An fMRI Study of English and Spanish Word Reading in Bilingual Adults

Abstract

Reading relies on a left-lateralized brain system, including occipito-temporal (OTC), temporo-parietal, and inferior frontal (IFC) cortices. Neuroimaging studies have investigated whether activation in these cortices is modulated by a language’s orthographic depth (consistency of grapheme-to-phoneme conversion). In Spanish-English bilinguals, some but not all studies have reported activation differences between the two languages during reading. Here, we studied Spanish-English early bilingual adults living in the United States (N = 25; 17 female, 8 male). We examined local activity, functional connectivity, and spatially distributed activity patterns during English and Spanish word reading. We found overlap in local activity for the two languages in the left IFC, but no differences in activation between them and few differences in functional connectivity (none of which were in pairs of regions known to be involved in reading); yet, there were spatially distributed patterns of brain activity that differentiate English and Spanish in regions of bilateral cerebellum/left OTC, the left superior occipital gyrus, the left IFC, and the left medial frontal gyrus. Taken together, we found no evidence for differences in local activation or functional connectivity during English versus Spanish word processing in regions known to be involved in reading, yet found evidence for the Spanish-English bilingual brain to distinguish between the two languages.

1. Introduction

Reading is a critical skill for academic and personal success. As such, studies have tried to understand the behavioral skills that support reading acquisition and its neurobiological bases. From the behavioral standpoint, it is widely accepted that grapheme-to-phoneme conversion (GPC), facilitated by phonological coding, is used to read novel words (Coltheart et al., 1993; Harm & Seidenberg, 1999; Seidenberg & McClelland, 1989; Wagner & Torgesen, 1987). The recognition of familiar and frequently encountered word forms, however, involves more direct mapping, facilitated by orthographic awareness (Badian, 1994; Badian, 2001). Therefore, word reading has been conceptualized under the so-called Dual Route Theory (Coltheart et al., 2001; Coltheart et al., 1993; Paap & Noel, 1991; Warrington & Shallice, 1980), involving an indirect, sublexical route that requires visual words to be transformed into their auditory counterparts (i.e., GPC), and a direct, lexico-semantic route that relies on the direct mapping between the visual word-forms (orthography) and their meanings. Neuroimaging studies have led to a brain-based model of word reading (Pugh et al., 2001) composed of three main areas in the left hemisphere: the occipito-temporal cortex (OTC) for memory-based visual word-form recognition (Cohen et al., 2000; Dehaene & Cohen, 2011; Dehaene et al., 2002); the temporo-parietal cortex (TPC) for GPC (Richlan, 2012, 2014); and, the inferior frontal cortex (IFC) for semantic processing, phonological decoding, and articulatory recoding of print (Bookheimer, 2002; Cutting et al., 2006; Price, 2012; Pugh et al., 2001). Meta-analyses have shown significant convergence across studies of reading in adults in these brain areas (Martin et al., 2015; Turkeltaub et al., 2002).

A revised version of this original brain-based model was advanced by Sandak et al. (2004), who argued, based on findings from neuroimaging studies, that both the IFC and the OTC are tuned for phonological processing. Specifically, the inferior frontal gyrus (region of the IFC) and the visual word form area (VWFA) (region of the OTC) are engaged during phonological priming and pseudoword processing compared to words, and show repetition-related reductions following phonologically analytic training. Of note, these authors highlighted the “new” role of the VWFA for phonological processing as logical, since individuals who struggle with reading fail to adequately develop this posterior ventral system, in addition to having phonological deficits. Sandak et al. (2004) further posited that some regions of the OTC (middle and inferior temporal gyri) and TPC (angular gyrus) are more activated during lexico-semantic processing. Moreover, some reports reveal that reading also relies on right-hemisphere regions, mainly for logographic writing systems (Hart et al., 2000; Mayall et al., 2001; Pugh et al., 1997; Tan, Feng et al., 2001; Tan et al., 2001). Taken together, this revised brain-based model is aligned with cognitive models of reading and recognizes that a range of pathways are utilized depending on the familiarity and GPC consistency of the word being processed (Frost, 2012).

Orthographic depth is the transparency, or consistency, of the GPC in a written language (Schmalz et al., 2015). In general, behavioral data have shown that reading words in languages with consistent GPC is easier (Ziegler et al., 2010; Ziegler & Goswami, 2005), and that reading is acquired at different rates in different countries due to the variability in each written language’s orthographic depth (Seymour, Aro, & Erskine, 2003). Neuroimaging studies have shown that the brain’s local activity during word reading is modulated depending on the language’s orthographic depth. The first study addressing this question involved two groups of monolinguals that were users of English or Italian (Paulesu et al., 2000). The authors found relatively greater activation in the left posterior inferior temporal gyrus and the left anterior inferior frontal gyrus in English (deep orthography) readers compared to Italian readers. They suggested that this observation may be due to the inconsistent mapping of English, requiring greater reliance on word-form recognition and lexico-semantic processing associated with the direct route. In contrast, there was relatively greater activity in the left posterior superior temporal gyrus in Italian (shallow orthography) readers compared to English readers. Paulesu et al. (2000) proposed that the highly consistent GPC in Italian might involve more phonological processing, and thus greater use of the indirect route. Because the study was conducted by comparing two groups of monolinguals, it raises the question whether the same observations would hold in bilinguals.

This question has been addressed in studies of Spanish-English early bilingual adults (Meschyan & Hernandez, 2006; Jamal et al., 2012; Hernandez, Woods, & Bradley, 2015). Meschyan & Hernandez (2006) found that English word reading, relative to Spanish, engaged more brain regions in the parietal and occipital lobes, potentially due to the visual analysis that is required given the many exceptions to the GPC rules. On the other hand, for Spanish, a language with shallow orthographic depth, the right supplementary motor area/cingulate, putamen, and insula, as well as the left superior temporal gyrus were more activated relative to English. Similar to Paulesu et al. (2000), the authors concluded that reading in a shallow orthography may be more phonologically driven, since the phonological forms are easily retrieved. Jamal et al. (2012) found that English word reading (relative to Spanish) activated the left middle frontal gyrus extending to the superior frontal gyrus, whereas Spanish word reading (relative to English) activated the left middle temporal gyrus extending into the superior temporal sulcus. The authors interpreted these results as English word reading requiring phonological decoding (given its highly inconsistent GPC), while Spanish word reading engaging regions related to semantic processing (due to low demand on letter-sound mapping), thereby casting a different perspective on the potential role of orthographic depth on the brain’s functional bases for word reading. More recently, Hernandez, Woods, & Bradley (2015) reported no difference in local activity between English and Spanish word reading in Spanish-English bilinguals. Taken together, three studies have investigated language-specific activation attributed to orthographic depth, and reached different conclusions. This inconsistency underscores the need for further studies.

Several aspects of these three studies could have led to the discrepancies in their results. One is the participants’ language proficiency; the participants in Meschyan & Hernandez (2006) and Hernandez, Woods, & Bradley (2015) deliberately did not have equal proficiency (spoken and written comprehension) in the two languages (less proficient in Spanish), whereas those in Jamal et al. (2012) were overall equally proficient. This factor, however, does not explain the pattern of results. A second aspect to consider is any differences in the tasks employed in each study. Meschyan & Hernandez (2006) used a single-word reading task in which participants were instructed to silently read words. Jamal et al. (2012) used an “implicit” single-word reading task, where participants pressed a button if the word had a tall letter (such as t, l, h) and another button if the word did not have a tall letter. This feature-detection task has been shown to activate brain regions known to be associated with aloud reading (Price, Wise, & Frackowiak, 1996; Turkeltaub et al., 2003). Finally, Hernandez, Woods, & Bradley (2015) used a similar single-word reading task as Meschyan & Hernandez (2006), but this time participants were instructed to press a button after silently reading each single word. While there are some differences across these three tasks, when considered in the greater context of published word reading tasks, they are fairly similar to each other. They are also similar to other published studies of reading, which are mostly limited to the presentation of single words (avoiding connected text). Further, none of them involved lexical decision making, which can affect cross-linguistic studies (Wang et al., 2015).

An additional aspect to consider is the sample size and statistical power. Meschyan & Hernandez (2006) and Jamal et al. (2012) each had 12 bilingual participants, while Hernandez, Woods, & Bradley (2015) had 20, the first two studies perhaps having cohorts too small to draw strong conclusions from. Moreover, each study used different statistical approaches, with Jamal et al. (2012) using the most liberal parameters. Therefore, it is difficult to say whether word reading in alphabetical languages with different orthographic depth, such as English and Spanish, relies on the same brain regions or not in bilinguals. This question was tested in the present study, using a larger cohort of Spanish-English early bilingual adults with equal overall proficiency in both languages (spoken and written comprehension) and measuring local activity as well as other analytical approaches, namely functional connectivity and multivariate pattern analysis.

Reading engages a distributed system of brain regions that have distinct functions and interact. As such, studies of word reading have benefited from investigations into functional connectivity, the correlation between distant activity (Biswal et al., 1995; Friston, 1994, 2011). Functional connectivity between reading-related regions has been shown during various reading tasks (Bitan et al., 2006; Hampson et al., 2006; Horwitz, Rumsey, & Donohue, 1998; Mechelli et al., 2005; Pugh & Mencl, 2000). Even at rest, many of the brain regions observed in activation studies of word reading are positively correlated with each other and negatively correlated with regions in the default mode network (Koyama et al., 2011, 2010). As reading relies on a cohesive, functionally connected brain system rather than a collection of brain regions working independently (Price, 2012), the current study sought to address functional connectivity in the context of English and Spanish word reading in bilinguals.

To date, little is known about whether the functional interactions between the constituents of the reading network are different between alphabetical languages with a deep versus a shallow orthographic depth. Moreover, there are no neuroimaging studies investigating the functional connectivity supporting English and Spanish word reading in early bilingual adults. Oliver, Carreiras, & Paz-Alonso (2016) examined functional connectivity differences associated with orthographic depth in Spanish-speaking late bilingual adults. This investigation, however, compared two groups based on their second languages differing in orthographic depth: English (deep) versus Basque (shallow). In the present study, we asked whether the functional network used for reading words in a deep (English) orthography differs from that used when reading words in a shallow (Spanish) orthography within the same group of participants.

In addition to examining brain activity with univariate analysis, the current investigation also examined the spatially distributed patterns of activity with multivariate pattern analysis (MVPA) (Haxby, 2012; Haxby et al., 2001; Haynes & Rees, 2005; Kamitani & Tong, 2005). This analysis approach reveals the pattern of activation across multiple voxels. Specifically, we used MVPA to test whether the two conditions of interest (English word reading vs. Spanish word reading) can be distinguished from one another on the basis of activity patterns that emerge for a set of voxels. Importantly, one can detect these patterns, even if the average level of activity does not differ between these two conditions (for a review, Tong & Pratte, 2011). For example, Hsu, Jacobs, & Conrad (2015) used an MVPA to examine spatially distributed activation patterns differentiating reading texts in English versus German in German-English late bilingual adults. The primary goal of their study was to test if the emotional response to a passage was the same in L2 (English) as in L1 (German) by varying the emotional content of the passages. One of their MVPAs, however, tested for patterns of activity distinguishing between English and German (without regard for the emotional content) and found results in cortical and subcortical areas. These observations were present in addition to differences in signal intensity reported between these two languages using a traditional univariate analysis. In contrast, a recent study tested whether Chinese word reading exhibits distinct neural representations from that of English word reading in Chinese-English late bilingual adults, using MVPA (Xu et al., 2017). The authors found spatially distributed patterns of activity in reading-related regions distinguishing between Chinese and English, even though the univariate activation analysis found little difference. Since it is currently unknown whether there are distinct neural representations for word reading in two alphabetic languages with different orthographic depth, the present study employed a classification MVPA to test whether there are spatially distributed patterns of brain activity that distinguish between reading words in a deep (English) versus a shallow (Spanish) orthography.

Taken together, the goal of the present study was to test whether reading words in languages with different orthographic depth is supported by different brain regions at the local level (activity), as well as at the network level (functional connectivity) in Spanish-English bilingual adults. We also examined spatially distributed activity patterns for the two languages (MVPA). Our participants were early bilinguals (i.e., both languages were acquired by age 6) with equal proficiency (spoken and written comprehension skills were balanced between the two languages). Furthermore, they were cultural bilinguals, meaning that they acquired both languages from their parents and their environment, rather than being cultural monolinguals that excelled at learning a second language.

Consistent with our prior study (Jamal et al., 2012), during English word reading (compared to Spanish), we expected to observe relatively greater local activity in regions of the indirect/sublexical route (i.e., TPC and IFC), and thus stronger positive functional connectivity between them. On the other hand, we predicted relatively greater local activity in regions of the direct/lexico-semantic route (i.e., OTC and IFC) and stronger positive functional connectivity between them during Spanish word reading (compared to English). We recognized, however, that the exact opposite may occur, as in Meschyan & Hernandez (2006), or that we find no differences at all, like Hernandez, Woods, & Bradley (2015) did. Lastly, we also expected to find spatially distributed patterns of brain activity in regions of the OTC, TPC, and/or IFC differentiating between these two orthographies. In other words, we predicted to observe distinct neural representation for English and Spanish word reading in some or all of these three cortical areas.

2. Material and Methods

2.1. Participants

Thirty-one healthy, young adults participated in this study. All participants were Spanish-English early bilinguals and biliterates (21 females; 10 males), who had acquired both languages orally by age 6 and via written form during schooling. None of the participants had major exposure to languages other than English and Spanish. One participant’s data were not analyzed because their anatomical image was not available, and another’s because the functional data were corrupted. Four participants’ data were excluded from the analyses because of excessive head motion during the scans, leaving a final group of 25 participants (17 females and 8 males with average age of 22.1 years; Table 1). All participants had normal or corrected-to-normal vision, and none of them had a history of neurological impairment or learning difference. Experimental procedures were approved by the Georgetown University Institutional Review Board, and written informed consent was secured from all participants at the beginning of the study. A subset of these participants’ data have been reported in the context of mean brain activation during English and Spanish word reading (Jamal et al., 2012) and structural (gray matter volume) differences compared to English monolinguals (Olulade et al., 2016). Specifically, 10 of the 25 final participants overlap with the Jamal et al. (2012) study.

Table 1.

Description of the participants and in-scanner task performance.

	M (SEM)	English M (SEM)	Spanish M (SEM)	p value
N	25	-	-	-
Sex (female/male)	17/8	-	-	-
Age (years)	22.1 (0.6)	-	-	-
Age range (years)	18.4 – 28.5	-	-	-
Language self-assessment (scale: 1–7; 7 = native-like competence)
	Listening Comprehension	6.7 (0.1)	6.7 (0.1)	n.s.
	Speaking	6.6 (0.1)	6.4 (0.2)	n.s.
	Reading Comprehension	6.6 (0.1)	6.2 (0.2)	n.s.
	Writing	6.7 (0.1)	6.0 (0.2)	< 0.05
	Overall Proficiency	6.6 (0.1)	6.3 (0.1)	n.s.
Language background
	Age of first exposure (years)	3.3 (0.4)	0.3 (0.1)	< 0.05
	Formal study (years)	14.9 (1.1)	12.0 (1.3)	n.s.
	Currently spoken per day (%)	72.2 (4.4)	27.8 (4.4)	< 0.05
Implicit Reading Task
	RW/FF Accuracy Difference (% correct)	0.8 (0.1)	−0.9 (0.2)	n.s.
	RW/FF Response Time Difference (ms)	−36.3 (12.1)	2.7 (10.6)	< 0.05

Abbreviations: M = mean; SEM = standard error of the mean; n.s. = non-significant (p = or > 0.05).

2.2. Behavioral Measures for Study Criteria

All participants engaged in psychoeducational testing to ensure that reading abilities in English and Spanish were in the typical range (standard score of 85 or above). To assess single-word reading in English, we used the Letter-Word Identification and Word Attack subtests from the Woodcock–Johnson III: Tests of Achievement (Woodcock, McGrew, & Mather, 2001). Single-word reading in Spanish was assessed with the “Identificación de letras y palabras” and “Análisis de palabras” subtests from the “Batería III Woodcock-Muñoz: Pruebas de aprovechamiento” (Munoz-Sandoval et al., 2005), which are the Spanish equivalents of the two aforementioned English subtests. All participants completed a language proficiency self-assessment questioner, as used in Meschyan & Hernandez (2006). It included questions about spoken and written comprehension proficiency in English and Spanish at the time of the study, as well as language background (Table 1).

2.3. fMRI Word Reading Task

Adapted from the original work of Price, Wise, & Frackowiak (1996) and as reported in previous studies examining the functional brain bases of reading (Evans et al., 2016; Olulade et al., 2013; Turkeltaub et al., 2003), we used an Implicit Reading (IR) Task in a functional magnetic resonance imaging (fMRI) block design paradigm. Inside the MRI scanner, participants were required to view single words in the middle of a screen and press one of two buttons to indicate the presence (right thumb) or absence (left thumb) of an ascender, or tall letter, in each word (e.g., “vowel” [tall letter] or “cream” [no tall letter] in English; and, “hebra” [tall letter] or “mueca” [no tall letter] in Spanish). During the active control condition, participants completed the same task, but this time in response to false-font strings (i.e., symbols created from real letter segments and rearranged with no resemblance to real letters). This task involved visual processing, response selection, and motor response, but not the orthographic, phonological, or semantic processing that automatically occurs when viewing real words. The stimulus presentation rate was 1 per 4.2 s (word duration = 1.2 s, fixation duration = 3.0 s).

Task stimuli consisted of 80 words (40 English and 40 Spanish) and 80 false-font strings (40 English equivalents and 40 Spanish equivalents); see Supplemental Table 1. For the Real Words (RW) condition, monomorphic, mono- or disyllabic five-letter, low-frequency (English: KF 8.05, SD 6.03; Spanish: LEXESP 7.95, SD 5.16) words were individually shown in black Arial font on a white screen. English words were selected from the MRC Psycholinguistic Database (Coltheart, 1981), and Spanish words were selected from the “BuscaPalabras” database (Davis & Perea, 2005). To avoid ambiguity, words with the letters i or j and words containing orthographic accents were not included.

Stimuli in the False Fonts (FF) condition were also shown in black on a white screen. These characters were matched to the RW condition stimuli for length and position of ascenders (e.g., l, t, b) and descenders, or hanging letters (e.g., y, p, g), in both languages. There was not, however, a strict correspondence between false-font characters and letters, to avoid the participants assigning phonemes to false-font strings. Although the RW condition does not explicitly demand the participants to process orthography, phonology, and semantic, these aspects of reading are processed implicitly, resulting in brain function that is similar to that published during explicit word reading tasks (Price et al., 1996; Brunswick et al., 1999; however, see Vogel, Petersen, & Schlaggar, 2013).

All participants were trained on the IR Task before the scanning session. Then, each participant was scanned during four runs (each run duration = 4 min, 27 s), two runs per language, with English and Spanish runs counterbalanced across participants. Each run consisted of two blocks with real words and two blocks with false-font strings, each block lasting 42 s. In addition, blocks of crosshair fixation occurred between the RW and FF blocks and lasted 18 s; crosshair fixation blocks were also used at the beginning (24 s) and at the end (21 s) of each run.

2.4. In-scanner Task Performance and Post-scanning Stimuli Recognition Test

While the participants were performing the tasks inside the MRI scanner, we collected responses on task accuracy (correctly answering whether the real word or false font stimuli had a tall feature or not) and response time.

Participants also completed a forced-choice pencil-and-paper recognition test in each language after the scanning session (Turkeltaub et al., 2003). This test measured their ability to recognize real words presented during the scan (versus foils), with the expectation that real words would be recognized more often than the false-font stimuli. Specifically, participants were presented with a list of 80 real words and 80 false fonts per language, half of which had been presented in the scanner, and asked to indicate which ones they had seen during the scan.

2.5. fMRI Data Acquisition and Preprocessing

MRI images were acquired on a Siemens Vision Magnetom 3.0-Tesla scanner with a circularly polarized head coil in the Center for Functional and Molecular Imaging at Georgetown University. For each participant, one high-resolution three-dimensional T1-weighted image was obtained for alignment of functional images with structure. For each experimental run, 89 whole-head echo planar imaging volumes were compiled under the following acquisition parameters: 3.0 s TR, 30 ms TE, 192 mm FOV, 50 contiguous axial slices, 2.8 mm slice thickness (0.2 mm inter-slice gap), 64×64 matrix (3.0 mm cubic voxels).

With a total of four runs (two runs per language) per participant, standard preprocessing steps, including slice-time correction, realignment, coregistration to structural image, normalization, and smoothing, were employed using the CONN toolbox v.18.a (https://www.nitrc.org/projects/conn) (Whitfield-Gabrieli & Nieto-Castanon, 2012). Differences in timing of slice acquisition were corrected and functional images were realigned to the first volume by means of rigid-body motion transformation. Normalized, bias-corrected T1-weighted images were generated and segmented into gray matter, white matter, and cerebrospinal fluid. Templates were based on the Montreal Neurological Institute (MNI) stereotaxic space (Cocosco et al., 1997), an approximation of Talairach space (Talairach & Tournoux, 1988). During normalization, the volumes were sampled to 3-mm cubic voxels. Global mean intensity and motion outliers (i.e., scan-to-scan motion greater than a threshold of 0.75 mm, which is 25% of the voxel size) in the fMRI time series were identified using the Artifact Detection Tools in CONN (ART; https://www.nitrc.org/projects/artifact_detect/), and all detected outliers were entered as regressors of no interest in the fMRI model. Moreover, participants that had 30% or more of outliers on either run were excluded from all analyses. Functional volumes were spatially smoothed with an 8-mm full width at half-maximum isotropic Gaussian kernel.

2.6. Brain Activity within English or Spanish Word Reading

The SPM12 toolbox (https://www.fil.ion.ucl.ac.uk/spm/) was used to construct and test the fit of the neuroimaging data to a general linear model (Friston et al., 1994). Statistical parametric maps were created corresponding with the time-courses for the following contrasts: English word reading greater than false fonts (English [RW > FF]); and, Spanish word reading greater than false fonts (Spanish [RW > FF]). Voxel-wise t-maps were constructed for each of the participants as a first-level analysis and the amplitude maps were then carried to a second-level analysis to test for significant group effects. Whole-brain random-effects analyses for each contrast were conducted by performing one-sample t-tests on the individual contrast images. Each of the contrasts are reported at a cluster-size threshold of p < 0.05 with false discovery rate (p-FDR) and a height threshold of p < 0.005, comparable to the statistical approach employed by the most recent study comparing local activation during English and Spanish word reading in bilingual adults and children (Hernandez, Woods, & Bradley, 2015), and significantly more stringent than previous studies from our laboratory using this IR Task (Turkeltaub et al., 2003; Jamal et al., 2012). Group-level activation maps were surface-rendered on the standardized MNI brain template in SPM12.

2.7. Brain Activity Common to English and Spanish Word Reading

To test whether there is brain activity shared while reading in each of these two languages, similar to an analysis reported in Hernandez, Woods, & Bradley (2015), we performed a standard conjunction analysis using the approach described by Nichols et al. (2005). We carried out a one-way analysis of variance at the group level with the English word reading map (English [RW > FF]) of each participant as one cell and the Spanish word reading map (Spanish [RW > FF]) of each participant as another cell. Then, both group-activation maps (English [RW > FF], and Spanish [RW > FF]) were selected (i.e., conjunction) to look at the local activation common to both languages. The statistical parametric maps are reported at a cluster-size threshold of p-FDR < 0.05 and a height threshold of p < 0.005.

2.8. Brain Activity Differences between English and Spanish Word Reading

To address our first prediction, we compared mean brain activity during English word reading greater than Spanish word reading, and vice versa. Using the SPM12 toolbox (https://www.fil.ion.ucl.ac.uk/spm/), the activation maps from the within-language analysis (English [RW > FF], and Spanish [RW > FF]) were carried to a second-level analysis to test for between-language differences. We performed a paired t-test with English word reading (English [RW > FF]) and Spanish word reading (Spanish [RW > FF]) activation maps, resulting in the following between-language comparisons: English [RW > FF] > Spanish [RW > FF]; and, Spanish [RW > FF] > English [RW > FF]. Group-level activation for each of these two contrasts was assessed under a cluster-size threshold of p-FDR < 0.05 and a height threshold of p < 0.005.

2.9. Establishing Seed Regions of Interest (ROIs) for Functional Connectivity Analysis

In order to test whether reading words in languages with different orthographic depth influences the functional connectivity between brain regions associated with reading in bilinguals, we selected eight seed ROIs on the basis of two meta-analyses of reading in alphabetic written languages (Martin et al., 2015; Bolger, Perfetti, & Schneider, 2005). First, using the coordinates reported by the most recent meta-analysis (Martin et al., 2015), we identified seven brain regions in the left hemisphere that are part of the OTC, TPC, and IFC: inferior temporal gyrus (L-ITG) (x = −48, y = −62, z = −20), inferior occipital gyrus (L-IOG) (x = −44, y = −74, z = −4), and middle occipital gyrus (L-MOG) (x = −42, y = −86, z = −2) in the OTC; intraparietal sulcus (L-IPS) (x = −42, y = −48, z = 48) in the TPC; and, inferior frontal gyrus, pars opercularis (L-IFG,oper) (x = −52, y = 18, z = 14), inferior frontal gyrus, pars triangularis (L-IFG,tri) (x = −52, y = 20, z = 18), and middle frontal gyrus (L-MFG) (x = −42, y = 4, z = 48) in the IFC. Second, we added an eighth ROI located in the posterior portion of the left superior temporal gyrus (L-STG) (MNI transformed: x = −54, y = −29, z = 10) in the TPC, reported in Bolger, Perfetti, & Schneider (2005) meta-analysis across 25 studies of reading in alphabetic languages (e.g., English, French, Italian, German), given the evidence that the left posterior superior temporal gyrus is involved in GPC (for a review, van Atteveldt & Ansari, 2014). Note that Martin et al. (2015) meta-analysis included 17 studies in deep orthographies and 3 in shallow orthographies, but did not report significant convergence of activation across these studies in the left superior temporal gyrus, potentially biased by the unbalanced deep-to-shallow ratio. For each ROI, we created a spherical seed with a 6-mm radius using the MarsBaR toolbox 0.44 (http://marsbar.sourceforge.net/). A brain map overlaying the eight functional seed ROIs with anatomy is found in Figure 1, all surface-rendered in MNI space.

Figure 1. Seed ROIs for the functional connectivity analysis.

Eight left-hemisphere seed ROIs representing brain regions involved in word reading in adults, located in the OTC (L-ITG, LIOG, and L-MOG), the TPC (L-STG and L-IPS), and the IFC (L-IFG,oper, L-IFG,tri, and LMFG).

2.10. Brain Functional Connectivity Differences between English and Spanish Word Reading

To test whether the functional network engaged during word reading is different for a deep versus a shallow alphabetic orthography, we performed a seed-to-voxel analysis. This method measures the temporal correlation between the seed ROIs and all the other voxels (cortex and cerebellum) after accounting for the common driving influence of task-derived activity on both a given ROI and a target voxel (Friston et al., 1997). This approach was chosen over an ROI-to-ROI analysis to ensure no regions were overlooked, since regions other than those eight defined a priori could also support word reading, depending upon key modulatory factors, such as differences in GPC consistency between two languages. Using the CONN toolbox v.18.a, task-based fMRI data were not band-pass filtered (0.008-Inf), so as to capture task-derived signals. For all data, we modeled nuisance covariates that include white matter and cerebrospinal fluid signals and their derivatives (CompCor in CONN: Behzadi et al., 2007).

We performed a generalized form of seed-to-voxel psycho-physiological interaction (gPPI) analysis for its sensitivity and specificity in detecting functional connectivity effects, compared to the standard PPI implementation in SPM, and its suitability to block designs (Cisler, Bush, & Steele, 2014; McLaren et al., 2012). The gPPI analysis allows the estimation of task-derived functional connectivity for more than one task condition, modelling all condition effects and interactions simultaneously in the same model instead of using a different model for every condition. For our gPPI analysis, we imported the eight ROIs into the CONN toolbox as seeds to test for changes in regional inter-hemispheric functional connectivity between task conditions (i.e., English versus Spanish word reading). At the first-level (individual) analysis, we included a psychological variable representing the two types of conditions (i.e., English RW and Spanish RW), one physiological variable (i.e., the time-course in each of the seed ROIs), and a PPI term (i.e., the interaction between these two regressors). Statistical significance was determined under a cluster-size threshold of p-FDR < 0.05 and a height threshold of p < 0.005; after a Bonferroni correction for multiple comparisons (i.e., 0.05 p-FDR/8 seed ROIs), the threshold was p < 0.006. The results that survived this more stringent threshold are marked with an asterisk (*) in Table 3. Group-level, seed-to-voxel statistical maps of correlation were surface-rendered on the standardized MNI brain template in CONN.

Table 3.

Results of Brain Functional Connectivity within and between Languages.

			MNI Coordinates
Language	Seed ROIs	Cluster location	x	y	z	Peak Z	Voxels
English
	L-IFG,tri	(−) L cerebellar lobule VI extending to crus I	−40	−48	−40	−5.17	440
Spanish
	L-MOG	(−) R thalamus extending to L thalamus*	10	−14	6	−5.34	696
		(−) L supplementary motor area extending to R supplementary motor area	−2	−2	52	−4.65	396
	L-IPS	(−) L postcentral gyrus extending to superior parietal lobule	−30	−26	48	−5.30	401
	L-IFG,oper	(−) L posterior cingulate cortex extending to anterior cingulate cortex*	−2	−28	44	−6.21	831
English > Spanish
	L-MOG	(+) R lingual gyrus extending to L lingual gyrus and R cerebellar lobule VI-V	12	−50	0	4.06	391
	L-IPS	(+) L superior parietal lobule	−30	−34	56	4.41	374
Spanish > English
	L-IFG,tri	(+) L cerebellar lobule VI*	−34	−48	−30	4.96	610
	L-STG	(+) L posterior cingulate cortex extending to precuneus	−2	−58	36	5.07	510

Abbreviations: L = left hemisphere; R = right hemisphere; (+) = positive FC; (−) = negative FC;

^*= survived Bonferroni correction p < 0.006.

2.11. Spatially Distributed Brain Activity Patterns Differentiating between English and Spanish Word Reading

To test for potential differentiation of English and Spanish word reading at the level of spatially distributed activation patterns, an MVPA was executed using The Decoding Toolbox (Hebart, Görgen, & Haynes, 2015). We performed a linear support vector machine classifier (Cortes & Vapnik, 1995) using the leave-one-run-out cross-validation procedure (Mahmoudi et al., 2012). Taking the unsmoothed functional data, the first run of the English and the Spanish IR Tasks were concatenated, and the same procedure was implemented for the second run of each task (English IR2 and Spanish IR2). The classification analysis was performed on the RW blocks of each language; therefore, concatenated runs 1 of the IR Task (containing both English and Spanish RW blocks) were used to train the classifier, and the concatenated runs 2 containing RW blocks of both languages were used to test it. The same procedure was repeated, but the opposite way (i.e., train in IR2 and test in IR1), and ultimately the average of these two analyses was obtained. A 4-mm radius spherical “searchlight” moved across the entire brain, taking each voxel in the volume as the searchlight center. Maps of statistically significant voxel-wise accuracies against a chance-level accuracy of 50% are shown under a cluster-size threshold of p-FDR < 0.05 and a height threshold of p < 0.005.

3. Results

3.1. In-scanner Task Performance and Post-scanning Stimuli Recognition Test

When looking at in-scanner performance accuracy, the difference of the difference (English [RW-FF] vs. Spanish [RW-FF]), which parallels the fMRI analyses (English [RW>FF] > Spanish [RW>FF], and Spanish [RW>FF] > English [RW>FF]), there was no statistically significant difference (p > 0.05), as shown in Table 1. For response time, there was a small but significant difference when comparing the difference of the difference. This result was due to participants taking longer to respond to false fonts than to real words in the English task. However, when comparing average response times for English versus Spanish real word reading, they did not differ.

The pencil-and-paper recognition test results showed that participants implicitly processed the word stimuli shown inside the scanner. Real words discrimination was significantly more accurate than false-font strings in each language (p < 0.01): 60% accuracy for English RW compared to 50% accuracy for English FF (n = 23; 2 participants’ data were not available for analysis); and, 67% accuracy for Spanish RW compared to 59% accuracy for Spanish FF (n = 22; 3 participants’ data were not available for analysis), similarly to previous findings from our laboratory (Turkeltaub et al., 2003).

3.2. Brain Activity within English or Spanish Word Reading

English:

English word reading invoked activity in one cluster on the left inferior frontal gyrus, pars triangularis extending to pars opercularis (Figure 2; Table 2).

Figure 2. Brain activity within languages.

English word reading (deep orthography) relative to false fonts engaged the left inferior frontal gyrus, pars opercularis extending to pars triangularis and pars orbitalis (top). Spanish word reading (shallow orthography) relative to false fonts activated the left inferior frontal gyrus, pars orbitalis extending to pars triangularis and the left middle temporal gyrus (bottom). Cluster-size p-FDR < 0.05, height threshold p < 0.005. L = left hemisphere, R = right hemisphere.

Table 2.

Results of Mean Brain Activity within, across, and between Languages.

		MNI Coordinates
Language	Cluster location	x	y	z	Peak Z	Voxels
English
	L inferior frontal gyrus, triangularis extending to opercularis	−38	−6	28	3.89	1,131
Spanish
	L inferior frontal gyrus, orbitalis extending to triangularis	−44	26	−6	5.05	3,340
	L middle temporal gyrus	−62	−34	2	4.72	1,554
English + Spanish
	L inferior frontal gyrus, orbitalis extending to triangularis and opercularis	−44	30	−2	3.73	833
English > Spanish
	n.s.
Spanish > English
	n.s.

Abbreviations: L = left hemisphere; n.s. = non-significant (p-FDR = or > 0.05).

Spanish:

Spanish word reading revealed brain activity in two left-hemisphere clusters: the inferior frontal gyrus, pars orbitalis extending to pars triangularis; and, the middle temporal gyrus (Figure 2; Table 2).

3.3. Brain Activity Common to English and Spanish Word Reading

English U Spanish:

Significant common activity for English and Spanish word reading was found in the left inferior frontal gyrus, pars orbitalis extending to pars triangularis and pars opercularis (Figure 3; Table 2).

Figure 3. Brain activity common to both languages.

English (deep orthography) and Spanish (shallow orthography) word reading relative to false fonts engaged a common area in the left inferior frontal gyrus, pars orbitalis extending to pars triangularis and pars opercularis. Cluster-size p-FDR < 0.05, height threshold p < 0.005. L = left hemisphere, R = right hemisphere.

3.4. Brain Activity Differences between English and Spanish Word Reading

English > Spanish:

The comparison of English greater than Spanish word reading revealed no significant difference (Table 2).

Spanish > English:

Spanish greater than English word reading also yielded no significant results (Table 2).

3.5. Brain Functional Connectivity within English or Spanish Word Reading

English:

There was a negative correlation between the seed in the left inferior frontal gyrus, pars triangularis (L-IFG,tri) and a cluster in the left cerebellar lobule VI extending to crus I (Figure 4; Table 3).

Figure 4. Brain functional connectivity within languages.

English word reading showed negative connectivity between seed L-IFG,tri (IFC) and left cerebellar lobule VI extending to crus I (top). Spanish word reading engaged negative correlation between seed L-MOG (OTC) and right thalamus extending to left thalamus, as well as left supplementary motor area extending to right supplementary motor area, seed L-IPS (TPC) and left postcentral gyrus extending to left superior parietal lobule, and seed L-IFG,oper (IFC) and left posterior cingulate cortex extending to left anterior cingulate cortex (bottom). Cluster-size p-FDR < 0.05, height threshold p < 0.005. L = left hemisphere.

Spanish:

Several negative correlations were found: between the seed in the left middle occipital gyrus (L-MOG) and a cluster in the right thalamus extending to the left thalamus, as well as a cluster in the left supplementary motor area extending to the right supplementary motor area; between the seed in the left intraparietal sulcus (L-IPS) and a cluster in the left postcentral gyrus extending to the left superior parietal lobule; and, between the seed in the left inferior frontal gyrus, pars opercularis (L-IFG,oper) and a cluster in the left posterior cingulate cortex extending to the left anterior cingulate cortex (Figure 4; Table 3).

3.6. Brain Functional Connectivity Differences between English and Spanish Word Reading

English > Spanish:

Relative to Spanish, English word reading had stronger positive correlations between: the seed in the left middle occipital gyrus (L-MOG) and a cluster in the right lingual gyrus extending to the left lingual gyrus and right cerebellar lobule VI-V; and, the seed in the left intraparietal sulcus (L-IPS) and a cluster in the left superior parietal lobule (Figure 5; Table 3).

Figure 5. Brain functional connectivity between languages.

Stronger positive functional connectivity between seed L-MOG (OTC) and right lingual gyrus extending to left lingual gyrus and right cerebellar lobule VI-V and seed L-IPS (TPC) and left superior parietal lobule during English word reading compared to Spanish (top), and between seed L-IFG,tri (IFC) and left cerebellar lobule VI and seed L-STG (TPC) and left posterior cingulate cortex extending to precuneus during Spanish word reading compared to English (bottom). Cluster-size p-FDR < 0.05, height threshold p < 0.005. L = left hemisphere.

Spanish > English:

During Spanish word reading compared to English, there was relatively stronger positive correlation between the seed in the left inferior frontal gyrus, pars triangularis (L-IFG,tri) and a cluster in the left cerebellar lobule VI; and, between the seed in the left superior temporal gyrus (L-STG) and a cluster in the left posterior cingulate cortex extending to the left precuneus (Figure 5; Table 3).

3.7. Spatially Distributed Brain Activity Patterns Differentiating between English and Spanish Word Reading

English vs. Spanish:

There were spatially distributed patterns of brain activity that distinguish between English and Spanish word reading in four clusters: beginning with the largest, left cerebellar lobule VI extending to left fusiform gyrus, right cerebellar lobule VI, and right middle occipital gyrus; left inferior frontal gyrus, pars orbitalis; left medial frontal gyrus; and, left superior occipital gyrus (Figure 6; Table 4).

Figure 6. Spatially distributed brain activity patterns between languages.

Multivariate pattern analysis revealed differentiation between English and Spanish word reading in four clusters: left cerebellar lobule VI extending to left fusiform gyrus, right middle occipital gyrus, and right cerebellar lobule VI; left inferior frontal gyrus, pars orbitalis; left medial frontal gyrus; and, left superior occipital gyrus. Cluster-size p-FDR < 0.05, height threshold p < 0.005. L = left hemisphere, R = right hemisphere.

Table 4.

Results of Spatially Distributed Brain Activity between Languages.

		MNI Coordinates
Language	Cluster location	x	y	z	Peak Z	Voxels
English vs. Spanish
	L cerebellar lobule VI extending to L fusiform gyrus, R cerebellar lobule VI, R middle occipital gyrus	−18	−54	−16	4.62	11,501
	L inferior frontal gyrus, orbitalis	−36	26	−6	3.95	1,958
	L medial frontal gyrus	−8	36	56	4.06	1,300
	L superior occipital gyrus	−14	−92	8	3.80	1,294

Abbreviations: L = left hemisphere; R = right hemisphere.

4. Discussion

The goal of this study was to test whether reading words in alphabetic languages with different orthographic depth has an effect on the functional brain bases of reading in Spanish-English early bilinguals. We examined and compared brain activity during single-word reading in a deep (English) and a shallow (Spanish) orthography, similar to prior neuroimaging studies (Meschyan & Hernandez, 2006; Jamal et al., 2012; Hernandez, Woods, & Bradley, 2015). A novel aspect of our study was the investigation into functional connectivity associated with processing words in a deep versus a shallow orthography, and the application of MVPA to examine spatially distributed patterns of brain activity that distinguish these two languages.

During English word reading, we observed engagement of the left inferior frontal gyrus, and during Spanish word reading, the left inferior frontal and middle temporal gyri. A conjunction analysis across these two languages with varying GPC consistencies revealed shared activity in the left inferior frontal gyrus. Between-language comparisons, however, revealed no difference in local activity. At the network level, we found relatively stronger positive functional connectivity between the left middle occipital gyrus and bilateral lingual gyri, and between the left intraparietal sulcus and the left superior parietal lobule, for the deep orthography. On the other hand, the shallow orthography had relatively stronger positive functional connectivity between the left inferior frontal gyrus and the left cerebellar lobule VI, and between the left superior temporal gyrus and the left posterior cingulate cortex. When examining spatially distributed activity patterns, we found clusters that distinguish between English and Spanish word reading in left cerebellum (extending into left fusiform, right cerebellum, and right middle occipital gyrus), as well as left superior occipital, inferior frontal, and medial frontal gyri.

4.1. Brain Activity underlying Word Reading in English and Spanish

Activity in the left inferior frontal gyrus, observed for each language separately and also when testing for overlap between them, is consistent with our own previous study (Jamal et al., 2012) that shared 10 participants with the present study. It is also consistent with the most recent study of English and Spanish word reading in Spanish-English bilinguals (Hernandez, Woods, & Bradley, 2015). While this latter study was primarily focused on bilingual adults versus children (reading in English and Spanish), when examining brain activity for each language in adults, the authors found engagement of the left inferior frontal gyrus (with extension into putamen and thalamus) for both English and Spanish word reading. However, the earlier study in Spanish-English bilinguals from the same research team did not report activity in the left inferior frontal gyrus for either language, despite similarities in the study design (Meschyan & Hernandez, 2006). The inferior frontal gyrus has been shown to be involved in semantic processing, phonological decoding, orthographic processing, and articulatory recoding of print (Bookheimer, 2002; Cutting et al., 2006; Fiez, 1997; Poldrack et al., 1999; Price, 2012; Pugh et al., 2001). Of note, our task is not designed to tease apart orthography, phonology, and semantics; rather it is thought to broadly engage all three processes. While it is not surprising that we found brain activity during both English and Spanish word reading in the left inferior frontal gyrus, we cannot make any specific attributions to orthographic, phonological, or sematic processing. In previous studies from our laboratory using this task, we reported activation of the left inferior frontal gyrus in typically reading adults and children (Evans et al., 2016; Turkeltaub et al., 2003). In one of these studies, activation of the pars orbitalis of the left inferior frontal gyrus, identified here as shared by English and Spanish, was also positively correlated with the ability to manipulate phonemes measured outside of the scanner (Turkeltaub et al., 2003).

Other than the left inferior frontal gyrus, there were no further findings during English word reading. For Spanish word reading, there was local activity in the left middle temporal gyrus, as reported in our earlier study by Jamal et al. (2012). Specifically, in this earlier study, the left middle temporal gyrus was identified during Spanish, but not English, word processing, as in the present study. In Hernandez, Woods, & Bradley (2015), however, local activation in the left middle temporal gyrus was observed for both English and Spanish word reading in bilingual adults, although it was located significantly more posterior and less lateral than the region reported for Spanish in the present study. None of these regions were reported in the study by Meschyan and Hernandez (2006), although they found activity in the left superior temporal gyrus during Spanish word reading compared to English.

Despite any apparent differences in the within-language maps, when directly comparing local activity between English and Spanish word reading, there were no differences. While this result was unexpected and inconsistent with two previous studies (Meschyan & Hernandez, 2006; Jamal et al., 2012), it aligns with the recent report by Hernandez, Woods, & Bradley (2015) that also did not show differences. Specifically, despite having a larger cohort (N = 20) than either of the two earlier studies (N = 12 for each study), Hernandez, Woods, & Bradley (2015) revealed no differences in activation between English and Spanish during a single-word reading task in early bilingual adults. Similarly, the present study, with 25 participants, also revealed no differences in activation between the two languages. Thus, these results lead us to conclude that reading words in alphabetic languages with different orthographic depth has no modulatory role on local brain fMRI signal intensity within bilinguals.

While all of these studies were conducted in Spanish-English early bilinguals, they were motivated by a landmark study on the role of orthographic depth conducted in monolingual speakers of English compared to monolingual speakers of Italian (Paulesu et al., 2000). Using a between-group comparisons, this study revealed that the deep orthography (English) made relatively greater use of regions in the OTC and the IFC, while the shallow orthography (Italian) engaged regions of the left TPC to a greater degree. One caveat with this study, however, is that the findings associated with the deep orthography were observed only during pseudoword reading and not during real word reading, while the findings associated with the shallow orthography were observed for both, pseudoword and real word reading. The authors concluded that the deep orthography relies more on semantic and orthographic processing. Yet, the lack of differences observed for real word processing weakens this claim, especially since pseudoword processing (for which there was a difference) is void of semantics and involves phonological processing. Future studies will need to combine the study of two monolingual groups using languages of different orthographic depth with the study of a group that is bilingual in these same two languages to better determine the role of the orthographies’ transparencies. Difference between two monolingual groups, like those revealed by Paulesu et al., (2000), may be due to factors other than those related to orthographic depth. Regardless, taken at face value, the findings in monolingual speakers of English and Italian do not extend to Spanish-English early bilingual adults reading real words, based on the current study and that by Hernandez, Woods, & Bradley (2015). Further, both of these studies indicate that the left inferior frontal gyrus is equally activated during English and Spanish word reading.

Some have argued against the notion of significant differences in brain function across languages, proposing a “universal signature” for reading (and language in general), as long as both languages were acquired simultaneously or early in life and are equally balanced (Rueckl et al., 2015; Cao et al., 2014; Pugh, 2006). For example, Rueckl et al. (2015) tested whether spoken (word listening) and written (word reading) languages activate the same brain regions across alphabetic and logographic languages. When they examined word reading across four languages, they found that the three alphabetic languages (English, Spanish, Hebrew) and the logographic language (Chinese) did not differ in local activity in brain regions known to be involved in reading (although there were small differences in bilateral postcentral gyrus and cingulate gyrus). These findings fit with the notion that there are more similarities than differences for word reading in languages with differing orthographic depth (Ziegler et al., 2010).

In the context of dual-language learners, Wong, Yin, & O’Brien (2016) argued that this notion of a universal brain system for spoken and written language holds, given that learning two languages will likely engage the same regions used in single language learning, an idea supported by numerous reports (Cummine & Boliek, 2013; Grogan et al., 2009; Leonard et al., 2010; Pugh, 2006). They also propose that a universal brain system supporting the two languages of bilinguals interacts with regions not essential for language processing, but for other higher-level cognitive domains, such as language switching, as shown by Abutalebi et al. (2012) and Hervais-Adelman, Moser-Mercer, & Golestani (2011). Therefore, differential properties of alphabetic languages, such as GPC consistency, may not manifest at the level of local mean activation in bilinguals. Nonetheless, this possibility does not implicate that identical computations are used (Rueckl et al., 2015), a process that can be clarified by examining signal intensity at the level of voxel-wise activation patterns, as will be discussed later.

4.2. Brain Functional Connectivity underlying Word Reading in English and Spanish

We found functional connectivity for some of the seed regions during English and also during Spanish word reading. Similar to studies examining brain function at the network level, both during reading tasks (Bitan et al., 2006; Hampson et al., 2006; Horwitz, Rumsey, & Donohue, 1998; Mechelli et al., 2005; Pugh & Mencl, 2000) as well as at rest (Koyama et al., 2011, 2010), the negative connections found in the present study for either language fell outside of those brain regions typically associated with reading. Specifically, English word reading showed decoupling between the left inferior frontal gyrus, pars triangularis seed region and the left cerebellum. Reading words in Spanish elicited decoupling between the left middle occipital gyrus seed region and cortical and subcortical regions associated with motor control (right thalamus and left supplementary motor area, each extending into their contralateral side); between the left intraparietal sulcus seed region and cortical regions engaged in sensory processing, information integration, and attention (left postcentral gyrus extending to superior parietal lobule); and, between the left inferior frontal gyrus, pars opercularis seed region and regions associated with executive function (left posterior cingulate extending to anterior cingulate cortex).

When directly comparing functional connectivity during English versus Spanish word reading, we found several differences (all positives) between the two languages. However, the regions that connected with some of our seed regions were not as expected. English word reading compared to Spanish revealed stronger positive connectivity between the left middle occipital gyrus seed region and an area spanning right lingual gyrus, left lingual gyrus, and right cerebellar lobule VI-V; and, between the left intraparietal sulcus seed region and the left superior parietal lobule. The lingual gyrus has been proposed to be involved in visual-word processing, mainly when participants are engaged in a stimulus-specific response (for a review, Mechelli et al., 2000). In the context of our task, participants may not only have stronger coupling between nearby visual regions during English word reading, given the potential for greater reliance on orthographic processing (Paulesu et al., 2000), but also due to the feature-detection demand. The left dorsal inferior parietal lobule/intraparietal sulcus has been implicated in serial decoding (Richlan, 2014), given its engagement in letter-to-letter shifting attention within a string (Behrmann, Geng, & Shomstein, 2004; Cabeza, Ciaramelli, & Moscovitch, 2012; Rosazza et al., 2009; Wager, Jonides, & Reading, 2004). Moreover, the superior parietal lobule, which has been associated with multiple cognitive processes, is important for visuospatial attention (for a review, Wang et al., 2015). However, further studies are needed to better understand these two sets of short range functional connections within the OTC and the TPC during English compared to Spanish word reading, especially since these connections were not found in the functional connectivity map for English word reading.

For the reverse comparison, we observed relatively stronger coupling between the left inferior frontal gyrus, pars triangularis seed region and the left cerebellar lobule VI during Spanish word reading compared to English. Evidence of anatomical connectivity in humans (McTavish & Franz, 2015) shows interconnections between inferior frontal regions and the cerebellum, although for the contralateral side (for a review, Buckner, 2013). Nevertheless, given that the left inferior frontal gyrus, pars triangularis is associated with lexico-semantic processing, it is noteworthy the cerebellum has been implicated in similar aspects of language processing, including semantic judgement and explicit memory retrieval (Fulbright et al., 1999; Desmond & Fiez, 1998; Ben-Yehudah & Fiez, 2008; Stoodley & Schmahmann, 2009), but these findings are usually in the right cerebellum. In the current analysis, since English word reading resulted in negative connectivity between the left inferior frontal gyrus, pars triangularis and the left cerebellar lobule VI, while for Spanish there was no such connection, this between-language difference is driven by the anticorrelation between these two regions for English. Lastly, Spanish compared to English word reading also engendered relatively stronger coupling between the left posterior superior temporal gyrus and the posterior cingulate cortex. This functional connection did not emerge in either group alone. Despite the evidence associating the posterior cingulate cortex with other regions in the so-called default mode network, a system mostly associated with “mind-wondering” (unconstrained rest), its specific function is still unclear (for a review, Leech & Sharp, 2014).

Together, the findings from this functional connectivity analysis were unexpected and provided no evidence to support the notion that English and Spanish differ in their functional connections between regions associated with reading due to their linguistic differences at the word processing level.

4.3. Distinguishing between English and Spanish Word Reading with Spatially Distributed Patterns of Brain Activity

In contrast to the univariate analysis of local activity described above, our MVPA revealed spatially distributed patterns of activation in regions of bilateral cerebellum/left OTC and left IFC, as well as in the left superior occipital and medial frontal gyri during English versus Spanish word reading. This finding suggests that there are clusters of voxels that specialize in distinguishing between a deep and a shallow alphabetic orthography, even within brain areas known to be involved in word reading, namely left OTC and IFC. Most likely each of the two languages recruits similar, but functionally independent neural populations within the same region(s), a mechanism that would not be revealed by mean activation maps.

Similar use of MVPA has been made in a study of Dutch-English bilinguals, where the univariate analysis failed to show between-condition effects on semantic representation of spoken language, but were revealed by the MVPA (Correia et al., 2014). Moreover, a recent study tested whether implicit word processing in Chinese versus English exhibits distinct neural representations in Chinese-English late bilingual adults using MVPA (Xu et al., 2017). The authors found spatially distributed patterns of activity in reading-related regions distinguishing between Chinese and English. Specifically, these patterns were observed in the left temporo-parietal cortex, as well as in the bilateral fusiform, occipital cortex, superior and middle temporal gyri, superior parietal lobule, precuneus, and inferior and middle frontal gyri. Of interest, as in the current study, this investigation did not identify differences in these regions when comparing activation between the two languages.

Our MVPA findings are consistent with this prior work and further support the existence of regions that code for orthography-specific information at a fine-grained level in ways that differ between the two alphabetic languages studied. The fact that we observed patterns of activation that distinguish between a deep and a shallow orthography in the left fusiform gyrus and the left inferior frontal gyrus, pars orbitalis, brain regions associated with the lexico-semantic route, raises the possibility that information about linguistic differences between English and Spanish word reading, such as the letter-to-sound correspondence consistency, is differentially represented in the bilingual brain without needing to recruit different regions for each language. Moreover, some of these representations were located in regions outside of the typical left-hemisphere cortical areas known to be involved in reading, specifically the left superior occipital gyrus (visual association cortex), the left medial frontal gyrus, and the expansions into right cerebellum and right middle occipital gyrus from the left cerebellar lobule VI. MVPA tends to be more sensitive to revealing processes outside of those required for the main task (Haynes & Rees, 2006; Norman et al., 2006; Pereira, Mitchell, & Botvinick, 2009). Therefore, these specific regions may represent spatially distributed patterns of brain activity distinguishing between English and Spanish word reading that are associated with complex pattern visual-stimuli processing (visual association cortex) and executive function (medial frontal gyrus).

4.4. Absence of Between-Language Differences in Brain Activity

Had we found differences in local activation between English and Spanish, we would have attributed them to differences primarily in orthographic depth of these two languages. The question arises, however, whether bilinguals truly simulate monolinguals of English and monolinguals of Spanish combined into the same brain. The bilingual experience may modulate the reading brain in several ways. First, there may be a strong interdependence between the two languages processed by bilinguals, resulting in literacy skills that transfer from one language (e.g., shallow) to the other (e.g., deep), and/or vice versa. For example, the Spanish-English bilinguals of our study may have read English words as if they were Spanish words. Such a strategy would explain the lack of differences in the brain’s local activity for English and Spanish word reading. Related to this possibility, there is evidence showing that Spanish-English (Kovelman, Baker, & Petitto, 2008) and Italian-English (Kremin et al., 2019; D’angiulli, Siegel, & Serra, 2001) bilinguals rely more heavily on phonological processing for reading and have overall better phonological abilities compared to English monolinguals. This observation suggests a “transfer effect”, where phonological skills developed through a shallow orthography can facilitate learning to read in a deep orthography (Proctor et al., 2010). There are also several studies that illustrate an effect of bilingualism on brain regions involved in language (Jasińska & Petitto, 2014; Parker Jones et al., 2012; Rodriguez-Fornells et al., 2002), as well as non-language processes (Costumero et al., 2015; Grady et al., 2015; Kroll & Bialystok, 2013). Most relevant in this regard is a study reporting differences in local activity during Spanish word processing in Spanish-Catalan bilinguals compared to Spanish monolinguals (Rodriguez-Fornells et al., 2002). These findings suggest that the bilingual experience, relative to the monolingual experience, modulates the reading of Spanish words. As such, it is possible that when comparing bilinguals during the reading of Spanish and English words, there will be differences attributed not to the differences in languages (e.g., orthographic depth), but due to being bilingual and the associated changes described above. However, since we did not find differences in local activity between the two languages, the participants’ bilingual experience is not a major concern for the interpretation of these null results. We acknowledge, however, there remains a small chance that their bilingual experience modulates local activation of brain regions involved in reading in one (or both) of the languages differentially. This potential concealment of a language-specific effect by the effect of bilingualism can only be ruled out by comparing bilinguals to monolinguals under the same task conditions, similar to Rodriguez-Fornells et al. (2002). In a separate study, we did this comparison with English monolinguals (Brignoni-Perez et al., unpublished results) and the findings do not alter the conclusions we draw in the current study.

Another aspect to consider is the similarity in linguistic characteristics for English and Spanish. One of the main shared characteristics is Romance (French and Latin-based) etymology. Therefore, one might be concerned that the lack of a between-language difference in local brain activity is due to the similarities between English and Spanish words. However, the words used in the study were not restricted to English and Spanish words with a common etymology. A post-hoc review revealed that only about a third of the English words in our study share the same root with their Spanish counterpart (e.g., “alarm” in English/“alarma” in Spanish). On the other hand, had we found differences in activity between English and Spanish, the question would arise whether other distinguishing linguistic properties of the word stimuli for each language could be confounding variables. For example, graphotactic patterns in words differ not only across languages, but also within languages, and monolingual readers of English have been shown to be sensitive to the graphotactic patterns that reflect a word’s origin (Treiman, Decker, & Kessler, 2019). Along the same lines, a recent study of a large lexical database in conjunction with information on age of acquisition norms showed that the etymology of English words (Germanic versus Latin-based) influences the age of acquisition, as well as reaction time and accuracy for that word, even when controlling for word frequency and length (Hernandez et al., 2019). These findings suggest that, in monolinguals, brain activity may differ while reading English words of different etymological origins. Such linguistic characteristics could also play a role in brain imaging studies of word processing when comparing two languages in bilinguals, especially if the linguistic characteristics of the word stimuli vary along the dimension of one of the two languages. Interestingly, despite the vast difference in the etymology of the two languages spoken by Chinese-English bilinguals, Xu et al. (2017) found no difference in local activation for word reading in Chinese and English.

In the face of a null result for local activity in the between-language comparison, it behooves us to also discuss the use of an implicit rather than an explicit reading task. The implicit reading task, first implemented by Price, Wise, & Frackowiak (1996), is thought to elicit similar brain activity as that observed during aloud word reading (Brunswick et al., 1999; Paulesu et al., 2000), but this may not be the case (Vogel, Petersen, & Schlaggar, 2013). Therefore, future studies need to be conducted with an overt reading task. As noted in the Introduction, however, the task used in the current study is similar to those used in prior studies on between-language differences in Spanish-English bilinguals (Meschyan & Hernandez, 2006; Jamal et al., 2012; Hernandez, Woods, & Bradley, 2015), as well as in Chinese-English bilinguals (Xu et al., 2017), thereby allowing us to draw comparisons to these studies in ways that would not be possible if we had used an overt reading task. Nevertheless, the choice of task can modify the results significantly, as demonstrated by Wang et al. (2015) who tested for language differences (English and Chinese) using two different tasks, a lexical decision task and a naturalistic story reading task. They found that during the lexical decision task there were more differences in brain activity between Chinese and English monolinguals than during the story reading task, suggesting that the choice of task is critical to the outcome.

Finally, the present study used a statistical approach that is similar to the most recent report by Hernandez, Woods, & Bradley (2015) and more stringent than prior studies (Meschyan & Hernandez, 2006; Jamal et al., 2012). These threshold differences are the most likely reason for the inconsistencies found between the two older studies and the two most recent studies (including the present one), and the convergence of results for Hernandez, Woods, & Bradley (2015) and the present study (i.e., no differences in local activity between the two languages). Some investigators advocate for the use of an even more stringent approach and thus might consider the one used here to be a limitation (i.e., cluster-size p-FDR < 0.05, height p < 0.005). If a less lenient height threshold was applied here (e.g., height p < 0.001 instead of height p < 0.005), however, it would not change the results when comparing brain activation between the two languages.

In the end, our observation of no difference in local activity is consistent with Hernandez, Woods, & Bradley (2015). Further, the report by Xu et al. (2017) also showed no difference in local activation during word processing between languages in Chinese-English late bilinguals. They found, however, significant spatially distributed patterns of activity distinguishing between the two languages, as we did in our Spanish-English early bilinguals.

5. Conclusions

We investigated brain function during word reading in a deep versus a shallow orthography within Spanish-English early bilingual adults. Our study revealed that single-word reading in both the deep (English) and the shallow (Spanish) orthography engages the left inferior frontal cortex. We found no differences when comparing English and Spanish for local activity, nor functional connectivity between regions used for reading. However, English and Spanish word reading were found to be differently represented in several regions of the brain, including some in the OTC and the IFC (part of the direct, lexico-semantic route, the latter also being associated with the indirect, sublexical route). Overall, these results suggest that word processing in alphabetic languages of different orthographic depth does not engage language-specific brain activity locally, or functional connectivity between regions supporting reading, but manifests as distinct neural representations in Spanish-English early bilingual adults.

Highlights:

• Reading in English and Spanish activates left inferior frontal cortex in bilinguals

• Local brain activity does not differ between English and Spanish word reading

• English and Spanish word reading rely on slightly different functional connectivity

• Distributed activity patterns distinguish reading words in English versus Spanish