The neurocognitive basis of morphological processing in typical and impaired readers

Abstract

Morphological awareness, or sensitivity to units of meaning, is an essential component of reading comprehension development. Current neurobiological models of reading and dyslexia have largely been built upon phonological processing models, yet reading for meaning is as essential as reading for sound. To fill this gap, the present study explores the relation between children’s neural organization for morphological awareness and successful reading comprehension in typically developing and impaired readers. English-speaking children ages 6–11 (N = 97; mean age = 8.6 years, 25% reading impaired) completed standard literacy assessments as well as an auditory morphological awareness task during functional near-infrared spectroscopy (fNIRS) neuroimaging, which included root (e.g., PERSON + al) and derivational (e.g., quick + LY) morphology. Regression analyses revealed that children’s morphological awareness predicted unique variance in reading comprehension above and beyond demographic factors, vocabulary knowledge, and decoding ability. Neuroimaging analyses further revealed that children with stronger reading comprehension showed greater engagement of brain regions associated with integrating sound and meaning, including left inferior frontal, middle temporal, and inferior parietal regions. This effect was especially notable for the derivational morphology condition that involved manipulating more analytically demanding and semantically abstract units (e.g., un-, -ly, -ion). Together, these findings suggest that successful reading comprehension, and its deficit in dyslexia, may be related to the ability to manipulate morpho-phonological units of word meaning and structure. These results inform theoretical perspectives on literacy and children’s neural architecture for learning to read.

The goal of reading is to comprehend meaning from text. However, for children with dyslexia, deficits in single-word reading may impede the ability to extract units of meaning, leading to impaired reading comprehension. Over the past several decades, we have gained substantial insight into the neurocognitive differences underlying dyslexia, most notably in phonological processing. However, much less is known about the role of morphological awareness, or children’s sensitivity to units of meaning, in readers with dyslexia. Accordingly, the present study investigates the role of morphological awareness both behaviorally and in the brain of young learners across a wide range of reading ability. We ask two main questions. First, how does morphological awareness contribute to reading comprehension in children with and without reading impairment? Second, how is the brain basis of morphological processing associated with skilled reading comprehension?

Morphological awareness may be understood along a developmental continuum from implicit knowledge of morphemes, and how meaningful units combine to make words, to an explicit understanding of morphology over the course of schooling (Carlisle, 2004). Prior to formal literacy instruction, children recognize morphemic regularities and can manipulate morphemes in speech (Berko, 1958). This tacit or implicit awareness deepens children’s word knowledge, connecting mental representations of sound, print, and meaning, and facilitating easier word recognition (Nagy et al., 2014). In more mature readers, tacit morphological awareness may act as the foundational knowledge that supports more explicit literacy strategies such as morphological analysis and decoding (Levesque et al., 2020). The present study examines children’s implicit morphological processing and its relation to reading comprehension skill.

In typical readers, morphological awareness plays a critical role in successful literacy. Theoretical perspectives (Levesque et al., 2020; Perfetti & Stafura, 2014) suggest that morphology makes both direct and indirect contributions to reading. First, morphology is integral to every word, such as in the words cat, snow + man, or creat + iv + ity. Morphological awareness thus contributes directly to single-word recognition by providing information about word segmentation, pronunciation, and meaning. Indeed, morphology may be seen as a “binding agent” that connects representations of sound, meaning, and print together, thereby strengthening mental representations of words themselves (Kirby & Bowers, 2017; Perfetti, 2007). Second, morphology is also a component of the general linguistic system and contributes directly to word and sentence-level language comprehension. In support of this framework, a growing body of evidence has revealed both direct and indirect associations between morphological awareness and reading comprehension (e.g., Deacon et al., 2014; Gilbert et al., 2014; James et al., 2021; Kieffer & Lesaux, 2012; Levesque et al., 2017; Nagy et al., 2006). Within the simple view of reading (SVR) framework, which posits that reading comprehension builds upon single-word decoding skill and broader language proficiency (Hoover & Gough, 1990), morphological awareness may contribute to both of these elements.

Of particular importance for reading comprehension in English is derivational morphology, which involves adding derived affixes to root morphemes to change the meaning or part of speech, as in beauti + ful + ly or in + decis + ion. As children progress through school, they encounter an increasing number of complex words in academic texts, over half of which are multi-morphemic derived words (Nagy & Anderson, 1984). In contrast to analyzing free root morphemes, derivational affixes are more semantically abstract and may be more analytically demanding for readers. Derivational morphological awareness has been shown to contribute directly to 6th graders’ reading comprehension, as well as indirectly through their vocabulary knowledge and word reading skill (Kieffer & Box, 2013). Furthermore, two separate studies found that children with poor reading comprehension skills underperform on derived items of a word analogy task (paint: painter:: bake: __) as compared to their word reading-matched peers with average reading comprehension skill (MacKay et al., 2017; Tong et al., 2011). Derivational morphological awareness is thus a promising area of inquiry, and may hold a valuable key to understanding literacy acquisition and reading comprehension difficulties. Yet despite its importance for successful reading, the role of morphological awareness in impaired reading, as well as the neurocognitive mechanisms that underlie morphological processing in typical reading and dyslexia, remains largely unexplored.

Morphological awareness in impaired readers

Little is known about the role of morphological awareness in children who struggle to read. Behavioral evidence suggests that children with dyslexia, a reading impairment associated with phonological deficits and word reading difficulty, consistently perform lower on morphological awareness tasks as compared to same-aged peers (Casalis et al., 2004; Kearns et al., 2016). For instance, retrospective analysis of 2nd graders with reading difficulties revealed that those children also had deficits in both phonological and morphological awareness in kindergarten, at the onset of learning to read (Law & Ghesquière, 2017). Similarly, preschoolers at family risk for developing dyslexia performed significantly lower on tasks of morphological awareness compared to an age-matched group of children with low risk for dyslexia (Law et al., 2017).

However, there are conflicting perspectives as to the etiology of these morphological difficulties. On one hand, these differences in morphological awareness may be a downstream consequence of impaired phonological awareness. In support of this perspective, Law et al. (2017) found that group differences in morphological awareness subside after controlling for phonological awareness. Similarly, Tsesmeli and Seymour (2006) found that adolescents with dyslexia underperformed on a morphological awareness task compared to age-matched controls, but not reading-matched controls. This evidence suggested that difficulties with morphological segmentation may simply be a cascading result of poor reading performance rather than a unique deficit (Tsesmeli & Seymour, 2006). On the other hand, it is possible that poor morphological awareness may be distinct from poor phonological awareness. In support of this perspective, children with unexpectedly poor comprehension skill despite adequate word reading ability demonstrate a specific morphological deficit independent of phonological difficulties (MacKay et al., 2017; Tong et al., 2011, 2014). Furthermore, a growing body of work suggests that morphological awareness contributes significantly to reading after controlling for phonological awareness, indicating that morphological processing is at least partially distinct from phonological processing (Carlisle & Nomanbhoy, 1993; Deacon & Kirby, 2004; Desrochers et al., 2018).

Neuroimaging has the potential to shed light on these conflicting perspectives regarding morphological awareness in reading (dis)ability. However, few studies have examined the neural correlates of lexical morphology in impaired developing readers. In a recent study of Finnish preschoolers with and without family risk of dyslexia, Louleli et al. (2020) asked participants to listen to sentences with correct and incorrect morphological derivations. This study revealed no significant differences between the control group and high-risk group, raising the possibilities that dyslexia-related differences in morphological processing may not be present in the auditory modality, or prior to reading instruction (Louleli et al., 2020). In contrast, two known studies of older children with dyslexia have revealed hypo-activation during morphological processing (Aylward et al., 2003; Richards et al., 2006), much like the well-documented hypo-activation during phonological processing (Norton et al., 2015). In both studies, children read two words and were asked to decide if those were related in meaning (builder-build vs. corner-corn) during fMRI neuroimaging. Aylward et al. (2003) found that children with dyslexia exhibited reduced activation in the same frontal, parieto-temporal, and occipital regions previously associated with phonological impairments (Norton et al., 2015). Richards et al. (2006) also found that brain activation in children with dyslexia was more bilateral, whereas in typical readers, it was more left-lateralized. However, because the tasks required word reading, questions remain about the extent to which these findings were driven by lower word reading abilities in children with dyslexia.

Studying morphological awareness in spoken language, which precedes and predicts successful reading, may help to shed light on the brain basis of morphology in typical and atypical readers independent of word reading skill. The dual-route model of language processing suggests the involvement of two neurocognitive pathways: a dorsal pathway, which is primarily involved in phonological processing, and a ventral pathway, which is engaged in efficient sound-to-meaning mapping (Hickok & Poeppel, 2007). Similarly, word reading relies on a dorsal circuit associated with phonological analysis and integrating phonological and orthographic information, as well as a ventral circuit associated with efficient mapping between orthography and semantics (Pugh et al., 2000; Shuai et al., 2019). Of particular interest to the current study is the relative contribution of phonological versus semantic mechanisms in morphological processing, and the extent to which each is associated with reading comprehension.

The aim of the present study was to examine the behavioral and neurobiological correlates of morphological awareness, and their relation to reading comprehension skill, in children with and without reading impairment. First, guided by the SVR model (Hoover & Gough, 1990), we ask: Does morphological awareness play a role in reading comprehension, above and beyond the contributions of vocabulary knowledge and word decoding skill? Second, what are the neural mechanisms underlying sensitivity to both root morphemes and derivations, and how are these mechanisms associated with reading comprehension skill? To answer these questions, we asked children in kindergarten through 6th grade, across a broad spectrum of reading ability, to complete an auditory task of morphological awareness during functional near-infrared spectroscopy (fNIRS). We hypothesized that morphological awareness would make a significant contribution to behavioral measures of reading comprehension, and that the brain basis of morphological processing would vary as a function of reading comprehension skill.

Method

Participants

Participants included 97 monolingual English-speaking children (M = 8.62, SD = 1.60, range: 5.92–11.97; 48 boys, 49 girls), across a wide range of reading ability. The sample was 78% white, 19% multi-racial or multi-ethnic, and 3% Black or African American. Participating families were of relatively high socioeconomic status, with mean parental educational attainment of 8.94 on an 11-point scale, corresponding to some post-baccalaureate or Masters’ level schooling. Primary guardians ranged from having some associate’s level or certificate training (5) to having a doctorate degree (11).

All participants were typically developing, with normal hearing and vision, and were proficient English speakers, as indicated by English vocabulary standard scores of 85 or above. Participants were also required to meet a minimum threshold of 62.5% accuracy on the experimental fNIRS task described below. While we do not use a categorical approach for our main analyses, the present study over-sampled children with dyslexia and reading impairment. Participants were considered reading impaired if they scored at least one standard deviation below the mean on at least two out of four standardized reading assessments and/or if their parent reported that they had a reading impairment. Fourteen children satisfied both criteria, and eight children were classified as reading impaired based on their task performance alone. Two were identified as reading impaired by their parent, although their performance fell within the typical range on the day of testing. According to these criteria, nearly a quarter of the sample (N = 24, 14 boys, 10 girls, M_age = 9.61, SD = 1.81) was considered reading impaired. Notably, none of the participants with reading impairment had disproportionately low comprehension skill as compared to their phonological awareness, word recognition, or decoding.

Behavioral measures of language and literacy

Children completed a 1-h battery of standardized language and literacy assessments. Measures of reading ability included the letter-word identification, passage comprehension, word attack, and sentence reading fluency subtests of Woodcock-Johnson IV (Schrank et al., 2014). Participants were considered reading impaired if their standard score fell at or below 85 on at least two of these measures.

Receptive vocabulary was assessed using the Peabody Picture Vocabulary Test Fifth Edition (Dunn, 2018). Children heard a word and were asked to match the meaning of the word to one of four corresponding pictures.

Phonological awareness was assessed using the Comprehensive Test of Phonological Processing (CTOPP-2) Elision subtest (Wagner et al., 2013). Children were asked to repeat a word while removing a phonetic unit. This assessment begins by asking participants to remove a whole syllable (e.g., “Say spider without saying spy”) and progresses to individual phonemes (e.g., “Say time without saying /m/”).

Morphological awareness was assessed using the Early Lexical Morphology Measure (ELMM), which was modeled after the extract the base task (Goodwin et al., 2012), and modified to be accessible to a broader range of children. Children heard a word and were asked to complete a sentence using part of that word (e.g., Noisy. Did you hear that __? [noise]). Notably, ELMM is designed to span the elementary school years and includes both derivational and compound morphology (Marks et al., 2021).

Working memory was assessed using the backward digit span task from the Wechsler Intelligence Scale for Children–Fifth Edition (WISC-V; Weschler, 2014). Children heard a series of numbers and were asked to repeat the series in reverse order. The first items included two numbers, and subsequent items included an increasing number of digits.

Brain basis of morphological processing

In the morphological awareness (MA) neuroimaging task, children heard three English words and were asked to indicate which two words shared a meaningful component. Two of the words presented shared a morpheme (classroom and bedroom), while one was a phonological distractor that shared the same sounds, but not the same meaning (mushroom). During the presentation of the first word (e.g., bedroom), children saw a colored rectangle appear at the top of a computer screen. Children then heard two more words in sequence, corresponding to the presentation of a rectangle in the bottom left corner (e.g., classroom), followed by a rectangle in the bottom right corner (e.g., mushroom) and a question mark. Participants were asked to indicate via button press whether the second word (classroom) or the third word (mushroom) was a better match for the first word.

The MA task consisted of three conditions. In the root morpheme experimental condition, children matched words with a shared free root (e.g., spaceship – battleship – friendship; or winner – winning – window). In the derivational affixes experimental condition, children matched words with a shared derivational affix (e.g., dancer – waiter – corner; or mistake – misspell – mister). In the control condition, children matched whole words (e.g., lady – lady – finish). The control task was designed to tap into whole-word processing, but not awareness of composite morphemes. There were 16 items in each condition, divided into four 30-s blocks of four items each, and separated by a fixed rest period (6 s). The final task had 48 items and was approximately 7.2 min long. The order of the blocks, as well as the order of correct responses, was randomized.

Children were trained on this “word matching game” by an experimenter immediately prior to neuroimaging. An experimenter first introduced the task visually, by presenting and naming three pictures (e.g., sand, sandwich, sandbox). Children then had a conversation with the experimenter about whether the second or third word was a better match for the first word; in this case, sandbox is the better match, because sand and sandbox both refer to real sand. After three trials, the experimenter introduced the button box and explained how words would be presented by a computer. All children understood the task after these three initial practice trials and were able to complete practice trials 4–8 on the computer with high accuracy before continuing on to the experiment itself. All words used in the task training were distinct from those used in the experimental task.

Functional NIRS data acquisition

We first established a priori brain regions of interest in the perisylvian language and literacy network by using published literature to identify dorsal and ventral inferior frontal, superior temporal, and middle temporal regions. We then used the international 10–10 system to build caps corresponding to these a priori regions, by mounting sources and detectors to a custom-built silicone headband with attached grommets. These caps were three different sizes to account for variability in head circumference. The final fNIRS probesets included 12 emitters of near-infrared light sources and 24 detectors spaced ~ 2.7 cm apart in a grid-like shape. This yielded 46 source-detector pairings or data channels, with 23 channels per hemisphere that covered frontal, temporal, and temporo-parietal regions (see Fig. 1). We digitized the geometric structure of the cap on a mannequin foam head using a Polhemus Patriot 6 Degree-of-Freedom Digitizer. The coordinates provided by the digitizer were processed in the AtlasViewer GUI, a MATLAB-based software (Aasted et al., 2015), and transformed to Montreal Neurological Institute (MNI) stereotactic space. Estimated regions covered by each channel and midpoint MNI coordinates are detailed in Table 1. Further details about MNI localization are available from Hu et al. (2020).

Fig. 1.

Functional NIRS probe configuration, visualized on a 3D brain template of left hemisphere reading-related circuits with probe-set overlay. Red and blue optodes correspond to source and detector placement, respectively

Table 1.

Estimated left hemisphere brain regions covered by the fNIRS probeset

			MNI coordinates						MNI coordinates
Source	Detector	Region	x	y	z	Source	Detector	Region	x	y	z
1	1	vIFG, MFG	− 56	50	− 17	4	7	SMG, STG, MTG, IPL	− 56	− 46	− 2
1	2	MFG, dIFG	− 53	49	− 1	4	8	IPL, SMG, AG	− 53	− 43	17
1	3	vIFG, precentral	− 62	29	− 14	4	9	MTG, AG, STG, SMG	− 46	− 59	1
1	4	dIFG, MFG	− 59	33	5	4	10	AG, precuneus, IPL, STG	− 45	− 53	15
2	3	Precentral, STG, IFG	− 65	12	− 11	5	5	MTG, STG	− 67	− 22	− 26
2	4	IFG, precentral, MFG	− 62	17	8	5	7	MTG, STG	− 63	− 36	− 23
2	5	Postcentral, STG, precentral	− 68	− 4	− 9	5	11	ITG, MTG, FG	− 58	− 40	− 40
2	6	Precentral, postcentral	− 64	− 1	11	6	7	MTG, STG, MOG, ITG	− 55	− 50	− 21
3	5	STG, postcentral, IPL, TTG	− 67	− 19	− 6	6	9	MOG ITG, FG, MTG	− 51	− 54	− 38
3	6	Precentral, postcentral, IPL	− 64	− 16	14	6	11	MOG, MTG, ITG	− 45	− 63	− 18
3	7	STG, SMG, IPL, postcentral	− 63	− 33	− 3	6	12	ITG, IOG, MOG, FG	− 44	− 64	− 34
3	8	IPL, postcentral, SMG	− 60	− 30	16

d dorsal; v ventral; IFG inferior frontal gyrus; MFG middle frontal gyrus; STG Superior Temporal gyrus; IPL inferior parietal lobule; MTG middle temporal gyrus; TTG transverse temporal gyrus; SMG supramarginal gyrus; AG angular gyrus; ITG inferior temporal gyrus; FG fusiform gyrus; MOG middle occipital gyrus; IOG inferior occipital gyrus. Regions are reported in the order of greatest probability for each channel. Each channel has a right hemisphere homologue

fNIRS data were collected using a TechEN-CW6 system with 690 and 830 nm wavelengths at a sampling frequency of 50 Hz. Techen-CW6 software signal-to-noise ratio (SNR) minimum and maximum were set to the standard 80 dB and 120 dB range, respectively. For each participant, probes were applied using the international 10–10 transcranial system positioning (Jurcak et al., 2007). Trained experimenters identified the nasion, inon, Fpz, and left and right pre-auricular points. Head circumference was measured, and F7, F8, T3, and T4 were anchored to a specific source or detector. Cardiac signal in the frontal channels was monitored to ensure the quality of optode placement.

Data processing and analysis

The subject- and group-level analyses were done using the NIRS Brain AnalyzIR Toolbox (Santosa et al., 2018), a MATLAB (Mathworks, MA) based software. At the subject level, we trimmed each raw data file to keep only 5 s of pre- and post-experimental task baseline data and downsampled the data from 50 to 2 Hz given that the fNIRS signal of interest lies in the frequency bands of 0–1 Hz. We converted the optical density data to hemoglobin concentration change data using the modified Beer-Lambert law. Each participant’s hemoglobin concentration data was then analyzed using a general linear model (GLM) with prewhitening and robust least square regression (Barker et al., 2013; Friston et al., 2007). We used an autoregressive filter combined with a weighted least square (WLS) estimation approach to eliminate the non-spherical noise structure caused by physiological and motion artifacts in the time series (Barker et al., 2013; Caballero-Gaudes & Reynolds, 2017; Friman et al., 2004). We modeled the canonical hemodynamic response function to peak 6 s after trial onset (Friston et al., 2007). The temporal and dispersion derivatives were added to the canonical HRF function as well as the discrete cosine transform (DCT) matrix to account for signal drift over time. The single subject GLM yielded estimated individual-level regression coefficients for HbO (oxygenated hemoglobin) and HbR (deoxygenated hemoglobin) signal, for each condition, and each channel.

Group-level analyses were then conducted using linear mixed-effects models for each data channel. In our first group-level GLM, we modeled task condition (control, free roots, and derivations) as a fixed effect, participant as a random effect, and the individual-level beta values for HbO and HbR as the predicting dependent variables. Age and socioeconomic status were included as covariates. In the second GLM, we included the interaction between task condition and reading comprehension ability and modeled vocabulary knowledge as a covariate. Estimated group-level channel-based effects were extracted for the experimental condition(s) > control contrasts. We then plotted the group-level effects (unstandardized betas) for each contrast on the MNI 152 brain template using the previously digitized MNI coordinates. The analyses presented below focus on HbO as it accounts for a larger portion of the signal (HBO 76%; HBR 19%), in part because fNIRS instruments such as TechEN CW6 capture the HBO signal with greater reliability (Gagnon et al., 2012). We only present the effects that survived FDR correction for multiple comparisons.

Results

Language and reading skill

Descriptive statistics for all language and literacy measures are presented in Table 2.

Table 2.

Descriptive statistics

	M	(SD)	Range
Age	8.62	(1.60)	5.92–11.97
Parental education	8.94	(1.71)	5–11
Vocabulary1	115.76	(16.20)	85–160
Word reading1	104.21	(18.77)	46–136
Decoding1	106.24	(15.59)	59–136
Passage comprehension1	98.85	(17.06)	40–127
Sentence reading fluency1	102.47	(18.01)	42–138
Phonological awareness2	9.93	(2.66)	3–15
Morphological awareness3	28.85	(8.56)	2–40
Working memory3	7.53	(1.96)	3–13

N = 97.

¹Standard score, typical range: 85–115;

²scaled score, typical range: 8–12;

³raw score

Participants had high average English language ability, with a mean vocabulary standard score of 115.76. Mean standard scores on standardized literacy assessments fell within the typical range, between 98 and 106. However, N = 24 participants were considered reading impaired (N = 22 of whom scored 85 or below on two or more literacy tasks, and two who were identified as reading impaired by their parents). At the other end of the spectrum, N = 25 were highly precocious readers, scoring 115 or above on two or more literacy tasks. Participants thus spanned a wide range of reading proficiency.

Children successfully completed the fNIRS morphological awareness task with high accuracy. The derivational affixes condition of the neuroimaging task was most challenging, with a mean accuracy of 63.88% (SD = 15.23%). Accuracies were significantly higher during the free roots condition (M = 84.47%, SD = 11.34%) compared to derivations (t(95) = 16.96, p < 0.001), and response times were significantly faster (t(95) = − 5.33, p < 0.001). Children performed near ceiling on the control condition, with a mean accuracy of 94.55% (SD = 8.21%). Correlations between language and literacy variables, as well as overall fNIRS task accuracy, are presented in Table 3. The fNIRS accuracy was most closely associated with word reading (r = 0.50, p < 0.001) and reading comprehension (r = 0.44, p < 0.001).

Table 3.

Partial correlations between language and literacy variables

	1	2	3	4	5	6	7	8
1. Vocabulary	-
2. Word reading	0.44***	-
3. Decoding	0.40***	0.86***	-
4. Passage comp	0.51***	0.85***	0.73***	-
5. Reading fluency	0.44***	0.73***	0.60***	0.73***	-
6. Phon. awareness	0.40***	0.65***	0.70***	0.64***	0.51***	-
7. Morph. awareness	0.35**	0.66***	0.57***	0.65***	0.47***	0.47***	-
8. Working memory	0.08	0.25*	0.16	0.24*	0.16	0.19	0.17	-
9. fNIRS task acc	0.20	0.50***	0.35**	0.44***	0.31**	0.27*	0.34**	0.12

Raw scores, controlling for age

Morphological awareness and reading comprehension

We used hierarchical regression to test our hypothesis that morphological awareness is related to reading comprehension (Table 4). At step 1, we entered children’s age, parental educational attainment, and whether or not they were classified as reading impaired. At step 2, guided by the simple view of reading, we entered vocabulary knowledge (PPVT; Dunn, 2018) and decoding skill (word attack; Schrank et al., 2014). Finally, at step 3, we entered two measures of morphological awareness: ELMM score and accuracy on the MA neuroimaging task. The final model accounted for 83% of the variance in reading comprehension. Results showed that age (β = 0.21, t = 2.32, p = 0.023), reading impairment (β = − 0.29, t = 4.14, p < 0.001), decoding ability (β = 0.26, t = 3.60, p = 0.001), and morphological awareness as measured by both the ELMM (β = 0.29, t = 3.54, p = 0.001) and accuracy on the fNIRS task (β = 0.13, t = 2.33, p = 0.022) were all significant predictors of reading comprehension skill. Once morphological awareness was added to the model, vocabulary was no longer a significant predictor of reading comprehension (β = 0.12, t = 1.56, p = 0.123).

Table 4.

Hierarchical regression explaining variance in reading comprehension

	Std β	t	p	R	R2	ΔR2
Step 1				0.825	0.681
Intercept		− 0.63	.533
Age	0.78	11.89	< .001
Parental education	− 0.01	− 0.10	.924
Reading impaired?	− 0.64	− 9.02	< .001
Step 2				0.880	0.755	0.094
Intercept		− 1.68	.098
Age	0.36	3.73	< .001
Parental education	− 0.02	− 0.39	.696
Reading impaired?	− 0.36	− 4.67	< .001
Vocabulary	0.19	2.19	.032
Decoding	0.39	4.96	< .001
Step 3				0.911	0.830	0.056
Intercept		− 2.56	.012
Age	0.21	2.32	.023
Parental education	− 0.01	− 0.23	.817
Reading impaired?	− 0.29	− 4.14	< .001
Vocabulary	0.12	1.56	.123
Decoding	0.26	3.60	.001
Morphological awareness	0.29	3.54	.001
fNIRS task accuracy	0.13	2.33	.022

N = 87 with complete data. Final F(7, 80) = 55.94, p < .001

To test whether morphological awareness made a similar contribution to reading comprehension across a broad range of reading skill, we conducted a post-hoc regression that included two additional reading impairment × morphological awareness interaction terms, one for each measure of morphology. Neither interaction was significant, indicating that both typical and impaired readers had similar associations between morphological awareness and their reading comprehension skill.

Brain basis of morphological processing

Our first step in analyzing the neuroimaging data was to examine the brain basis of morphological processing across all participants. Compared to the resting baseline, our MA task incurred widespread activation in perisylvian language regions, including the bilateral inferior frontal gyrus, primary auditory cortex, superior/middle temporal gyrus, and supramarginal gyrus (Fig. 2).

Fig. 2.

Brain activity during the morphological awareness task conditions

We then examined brain activity specific to the free root morpheme condition (winner – winning – window) and the derivational affixes condition (dancer – waiter – corner). The within-group comparison for the roots > control contrast revealed that, compared to whole-word processing, attention to free root morphemes incurred significantly greater activity in the bilateral middle/inferior frontal gyri (MFG/IFG), as well as the left anterior superior temporal gyrus (STG), and left middle/inferior temporal gyrus (MTG/ITG). The affixes > control contrast revealed that, compared to whole-word processing, attention to derivations also involved significantly greater activation in left MFG/IFG and posterior temporal regions, as well as right precentral gyrus and STG. The comparison of affixes > roots revealed greater left inferior frontal/precentral, and left occipito-temporal activation for affixes, and greater left temporal lobe activation for free roots. Affixes also incurred less substantial occipito-temporal deactivation than root morpheme processing. Beta values for the comparisons between conditions are detailed in Table 5.

Table 5.

fNIRS channels significantly activated during morphological awareness task conditions

H	Channel	Region	β	T-stat	p	q
Root morphemes > whole-word processing
L	1.1	vIFG, MFG	3.25	8.93	< .001	< .001
L	1.3	vIFG, precentral	1.32	3.37	.001	.003
L	1.4	dIFG, MFG	0.74	3.01	.003	.008
L	2.3	Precentral, STG, IFG	2.08	3.99	< .001	< .001
L	5.5	MTG, STG	2.49	8.05	< .001	< .001
L	5.11	ITG, MTG, FG	0.71	3.24	.001	.004
R	6.12	ITG, IOG, MOG, FG	0.88	3.25	.001	.004
Derivational affixes > whole-word processing
L	1.1	vIFG, MFG	2.58	6.14	< .001	< .001
L	1.3	vIFG, precentral	2.24	4.91	< .001	< .001
L	1.4	dIFG, MFG	1.63	5.67	< .001	< .001
R	2.4	IFG, precentral, MFG	1.15	2.83	.005	.016
L	4.9	MTG, AG, STG, SMG	0.70	2.69	.007	.022
L	6.12	ITG, IOG, MOG, FG	0.84	2.60	.010	.027
Roots > affixes
R	1.1	vIFG, MFG	1.06	− 3.21	.001	.007
L	2.3	Precentral, STG, IFG	1.74	− 2.66	.008	.032
L	2.5	Postcentral, STG, precentral	1.00	− 2.50	.013	.047
L	5.5	MTG, STG	2.21	− 5.54	< .001	< .001
R	6.9	MOG ITG, FG, MTG	2.16	− 4.58	< .001	< .001
R	6.11	MOG, MTG, ITG	2.25	− 5.29	< .001	< .001
R	6.12	ITG, IOG, MOG, FG	4.42	− 15.68	< .001	< .001
Affixes > roots
L	1.4	dIFG, MFG	0.89	2.84	.005	.021
R	2.5	Postcentral, STG, precentral	1.06	2.76	.006	.026
L	4.7	SMG, STG, MTG, IPL	1.03	3.90	< .001	.001
L	4.8	IPL, SMG, AG	0.86	2.47	.014	.048
L	4.9	MTG, AG, STG, SMG	1.53	5.44	< .001	< .001
L	6.7	MTG, STG, MOG, ITG	1.10	3.21	.001	.007
L	6.11	MOG, MTG, ITG	1.59	4.42	< .001	< .001

L left hemisphere; R right hemisphere. Channel Source.Detector. q significance level after FDR correction. d dorsal; v ventral; IFG inferior frontal gyrus; MFG middle frontal gyrus; STG superior temporal gyrus; IPLinferior parietal lobule; MTG middle temporal gyrus; SMG supramarginal gyrus; AG angular gyrus; ITG inferior temporal gyrus; FG fusiform gyrus; MOG middle occipital gyrus; IOG inferior occipital gyrus. Regions are reported in the order of greatest probability for each channel

Interaction with reading comprehension skill

Our next aim was to examine the association between reading comprehension skill and the brain basis of morphological processing. To do this, we ran another GLM in which we modeled the main effects and interaction between task condition (derivational affixes, free roots, and control) and children’s reading comprehension standard score, including vocabulary standard score as a covariate. The standard scores reflect children’s reading and vocabulary skills as they relate to age-appropriate norms (Dunn, 2018; Schrank et al., 2014). As children at the lower end of this distribution are those with reading comprehension skills far below grade-level expectations, we did not include age as a separate regressor.

Results demonstrated a significant interaction between reading comprehension and morphological processing in largely left-lateralized language regions of the brain. Common to both the roots and affixes condition, children with better reading comprehension for their age demonstrated increased activation in the left ventral IFG/anterior STG and right posterior STG as compared to impaired readers. Better reading comprehension was also associated with lower engagement of the right posterior temporal/occipital cortex. During the derivations condition specifically, we observe additional brain-behavior associations in the left vIFG, MTG, and IPL, as pictured in Fig. 3 (see also Table 6). Children with better reading comprehension demonstrated a greater increase in activation for derivations in the IFG and MTG and less deactivation in the IPL. Separate brain-behavior associations with reading comprehension raw scores, vocabulary raw scores, and age are presented in Supplementary Fig. 1, revealing that skilled reading comprehension is consistently positively associated with left parietal activity.

Fig. 3.

Interaction between reading comprehension and morphological awareness task activity. MACondition: control, roots, or derivations. ReadingCompSS = reading comprehension standard score (age adjusted); VocabSS = vocabulary standard score (age adjusted). vIFG: ventral inferior frontal gyrus; IPL: inferior parietal lobe; MTG: middle temporal gyrus

Table 6.

Brain-behavior interactions with reading comprehension

H	Channel	Region	β	T-stat	p	q
Root morphemes × reading comprehension
L	1.1	vIFG, MFG	0.08	3.08	.002	.040
R	3.5	STG, postcentral, IPL, TTG	0.09	3.48	.001	.024
R	4.9	MTG, AG, STG, SMG	− 0.09	− 3.28	.001	.027
R	6.9	MOG ITG, FG, MTG	− 0.13	− 3.91	< .001	.009
Derivational affixes × reading comprehension
L	1.1	vIFG, MFG	0.13	4.86	< .001	< .001
L	1.3	vIFG, precentral	0.08	2.89	.004	.026
R	1.4	dIFG, MFG	0.07	3.04	.002	.018
R	2.5	Postcentral, STG, precentral	0.09	3.03	.003	.018
L	3.5	STG, postcentral, IPL, TTG	0.12	4.42	< .001	< .001
R	3.5	STG, postcentral, IPL, TTG	0.08	2.85	.004	.027
L	3.7	STG, SMG, IPL, postcentral	0.09	3.40	.001	.006
R	3.8	IPL, postcentral, SMG	0.13	3.84	< .001	.002
L	4.10	AG, precuneus, IPL, STG	0.13	4.86	< .001	< .001
L	5.5	MTG, STG	0.10	3.78	< .001	.002
L	6.9	MOG ITG, FG, MTG	− 0.15	− 4.20	< .001	< .001
R	6.9	MOG ITG, FG, MTG	− 0.17	− 4.68	< .001	< .001
R	6.12	ITG, IOG, MOG, FG	− 0.11	− 4.64	< .001	< .001

L left hemisphere; R right hemisphere. Channel Source.Detector. q significance level after FDR correction. d dorsal; v ventral; IFG inferior frontal gyrus; MFG middle frontal gyrus; STG superior temporal gyrus; IPL inferior parietal lobule; MTG middle temporal gyrus; TTG transverse temporal gyrus; SMG supramarginal gyrus; AG angular gyrus; ITG inferior temporal gyrus; FG fusiform gyrus; MOG middle occipital gyrus; IOG inferior occipital gyrus. Regions are reported in the order of greatest probability for each channel

Discussion

This study asked two main questions. First, how does morphological awareness support reading comprehension in children with and without reading impairment? Second, how is the brain basis of morphological processing associated with skilled reading comprehension? To answer these questions, we analyzed behavioral and fNIRS neuroimaging data from a large sample of children across a wide range of reading ability. At the behavioral level, we found that morphological awareness made a substantial contribution to reading comprehension in typical and impaired readers. At the neural level, we discovered an interaction between reading comprehension skill and brain activation during morphological processing in left inferior frontal, middle temporal, and inferior parietal brain regions often associated with reading development. These results shed new light on the role of morphological awareness in reading comprehension and the neural mechanisms supporting this association in children across a broad range of reading abilities.

Morphological awareness contributes to reading comprehension

There has been some disagreement in the field as to the role of morphological awareness in reading for children with dyslexia. Many studies have reported lower performance on tasks of morphological awareness compared to age-matched controls (Berthiaume & Daigle, 2014; Casalis et al., 2004; Kearns et al., 2016; Tsesmeli & Seymour, 2006). At the same time, others suggest that morphological awareness might be a relative strength or a compensatory factor for impaired readers (Deacon et al., 2019; Law et al., 2018). This study examined the contribution of morphological awareness to reading comprehension in both typical and impaired readers. We examined the possibility of an interaction between a binary classifier of reading impairment and morphological awareness and found no evidence for an interaction, suggesting that both typical and impaired readers were relying on morphological awareness to a similar extent.

More specifically, our findings suggest a relation between reading comprehension and morphological awareness, measured both within the sentence context and at the single-word level. Our sentential measure asked children to complete a sentence by extracting the base of a multimorphemic word, as in, “Colorful. That flower is such a pretty ___ [color].” Our neuroimaging task presented children with three individual words (e.g., farmer—waiter—corner) and asked children to identify the two that shared a morpheme (farm + ER and wait + ER). Both measures assessed children’s sensitivity to compound and derivational morphology. We found that each morphology measure made a unique contribution to participants’ reading comprehension, and together accounted for an additional 5.6% of variance explained, above and beyond the contributions of demographic factors, vocabulary knowledge, and decoding skill.

This new evidence contributes to the growing body of work advocating for the importance of morphological awareness in the reading process across a wide range of ages, skill levels, and languages (e.g., Arabic: Vaknin-Nusbaum & Saiegh-Haddad, 2020; Chinese: Cheng et al., 2017; Pan et al., 2016; French and Greek: Desrochers et al., 2018). In English, numerous studies have revealed both direct and indirect relations between morphological awareness and reading comprehension (Deacon et al., 2014; Kieffer & Lesaux, 2012; Levesque et al., 2017). Our present findings extend this evidence, demonstrating a robust association between morphological awareness and reading comprehension in readers across a wide range of literacy skill.

Brain basis of morphological processing

Current models of the neurobiology of dyslexia have been largely informed through phonological reading tasks, such as rhyme judgments (Hoeft et al., 2007; Kovelman et al., 2012; Tanaka et al., 2011). While phonological processes are often seen as a stepping stone to learning to read, the ability to recognize larger units of meaning is essential for later literacy and successful reading comprehension (Rastle, 2018). Yet, little is known about the brain basis of morphological awareness or the extent to which it might vary in impaired readers.

The present study used fNIRS to investigate the neurocognitive mechanisms associated with morphological processing in the auditory modality. Children heard three words and identified the two words that shared a meaningful component. Our experimental design contrasted three conditions: a free root morpheme matching condition (teacup—teapot—T-rex), a derivational affix matching condition (reset—replay—reading), and a whole-word processing control condition. We focus on the auditory modality for two reasons: first, because the brain basis of spoken word processing precedes and predicts successful reading development (Marks et al., 2019), and second, to ensure that performance was not confounded by single-word reading ability in participants with dyslexia.

Examining the neural processes associated with each condition revealed both common and condition-specific patterns of brain activity. Whole-word, root morpheme, and derivational affix processing all recruited bilateral auditory language processing regions (IFG and STG) and demonstrated relative deactivation of posterior brain regions. Compared to the whole-word processing control, the two morphological awareness conditions both revealed greater engagement in several hubs of the semantic system (Binder et al., 2009). Processing free roots incurred greater left temporal activity than whole words, especially in the left MTG, a region associated with lexical or semantic retrieval (Binder et al., 2009). Roots also incurred greater activation in the left IFG and anterior STG. The IFG has been associated with phonological, semantic, and syntactic processing (Vigneau et al., 2006). More specifically, the ventral aspect of the IFG (BA 47) is typically associated with complex semantic retrieval analyses, and the more dorsal aspect of the left IFG (BA 45/44) region is typically associated with both phonological (Ip et al., 2019) and morpho-syntactic (Kovelman et al., 2008; Skeide & Friederici, 2016) language processes. Activity in the anterior STG is associated with syntactic complexity (Brennan et al., 2012). Notably, Arredondo et al. (2015) similarly reported IFG and aSTG activation during a morphological judgment task with children. In other words, the free root morpheme condition relied on brain regions associated both phonological and semantic processes, as well as those implicated in segmentation and word analysis.

Similar to the roots condition, derivations also incurred greater left frontal and temporal activity than whole-word processing. In particular, wider-spread prefrontal activity suggests this condition was more effortful, as also demonstrated by children’s lower accuracy and higher response times. This is logical as derivational affixes are more semantically abstract compared to free roots and may be more analytically demanding. Together, these results speak to the multi-faceted nature of morphological processing, which requires one to connect analysis of a word’s underlying structure to representations of meaning.

Brain-behavior associations with reading comprehension skill

To uncover the relation between reading comprehension and the neural bases of morphological awareness, we examined the interaction between participants’ reading comprehension and their brain activity during the MA task. We observed an interaction between reading ability and engagement of the left IFG/anterior STG and right posterior STG regions common to both the roots and derivations condition. Furthermore, children with better reading comprehension showed greater engagement of the left ventral IFG, left MTG, and left IPL associated with derivations specifically. Notably, hypoactivation has frequently been reported the left IFG, MTG, and IPL in studies of dyslexia (Richlan, 2014; Shuai et al., 2019).

The locations of these brain-behavior interactions are closely aligned with theoretical models of auditory word processing and literacy development. In particular, beginning word reading largely relies on frontal (IFG) and dorsal (STG/IPL) regions associated with phonological analysis and integrating phonological and orthographic information (Pugh et al., 2000; Shuai et al., 2019). Furthermore, the left ventral IFG/anterior STG region is thought to be involved in building syntactic structure during language processing (Bemis & Pylkkanen, 2013; Brennan et al., 2012). In our MA task, both the roots and derivations conditions require structural analyses for the morpho-phonological segmentation of spoken words and specific attention to mapping units of sound onto units of meaning. Our findings suggest that perhaps better readers can more effectively tap into the phonological and structural analyses associated with these brain regions, a processing feature that cascades to benefit their text reading abilities.

Of particular note are the derivations-specific associations with reading comprehension in the parietal lobe. Numerous studies have pointed towards functional and structural differences in the inferior parietal region in dyslexia across a variety of tasks. This leads to several possible explanations for this association between reading comprehension and IPL activity. One possibility is that reduced IPL activity may be associated with phonological deficits. Studies of impaired readers frequently report relatively lower engagement—or greater task-related deactivation—of the left IPL during phonological awareness tasks. For instance, Hoeft et al. (2007) reported greater left parietal deactivation among children with dyslexia during a visual word rhyme judgment task, as well as reduced gray matter volume in this region compared to both age-matched and reading-level-matched controls. Alternatively, this brain-behavior association could be associated with a specific semantic deficit. Reduced IPL activity has been observed during semantic judgments, both at the single-word level (Booth et al., 2007) and in sentence processing (Schulz et al., 2008). Landi et al. (2010) discovered that adolescents with dyslexia under-activate left inferior parietal regions during both phonological and semantic processing, in both the auditory and visual modalities. In line with this prior work, our findings similarly suggest that impaired readers show greater deactivation in the left inferior parietal cortex during a morphological awareness task. This effect was more robust during the derivations condition than the roots condition, likely due to the semantically abstract and analytically complex nature of derivational morphology. As the IPL is classically associated with integrating phonological, semantic, and orthographic representations, these results suggest that poor readers may struggle to efficiently manipulate and integrate units of sound and meaning.

Theoretical implications

Our neurocognitive findings support and extend theories of reading comprehension. The reading systems framework (Perfetti & Stafura, 2014) suggests that morphology should contribute to reading comprehension at both the single-word level and the sentence level. We complement this perspective by demonstrating that morphological awareness makes a significant contribution to passage comprehension, likely through its role at both single-word and sentence processing levels, in both typical and impaired readers. Going one step further, Kirby and Bowers’ (2017) binding agent theory suggests that morphology may be the “glue” that connects and integrates mental representations of phonology and semantics to one another, and to orthography. The robust engagement of the neural networks often associated with syntactic, semantic, phonological, and orthographic language processes further reinforces this perspective. Binding may take place at the level of these language processes, as suggested by activations in language-associated regions and/or in the IPL region classically associated with speech-to-print mapping. These findings suggest that successful reading comprehension, and its deficit in impaired readers, may relate to children’s ability to effectively integrate the units of sound and meaning in speech. The present study thus helps to bridge our understanding of the neurobiology of language, and children’s sensitivity to morpho-phonological language structure, to theories of reading comprehension.

Future directions

This study examined the neurocognitive bases of morphological processing in typically developing and impaired readers, ages 6–11. To the best of our knowledge, the present work is the first to suggest that functionality of the left parietal region modulates the relation between morphological awareness and reading comprehension. One particular strength of the current study is its methodological approach, which combines neural and behavioral measures in a relatively large sample of readers across a wide range of reading proficiency. As reading (dis)ability falls along a broad spectrum, with many possible areas of weakness for struggling readers, we did not dichotomize our sample to compare the association between morphology and reading comprehension across groups. Nevertheless, future research may be interested in a direct comparison between clinically impaired and typically developing readers. Future studies might also consider examining developmental differences between younger and older students, as children’s knowledge and application of derivational morphology continue to develop over time (e.g., Kuo & Anderson, 2006).

The present findings, which shed new light on morphological processing at the single-word level, inspire further questions about the neurocognitive processes involved in comprehension beyond single-word recognition. Recent theoretical perspectives suggests that morphology makes a multi-dimensional contribution to reading comprehension. In particular, the morphological pathways framework (Levesque et al., 2020) notes the importance of both implicit and explicit morphological skills in supporting comprehension through underlying reading processes such as morphological analysis and morphological decoding. Although this study reveals the brain bases of implicit morphological processing, neither experimental task in this study required explicit or conscious knowledge of morphemes. We may deepen our theoretical understanding of these reading processes by studying the brain bases of both implicit and explicit morphological awareness, in spoken language (as in the present study), as well as in print processing.

Our behavioral measures are not as extensive as might be possible in a behavioral-only approach. The present measures were limited to only two tasks of morphological awareness, although there are many others that tap into morphological analyses in greater detail (Goodwin et al., 2017; Levesque et al., 2019). The relation between these morphological skills, and the neurocognitive processes that support them, are important directions for future research. We might also ask how processing free root morphemes, as studied here, might differ from processing bound roots (e.g., in-ject, e-ject-ion). Furthermore, we recognize that our sample was relatively demographically homogeneous and of predominantly high SES, which may impede the generalizability of our findings. Finally, although locations of each fNIRS channel have been systematically mapped to MNI space using both photogrammetry and MRI (see Hu et al., 2020), we should nevertheless be cautious when mapping pediatric data to standard space. However, because age-related differences in brain morphology are most pronounced in subcortical regions (e.g., Fonov et al., 2011), this limitation may be less of a concern for our fNIRS analysis. fNIRS provides rich information about cortical activity in the perisylvian language network, but fNIRS data cannot support inferences about subcortical regions or regions outside the scope of the probeset.

Conclusions

The present study sheds light on the association between morphological awareness and reading comprehension, and the neurocognitive mechanisms underlying this association, in typical and impaired readers. Our findings add to a growing body of knowledge indicating the importance of morphology for successful reading. We provide some of the first evidence of distinct neurocognitive processes underlying free root and derivational morphological processing in developing readers of English, and reveal an interaction between morphological processing and reading skill. Our findings indicate that better reading comprehension is associated with increased activation in left hemisphere brain regions associated with language processing and speech-to-print mapping. These findings not only underscore the importance of morphological awareness for successful reading development, but highlight the specific importance of derivational morphology for reading comprehension in English.

Data and code availability

Data will be made available upon reasonable request.