes a good Orthographic and Associative 10 benchmark to which to compare the effect of orthographic neighborhood density on the ERP. As a supposedly semantic variable, we expect an N400 effect for NoA and if this is comparable to the orthographic neighborhood density N400 effect, this would support the notion that the orthographic neighborhood density effect reflects overall semantic activation. However, should the two effects differ then the orthographic neighborhood density N400 would seem to be of a different nature. A difference in the topography of the two effects would provide some evidence that different neural generators would be involved or the same neural generators to a different degree (McCarthy & Wood 1985; but see Urbach & Kutas, 2002). Orthographic neighborhood density and NoA might, for example, differentially engage brain areas linked to orthographic and semantic processing, which could lead to different scalp topographies of the ERP. There is some evidence that increases in N density lead to bigger competition at the orthographic level, whereas NoA does not. An increasing number of orthographic neighbors would not only result in more global lexical activation but also in a bigger set of orthographic candidates among which the word recognition system has to choose. The latter would be especially important in tasks requiring unique identification, such as perceptual identification or continuous word reading as measured by eye tracking. Indeed, studies employing these tasks have found an inhibitory effect for N density (Carreiras et al., 1997; Pollatsek et al. 1999; Snodgrass & Mintzer, 1993; but see Sears, Lupker, & Hino, 1999). By contrast, Duñabeitia et al. (2008) found facilitatory effects of NoA in perceptual identification and eye-movement measures, suggesting that NoA does not strongly affect competitive processes in word recognition (in contrast to the competitive processes seen in memory tasks). Furthermore, it is known that the N400 window contains various negativities differentiated by their topography. The effect of concreteness on the ERP, for example, occurs as an enhanced negativity around 400 msec after target presentation with a frontal maximum. Orthographic and Associative 11 However, the N400 effect related to other semantic manipulations is described as having a posterior and slightly right-lateralized maximum (Kounios & Holcomb, 1994; Kutas & Van Petten 1994; Swaab, Baynes, & Knight, 2002). Thus, the orthographic and associative neighborhood manipulations might differentially affect these negativities in the N400 window. One could also speculate that an eventual ERP effect of NoA would set in later than the effect of orthographic neighborhood density. In terms of an interactive activation model (McClelland & Rumelhart, 1981), the presentation of a word with many orthographic neighbors would first lead to an increased level of overall lexical activation because of the initial co-activation of orthographic neighbors based on the strong overlap in letter identity and letter position between such a word and its orthographic neighbors. The feedback to orthographic word units via shared letter units could then lead to a further increase in overall lexical activation. Assuming that the activation of associates takes place at the semantic level, the first moment that NoA would have an impact is, so to say, one level later than for orthographic neighborhood density. Apart from the comparison with the ERP effect of orthographic neighborhood density, an ERP study on NoA is interesting in its own right and represents a valuable extension of the existing research on semantic processing in visual word recognition. The ERP, with its high temporal resolution, can give us information about the timing of the NoA effect. With the N400 we also have a well-studied component at hand that is related to semantic processing. Finding an effect of NoA on the N400 would reassure us that we are dealing with a semantic effect. To summarize, this study has two purposes. First, to test the semantic-level interpretation of the orthographic neighborhood density N400 effect as proposed by Holcomb et al. (2002) by comparing the ERP elicited by a orthographic neighborhood density manipulation and a semantic manipulation as exemplified by associative neighborhood Orthographic and Associative 12 density. The respective hypothesis can be formulated like this: If the semantic-level explanation of the orthographic neighborhood density N400 effect is correct, manipulations of orthographic and associative density should result in similar N400 effects. The second purpose is to find an electrophysiological signature of NoA, which would potentially provide valuable information on the time course of this effect.