Elsevier

NeuroImage

Volume 42, Issue 3, September 2008, Pages 1185-1195
NeuroImage

Modulation of brain activity by multiple lexical and word form variables in visual word recognition: A parametric fMRI study

https://doi.org/10.1016/j.neuroimage.2008.05.054Get rights and content

Abstract

Psycholinguistic research has documented a range of variables that influence visual word recognition performance. Many of these variables are highly intercorrelated. Most previous studies have used factorial designs, which do not exploit the full range of values available for continuous variables, and are prone to skewed stimulus selection as well as to effects of the baseline (e.g. when contrasting words with pseudowords). In our study, we used a parametric approach to study the effects of several psycholinguistic variables on brain activation. We focussed on the variable word frequency, which has been used in numerous previous behavioural, electrophysiological and neuroimaging studies, in order to investigate the neuronal network underlying visual word processing. Furthermore, we investigated the variable orthographic typicality as well as a combined variable for word length and orthographic neighbourhood size (N), for which neuroimaging results are still either scarce or inconsistent. Data were analysed using multiple linear regression analysis of event-related fMRI data acquired from 21 subjects in a silent reading paradigm. The frequency variable correlated negatively with activation in left fusiform gyrus, bilateral inferior frontal gyri and bilateral insulae, indicating that word frequency can affect multiple aspects of word processing. N correlated positively with brain activity in left and right middle temporal gyri as well as right inferior frontal gyrus. Thus, our analysis revealed multiple distinct brain areas involved in visual word processing within one data set.

Introduction

Word recognition poses particular challenges to cognitive neuroscientists: Information about specific words can be retrieved from a vast memory store, and combined in extremely flexible ways, within fractions of a second. Until recently, the neuronal networks subserving word recognition were often described by a relatively small set of brain areas, motivated by classical neurological models of language processing (e.g. Damasio and Geschwind, 1984, Geschwind, 1970), such as left inferior frontal, left inferior temporal or angular gyrus (Binder et al., 2005, Bookheimer, 2002, Price, 2000). However, the specific role of individual brain areas in the word recognition process is still a matter of debate. A common procedure is to focus on one or a few aspects of word recognition, and control for potentially confounding variables as well as possible. This has the disadvantage that usually only few variables can be studied within the same set of subjects, such that the “bigger picture” can only be created by comparing results across many different studies. It therefore introduces considerable variation with respect to recording and analysis techniques, tasks, stimulus and subject selection etc. It would obviously be desirable to be able to study as many psycholinguistic features as possible using the same group of subjects, study design and analysis technique. Furthermore, a recent study showed that activations reported for words compared to pseudowords might have resulted from deactivation for pseudowords, suggesting that results for factorial contrasts can be difficult to interpret (Mechelli et al., 2003).

A more promising way of disentangling the contributions of different parts of the networks would be to study their modulation by specific well-defined psycholinguistic properties. In this study, we therefore investigated the effect of several psycholinguistic parameters on metabolic brain activity, using multiple linear regression in the analysis of event-related fMRI data. Multiple linear regression analysis of neuroimaging data with respect to psycholinguistic variables has two main advantages over factorial approaches: 1) It uses information about the continuous distribution of values, e.g. of word frequencies for individual items. Thus, for continuous variables, regression designs may have greater statistical power than factorial designs (Cohen, 1983). 2) For factorial designs, “awkward” items on the extremes of the parameter distributions might have to be chosen (e.g. very high or low frequency items), which may not be representative for the stimulus population under study (Baayen et al., 1997, Ford et al., 2003). As Balota et al. (2004) put it: Regression analysis allows “the language, instead of the experimenter, to define the stimulus set”. Parametric analysis has already been applied in other research areas (e.g. picture naming, see Graves et al., 2007), but not in neuroimaging studies on visual word recognition. For the present study, an initial set of 21 psycholinguistic features was analysed with respect to its intercorrelation pattern. From these we constructed five composite variables related to different aspects of the visual word recognition process, 3 of which will be the focus of this paper. These entered a multiple regression analysis of the event-related fMRI response, which allowed us to determine their individual contributions to the signal.

One of the most extensively studied variables in visual word recognition is word frequency. Many studies have shown that high frequency words are responded to faster (Gernsbacher, 1984, Scarborough et al., 1977, Whaley, 1978) and generally produce lower amplitudes in electrophysiological (e.g. Assadollahi and Pulvermüller, 2003, Hauk and Pulvermüller, 2004a, Sereno et al., 1998) or metabolic (Carreiras et al., 2006, Chee et al., 2003, Fiebach et al., 2002, Fiez et al., 1999, Kronbichler et al., 2004) responses than their low frequency counterparts. It has been suggested that these effects reflect facilitation of early access to lexico-semantic information (Cleland et al., 2006, Hauk and Pulvermüller, 2004a). Word frequency therefore appears to be a crucial parameter for revealing core language areas involved in word recognition.

However, some authors have argued that effects of word frequency reflect post-access decision or verification processes. For example, McCann et al. (2000) reported that word frequency effects persisted in a dual-task paradigm, where a distractor task is assumed to interfere with early stages of word recognition processes. They interpreted this result as evidence for a late locus of word frequency effects. Similarly, Balota and Chumbley (1984) found word frequency effects to be larger in a lexical decision compared to a category verification and pronunciation task. They therefore suggested that these effects depend on the familiarity-based decision process, rather than word identification per se. The insensitivity of naming or lexical decision times to pseudohomophones (e.g. “brane”) with respect to base-word frequency (“brain”) has also been interpreted as evidence that this variable does not affect lexical access (Mccann et al., 1988). However, a recent experiment using dual-task methodology similar to McCann et al. (2000) demonstrated that word frequency does produce effects at early stages of word recognition (Cleland et al., 2006). Other authors also reported effects of word frequency even for the case where the task was chosen in order to minimise them, e.g. using very short exposure durations (Allen et al., 2005). This is in agreement with a number of recent electrophysiological studies that reported early (i.e. < 200 ms) effects of word frequency in the event-related potential or field (ERP/F) (Assadollahi and Pulvermüller, 2003, Dambacher et al., 2006, Hauk et al., 2006a, Hauk and Pulvermüller, 2004a, Sereno and Rayner, 2003). We conclude from these data that word frequency affects early stages of word processing, which does not exclude the possibility that it affects later stages as well, or that these effects can be modulated by task demands. A more detailed analysis of the brain areas affected by word frequency might therefore reveal lexico-semantic areas without the requirement to contrast word stimuli to a non-word control condition. Importantly, neuroimaging allows us to measure the brain response without the need of overt responses, as in lexical or semantic decisions. We therefore employed a silent reading task, rendering it unlikely that the effects of psycholinguistic variables in this study can be explained on the basis of decision or verification processes.

The brain areas modulated by word frequency are still a matter of debate. Several previous studies using factorial designs reported left inferior frontal activation for word frequency, in silent reading (Joubert et al., 2004, Kronbichler et al., 2004), reading aloud (Fiez et al., 1999) and in visual lexical decision tasks (Fiebach et al., 2002, Nakic et al., 2006). Right hemispheric brain structures modulated by word frequency were reported by only few studies, namely right inferior frontal cortex (Joubert et al., 2004, Nakic et al., 2006), and bilateral insula (Fiebach et al., 2002). Similarly, areas in left inferior temporal or fusiform gyrus were found to be modulated by word frequency in two studies (Joubert et al., 2004, Kronbichler et al., 2004). Further cortical areas associated with word frequency were left middle (Kronbichler et al., 2004) and superior (Fiez et al., 1999) temporal gyrus, as well as left precentral gyrus and bileratal occipital gyri (Kronbichler et al., 2004). This brief overview suggests some consistency across studies with respect to word frequency effects in left inferior frontal cortex, but considerable inconsistency with regard to other language-related areas, e.g. in inferior temporal cortex or in the right hemisphere. The situation is further complicated by findings in studies on the task-dependency of word frequency effects: In the study of Carreiras et al. (2006), word frequency effects were found in left frontal cortex only for lexical decisions, but not for reading aloud. Similarly, Chee et al. (2002) found left inferior frontal activation for low frequency words for semantic judgments, but not for silent reading. These results would endorse the view already formulated in the behavioural literature that word frequency effects are more related to task-specific decision processes, rather than lexico-semantic processing. This, however, is in contrast to several aforementioned studies that have reported word frequency effects in overt and covert reading, i.e. in tasks that do not require a decision.

In general, the areas that have been reported as being sensitive to word frequency correspond to “classical” language-related areas, such as left inferior frontal, middle/superior temporal and inferior temporal/fusiform cortex (e.g. Price, 2000). Note also that these activations were consistently larger for low compared to high frequency words, which corresponds well to the above-mentioned behavioural and electrophysiological results. This supports the view that word frequency effects have the potential to reveal the cortical network of lexico-semantic processing. Thus, word frequency effects are of great importance for research on the cortical basis of word processing, but existing neuroimaging data are as yet inconsistent. We therefore investigated word frequency effects on metabolic brain activation using a silent reading task and a sensitive parametric analysis.

Word frequency is correlated with a number of other variables that may play a significant role in word recognition. For example, it has long been known that longer words are generally of lower frequency than shorter ones (Whaley, 1978, Zipf, 1935). In order to allow us to identify brain areas involved in the retrieval of lexical information, it is necessary to rule out confounding variables that are potentially related to other processes, e.g. those underlying processing of the orthographic word form. We therefore included two variables into our study that reflect different aspects of letter string processing. This will also allow us to define the brain areas related to these variables in more detail.

The first surface form variable is orthographic typicality, defined by the frequencies of letter bi- and trigrams (e.g. “ert” occurs more frequently in written English than “cht”). This variable does not rely on semantic information, or even on knowledge whether a letter string is a word or not. Although this variable is often controlled for in neuroimaging experiments on word recognition, reports on actual effects for this variable are scarce. One fMRI study found larger activation for more “word-like” nonwords compared to nonwords with low frequency letter constellations in a letter detection task (Binder et al., 2006). Syllable frequency, which is related to typicality, has been shown to produce effects in a left anterior inferior temporal region in a lexical decision task, but in left anterior insula for reading aloud (Carreiras et al., 2006). A recent fMRI study found different areas in inferior temporal and fronto-insular cortex to be sensitive to different levels of orthographic typicality (Vinckier et al., 2007). Two recent ERP studies found the earliest effects of orthographic typicality around 100 ms after word onset, and localised them into left inferior temporal areas (Hauk et al., 2006a, Hauk et al., 2006b). Further data are necessary to corroborate these findings and specify the role of the corresponding brain areas in more detail.

Word length (usually quantified as number of letters) has been reported to be positively correlated with reaction times (Ellis, 2004, Weekes, 1997, Whaley, 1978). ERP/ERF studies found earliest effects of word length around 100 ms after word onset, with long words producing larger amplitudes than short ones (Assadollahi and Pulvermüller, 2003, Hauk and Pulvermüller, 2004a). In one recent PET study using a silent reading task, word length has been associated with early visual processing in fusiform and lingual gyrus (Mechelli et al., 2000).

The situation is further complicated by the fact the word length is generally negatively correlated with orthographic neighbourhood size (“N”), i.e. for a given word the number of words that can be created by exchanging only one letter (e.g. “cat” into “can”) (Coltheart et al., 1977). The pattern of results for N in behavioural tasks has been reported to differ between words and pseudowords: For words, higher N usually yields faster lexical decision times, while for pseudowords rejections take longer for higher N's (Forster and Shen, 1996, Sears et al., 1995). This pattern has also been found in the behavioural data of (Holcomb et al., 2002), while their ERP data showed the same increase of N400 amplitudes with N for both words and pseudowords. It has been concluded that N facilitates the lexical retrieval process by means of competition between orthographically similar words (Andrews, 1989, Andrews, 1997, Grainger and Jacobs, 1996). However, this interpretation has been challenged by other researchers, who found N effects to be absent in a semantic decision task (Forster and Shen, 1996), or to depend on the matching between word and non-word stimuli (Siakaluk et al., 2002). These doubts are further supported by the fMRI results of Binder et al. (2003), who did not find any brain areas for which activation significantly increased with neighbourhood size. Instead, they found that higher N decreased activation to words in left prefrontal, angular and ventrolateral temporal cortex. In contrast, Fiebach et al. (2007) found differential effects of N for words and pseudowords in medial prefrontal and mid-dorsolateral cortex. Because these areas are commonly related to executive control functions rather than lexico-semantic processing, the authors argue that effects of N might arise only at a late post-lexical level. Effects of N are therefore of great interest for psycholinguistic theories of lexical access, and more data are needed to establish its effect on brain activation. In our study, N was highly negatively correlated with word length (i.e. number of letters). In order to rule out both N and word length as confounds for effects of word frequency and orthographic typicality, we combined these two variables into one regressor variable (correlating positively with length and negatively with N). We did not attempt to fully disentangle the effects of N and word length in this study, because our stimulus set was not optimised for this purpose. However, the brain areas modulated by this variable can be compared to those found in previous studies reviewed above.

In the present study, we aimed at corroborating and extending previous neuroimaging results on variables related to lexico-semantic and orthographic processing, i.e. word frequency, orthographic typicality, and a combined variable for orthographic neighbourhood size and word length. We chose a silent reading task in order to minimise effects related to task-dependent decision or verification processes, as in several previous studies (Hauk et al., 2004, Joubert et al., 2004, Kronbichler et al., 2004, Mechelli et al., 2003). Data were analysed using a multiple linear regression design in order to extract results for several variables from the same data set, to gain statistical sensitivity, and to avoid problems associated with classical factorial designs.

Section snippets

Stimuli and experimental design

250 mono-syllabic and mono-morphemic English word stimuli were employed in the study, which mainly contained concrete action-related and visual-related words (see Hauk et al., 2004). 150 baseline trials consisting of strings of hash marks varying in length were interspersed among the word stimuli. The average length of words and hash marks was matched (2-tailed unpaired t-test p > 0.1). In addition, 50 null events were included in which a fixation cross remained on the screen. The stimulus set

Effects of all words vs baseline

In the first analysis step, we contrasted activation for all words to the baseline condition (strings of hash marks) in a factorial design. Several activation spots correspond to classical language areas in the left hemisphere and will be compared to activation obtained for the other psycholinguistic variables reported below. These occurred in left fusiform, left precentral, left middle temporal and left inferior frontal gyrus (Fig. 2 and Table 1). Further activation spots appeared exclusively

Effects of Frequency and All Words

Word frequency has been widely reported to be negatively correlated with reaction times (e.g. Cleland et al., 2006), ERP (e.g. Hauk and Pulvermüller, 2004a) or BOLD amplitudes (e.g. Fiebach et al., 2002). Based on behavioural and electrophysiological results, it has been argued that these effects reflect facilitated access to lexical information (Allen et al., 2005, Assadollahi and Pulvermüller, 2003, Cleland et al., 2006, Hauk and Pulvermüller, 2004a, Sereno and Rayner, 2003). Determining

References (81)

  • EllisA.W.

    Length, formats, neighbours, hemispheres, and the processing of words presented laterally or at fixation

    Brain Lang.

    (2004)
  • FiebachC.J. et al.

    Inhibition and facilitation in visual word recognition: prefrontal contribution to the orthographic neighborhood size effect

    NeuroImage

    (2007)
  • FiezJ.A. et al.

    Effects of lexicality, frequency, and spelling-to-sound consistency on the functional anatomy of reading

    Neuron

    (1999)
  • FristonK.J. et al.

    Event-related fMRI: characterizing differential responses

    NeuroImage

    (1998)
  • FristonK.J. et al.

    Characterizing dynamic brain responses with fMRI: a multivariate approach

    NeuroImage

    (1995)
  • HaukO. et al.

    The time course of visual word recognition as revealed by linear regression analysis of ERP data

    NeuroImage

    (2006)
  • HaukO. et al.

    Somatotopic representation of action words in human motor and premotor cortex

    Neuron

    (2004)
  • HaukO. et al.

    Effects of word length and frequency on the human event-related potential

    Clin. Neurophysiol.

    (2004)
  • JoubertS. et al.

    Neural correlates of lexical and sublexical processes in reading

    Brain Lang.

    (2004)
  • KronbichlerM. et al.

    The visual word form area and the frequency with which words are encountered: evidence from a parametric fMRI study

    NeuroImage

    (2004)
  • McCandlissB.D. et al.

    The visual word form area: expertise for reading in the fusiform gyrus

    Trends Cogn. Sci.

    (2003)
  • NakicM. et al.

    The impact of affect and frequency on lexical decision: the role of the amygdala and inferior frontal cortex

    NeuroImage

    (2006)
  • OldfieldR.C.

    The assessment and analysis of handedness: the Edinburgh inventory

    Neuropsychologia

    (1971)
  • PoldrackR.A. et al.

    Functional specialization for semantic and phonological processing in the left inferior prefrontal cortex

    NeuroImage

    (1999)
  • PriceC.J. et al.

    The myth of the visual word form area

    NeuroImage

    (2003)
  • SerenoS.C. et al.

    Measuring word recognition in reading: eye movements and event-related potentials

    Trends Cogn. Sci.

    (2003)
  • Thompson-SchillS.L.

    Neuroimaging studies of semantic memory: inferring "how" from "where"

    Neuropsychologia

    (2003)
  • Tzourio-MazoyerN. et al.

    Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain

    NeuroImage

    (2002)
  • VinckierF. et al.

    Hierarchical coding of letter strings in the ventral stream: dissecting the inner organization of the visual word-form system

    Neuron

    (2007)
  • WhaleyC.

    Word–nonword classification time

    J. Verbal Learn. Verbal Behav.

    (1978)
  • AllenP.A. et al.

    Evidence for an activation locus of the word-frequency effect in lexical decision

    J. Exp. Psychol. Hum. Percept. Perform.

    (2005)
  • AndrewsS.

    Frequency and neighborhood effects on lexical access — activation or search

    J. Exper. Psychol., Learn., Mem., Cogn.

    (1989)
  • AndrewsS.

    The effect of orthographic similarity on lexical retrieval: Resolving neighborhood conflicts

    Psychon. Bull. Rev.

    (1997)
  • AssadollahiR. et al.

    Early influences of word length and frequency: a group study using MEG

    NeuroReport

    (2003)
  • BaayenR.H. et al.

    The CELEX Lexical Database (CD-ROM)

    (1993)
  • BalotaD.A. et al.

    Are lexical decisions a good measure of lexical access? The role of word frequency in the neglected decision stage

    J. Exp. Psychol. Hum. Percept. Perform.

    (1984)
  • BalotaD.A. et al.

    Visual word recognition of single-syllable words

    J. Exp. Psychol. Gen.

    (2004)
  • BinderJ.R. et al.

    Neural correlates of lexical access during visual word recognition

    J. Cogn. Neurosci.

    (2003)
  • BookheimerS.

    Functional MRI of language: new approaches to understanding the cortical organization of semantic processing

    Annu. Rev. Neurosci.

    (2002)
  • BrettM. et al.

    Region of interest analysis using an SPM toolbox

  • Cited by (67)

    • Modulation of brain activity by psycholinguistic information during naturalistic speech comprehension and production

      2022, Cortex
      Citation Excerpt :

      In addition, factorial designs frequently do not exploit the full range of values available for continuous variables. To combat this, an alternative approach is to investigate the neural response to the property of interest in a parametric fashion and some studies have done this with great success (Graves, Desai, Humphries, Seidenberg, & Binder, 2010; Hauk, Davis, & Pulvermuller, 2008; Mummery, Ashburner, Scott, & Wise, 1999; Price et al., 1992; Wise et al., 2000). In recent years, researchers have often moved beyond univariate analysis of single-voxel responses, in favour of sophisticated multivariate pattern analyses (MVPA) (Kriegeskorte & Kievit, 2013; Kriegeskorte, Mur, & Bandettini, 2008; Norman, Polyn, Detre, & Haxby, 2006).

    View all citing articles on Scopus
    View full text