Behavioral and neurocognitive studies of reading are increasingly being enlisted for their potential contribution to efforts to develop evidence-based interventions for dyslexic readers [Gabrieli, 2009
]. On a theoretical level, reading researchers have long debated the extent and nature of the role of phonological processing in visual word recognition. The present study offers an initial neurobehavioral investigation of this issue with particular reference to the variable of phonological neighborhood density. Phonological neighborhood density (PND) is a measure of neighborhood size that is based on the number of words that can be generated by replacing a phoneme in a target word with another phoneme in the same position [Yates, 2005
]. For example, phonological neighbors of the word urge are edge, age, earl, earn, and earth. Phonological neighborhood density effects provide potentially important insights into the organization and operation of the mental lexicon [Grainger et al., 2005
; Yates, 2005
The nature of the phonological neighborhood effect is currently not well understood. In one of the first studies of this variable, Yates et al. 
examined phonological neighborhood density (PND) while controlling for orthographic neighborhood characteristics. They found a facilitative effect on a lexical decision task; that is, participants were faster at making lexical decisions to words with many vs. few phonological neighbors [Yates et al., 2004
]. Using a different set of stimuli, Yates 
again obtained a facilitative phonological neighborhood density effect in lexical decision, as well as in a naming task and a semantic categorization task.
Nevertheless, the presence of an apparent facilitative effect of phonological neighborhood density runs contrary to predictions of interactive activation models of visual word recognition. These models propose that inhibitory connections exist between lexical representations within a given layer of the mental lexicon [Andrews, 1997
; Coltheart et al., 2001
; Grainger and Jacobs, 1996
]. Similarly, the cross-code consistency view of lexical representation proposed by Grainger et al. 
argues against a facilitative effect of phonological neighborhood density. The notion of cross-code consistency was invoked by Grainger et al. to account for a pattern of findings they observed when examining the relationship between orthographic neighborhood density and phonological neighborhood density in visual word recognition. Orthographic neighborhood density refers to the number of words that can be generated by replacing one letter from a target word in the same letter position [Coltheart et al., 1977
]; for example, gap, cup, and cat are all orthographic neighbors of the target word, cap. In a lexical decision task with readers of French, Grainger et al. 
found that the effect of phonological neighborhood density was facilitative for words with high orthographic neighborhood density but was inhibitory for words with low orthographic neighborhood density.
According to the cross-code consistency account proposed by Grainger et al. 
, words with high orthographic and high phonological neighborhood density and words with low orthographic and low phonological neighborhood density tend to have more consistent orthographic to phonological representations compared with words that are high in one type of neighborhood density but low in the other. Higher cross-code consistency is in turn associated with faster response times. In this view, words that have a high orthographic neighborhood density will show more consistent cross-code mapping if they also have high phonological neighborhood density, which should, in turn, lead to faster response latencies and a facilitative phonological neighborhood density effect, as has been reported by Yates et al. 
and Yates 
. Conversely, when words have a low orthographic neighborhood density, the higher the phonological neighborhood density of a word, the less consistent the cross-code mapping and thus the slower the response should be. This should result in an inhibitory phonological neighborhood density effect. This latter prediction has not been tested to date. The present study, therefore, tested the prediction of an inhibitory phonological neighborhood density effect when orthographic neighborhood density is kept very low.
In addition to testing cross-code consistency-based predictions, this study sought to explore whether there is a neural correlate of phonological density effects detectable using neuroimaging techniques. Recent developments in brain activity recording methods have made it possible for researchers to examine the neural correlates of a number of psycholinguistically-relevant variables influencing word recognition. Two neuroimaging studies using magnetoencephalography (MEG) have studied phonological density effects. Pylkkänen et al. 
found that, of the different components studied, the M350 response component was particularly sensitive to phonological neighborhood density [see also Pylkkänen and Marantz, 2003
]. Pylkkänen et al. 
also found that people responded to words with high phonological neighborhood density more slowly than they did to words with low phonological neighborhood density. However, this finding was not replicated in a subsequent study by Stockall et al. 
, in which a null effect of phonological neighborhood density was obtained. It is important to note that these MEG studies of phonological neighborhood density did not control for orthographic neighborhood size, making it difficult to know whether the effects were solely due to phonological neighborhood density.
This study explored neural correlates of phonological neighborhood density effects using the hemodynamic based measure of near infrared spectroscopy (NIRS). To our knowledge, no previous hemodynamic study in visual word recognition has specifically examined this variable. Hemodynamic changes in the brain can potentially provide corroborating evidence for the existence of a distinct phonological neighborhood density effect. NIRS measures changes in the concentration of oxy-hemoglobin and deoxy-hemoglobin in the brain regions of interest by shining near-infrared light (650–950 nm) through the scalp and analyzing the characteristics of its subsequent absorption and scattering. When a brain area engages in a mental operation, an increase in the concentration of the oxy-hemoglobin should be observed [see Strangman et al., 2002a
, for a review]. Previous studies suggest that NIRS data are highly consistent with fMRI data [Strangman et al., 2002b
]. NIRS is a portable and affordable alternative to fMRI, and provides reasonable temporal and spatial resolution [Hochman, 2000
; Strangman et al., 2002a
]. The temporal resolution of the NIRS system is a 200 Hz sampling rate, which is better than fMRI but is worse than that of EEG. The spatial resolution of NIRS can be as precise as 5 mm, which is better than EEG but is worse than fMRI.
Because it is not possible to monitor blood oxygenation changes in the whole brain using NIRS, one needs to identify a brain region of interest (ROI). Two recent meta-analyses of functional neuroimaging studies using fMRI have identified several brain areas related to phonological processing [Bolger et al., 2005
; Tan et al., 2005
]. With English words, brain activation was found to be more pronounced in the left dorsal temporoparietal system, including posterior regions of the left superior temporal gyrus (BA22), the angular gyrus (BA39), and the supramarginal gyrus (BA40). These areas are thought to mediate grapheme-to-phoneme conversion and fine-grained phonemic processing in alphabetic writing systems [Bolger et al., 2005
; Tan et al., 2005
]. Because phonological neighborhood effects, as conceptualized by the cross-code consistency account, relate to fine-grained phonemic processing and to grapheme-to-phoneme conversion, the brain areas of BA22/39/40 were selected as the ROI in this study which examined whether hemodynamic measures are sensitive to phonological neighborhood density effects in English readers.
To examine the effect of phonological neighborhood density effect as cleanly as possible, we controlled for orthographic neighborhood density by keeping it very low. This allowed us to test the assumption of interactive activation models of inhibitory connections between similar representations and the prediction based on cross-code consistency of an inhibitory effect of phonological neighborhood density. Moreover, whereas previous NIRS studies of language processing have used a block design approach, we used an event-related design. Although a block design provides superior statistical power to detect subtle differences [Friston et al., 1999
], an event-related design has the advantage of allowing randomized presentation of stimuli. This is desirable in investigations where the carryover effects typical of block designs may introduce response artifacts [Chee et al., 2003
]. Using an event-related design in the present study also allowed us to determine whether event-related designs using NIRS are sensitive to the subtle changes typically observed in word recognition experiments.