Abnormal brain activation during speech perception and production in children and adults with reading difficulty

Reading difficulty (RD) is a common neurodevelopmental disorder (5–17% prevalence) characterized by persistent impairments in accurate and fluent reading despite adequate intelligence and instruction. A core feature is a phonological processing deficit that manifests early in speech processing and can predict later literacy outcomes. Prior behavioral, EEG, and neuroimaging work has shown reduced phoneme discrimination, less distinct phonemic representations, and atypical activation in temporal and frontal speech-related cortices in individuals with RD. Fewer studies have examined speech production, especially using objective acoustic measures, and little is known about how phonological deficits are reflected in foreign speech perception and production across development. This study asks whether foreign (Spanish) speech perception and production differ between Chinese children and adults with RD versus controls, whether age effects differ by RD status, and whether fNIRS activation patterns during these tasks can serve as biomarkers of RD. The authors hypothesized group differences in fronto-temporo-parietal regions, altered developmental (age) effects in RD, and that multivariate patterns would classify RD status.

Extensive prior research links RD to phonological deficits evident in both perception and production. Behaviorally, individuals with RD show poorer discrimination of minimal pairs and less sharply defined phoneme boundaries. Neuroimaging reveals reduced activation in prefrontal and superior temporal cortices during auditory phonological tasks, and multivoxel analyses indicate less distinct phonemic representations in at-risk beginning readers. EEG studies report smaller and delayed MMN/MR and LDN responses to speech deviants in children and adults with RD, with some evidence of persistent or stronger deficits in adults. In Chinese RD, reduced activation in dorsal left IFG during auditory rhyming and diminished MMN for cross-category deviants indicate impaired categorical perception, though developmental differences between adults and children in Chinese RD had not been explored. Speech production studies emphasize slower speaking rates, longer pauses, and difficulties with multisyllabic repetition in children at risk and those with RD, with some deficits persisting into adulthood. However, objective acoustic measures such as VOT and vowel formants have been underused, and brain activity during production in RD is understudied. In foreign-language contexts, RD has been associated with weaker MMN to L2 words and mixed findings for production (e.g., deficits in English vowel contrasts in French RD adults). Language distance moderates L2 perception/production difficulties; Spanish was selected here due to its phonological contrast with Chinese to maximize detection of RD-related deficits.

Participants: 87 native Chinese, right-handed participants with no Spanish learning history were recruited: 20 children with RD (mean age 11.00, 12 males), 24 age-matched children without RD (mean 10.58, 9 males), 20 adults with RD (mean 19.63, 9 males), 23 age-matched adults without RD (mean 19.65, 10 males). RD inclusion: Raven standard score > 80 and z < −1.5 on at least one of three reading tests (Chinese character naming, Chinese sentence reading fluency, one-minute Chinese character naming). Controls: Raven > 80 and z > −1 on all three tests. IRB approval and consent/assent obtained. Behavioral assessments: Screening tests included Chinese character naming (accuracy), Chinese sentence reading fluency (comprehension under time limit), and one-minute character naming (regular and irregular). Additional measures: phonological awareness (English initial sound deletion; pseudoword rhyming), morphological awareness (Chinese homophonic and homographic morpheme tasks), orthographic awareness (character correction; delayed copy), working memory (forward/backward digit spans), and rapid automatized naming (digits, pictures). Adult and child norms were established where needed. Tasks and stimuli: fNIRS speech perception task used a rapid event-related design with 120 pairs of Spanish CV syllables across four conditions: identical/different and high/low similarity to Chinese, plus 60 baseline trials. Each trial presented two syllables (800 ms each, 200 ms gap), with jittered SOA (3.5–4 s). Speech production task required imitation of 26 multisyllabic Chinese pseudowords and 26 multisyllabic Spanish words, each repeated three times (156 trials total across two runs). Stimuli presentation included 1500 ms audio followed by 1500 ms imitation cued by a red cross; ITI jittered (250–1000 ms). Baseline trials were interleaved. Phonetic analyses: For production, voice onset time (VOT) for /b/ and /d/ and vowel formants (F1, F2) for five Spanish vowels (a, o, i, e, u) were measured from five Spanish words (dificil, dado, brazo, bueno, bebe). Formants were extracted in Praat and normalized with the Lobanov method using the Vowels R package to mitigate anatomical differences. Euclidean distance in F1–F2 space from a native Spanish model speaker quantified vowel similarity. fNIRS acquisition and preprocessing: CW-NIRSport2 system (760/850 nm), 4.4 Hz sampling, 3.0 cm source-detector distance; two 4×4 probe sets over bilateral frontal, parietal, and temporal areas (48 channels). Optode placement followed the 10–20 system; anatomical localization used T1-weighted MRI (MNI normalization; AAL atlas). Quality control excluded channels/participants lacking clear cardiac components; approximately three channels per participant were excluded on average. Processing (Homer2): intensity to OD conversion, wavelet motion correction (iqr=0.8), bandpass filter (0.02–0.5 Hz), conversion to HbO/HbR via modified Beer–Lambert (partial pathlength factor 6.0). Analyses focused on HbO due to higher SNR. GLM analysis: In NIRS-KIT, design matrices modeled perception conditions (identical/different × high/low similarity; baseline) and production conditions (Chinese, Spanish; baseline). Regressors were convolved with the canonical HRF; contrasts (lexical minus baseline) provided channel-wise beta values. Group-level statistics used repeated-measures ANCOVAs with factors of Group (RD, controls), Age (children, adults), and relevant within-task factors; Raven was a covariate. Multiple comparisons were controlled with FDR across 48 channels. Classification: Feature vectors comprised 48-channel beta values per task. RBF-SVM (scikit-learn) with LOOCV evaluated performance. Hyperparameters C and γ were grid-searched (powers of 2) within training folds. Recursive feature elimination (nested LOOCV) identified optimal feature subsets. Permutation testing (1000 shuffles) estimated chance distributions; p-values were the proportion of permuted accuracies exceeding observed accuracy.

Behavioral and phonetic results: Age and Raven were comparable within child and adult groups except adults with RD had lower Raven than adults without RD (t(27.48)=4.91, p<0.001); Raven was included as a covariate and subgroup analyses matched on Raven replicated the whole-sample findings. Vowel formants: A Group×Age ANCOVA (Raven covariate) showed a significant main effect of Age on average distance to a native speaker across vowels (F(1,84)=12.10, PFDR=0.003, ηp²=0.13): children were closer to the native model than adults. No main effect of Group and no interaction. Simple effects indicated children without RD were more similar to the native speaker than adults without RD (t(25.25)=−3.66, p=0.001), but no child–adult difference within RD (t(38)=−1.52, p=0.14). Critically, children without RD outperformed children with RD (t(40)=−4.69, p<0.001); no adult group difference (t(40)=−0.29, p=0.77). VOTs: All groups differed from the native model for /b/ and /d/, but neither Age nor Group main effects nor their interaction reached significance (e.g., /b/: F(1,82)=1.10, PFDR=0.30, ηp²=0.01; Group F(1,82)=0.03, PFDR=0.97; interaction F(1,82)=4.23, PFDR=0.093). Perception fNIRS: Repeated-measures ANCOVA (Group×Age×Similarity×Syllable Consistency; Raven covariate) found no main effects but a significant three-way interaction (Age×Group×Similarity) in right IFG (CH31: F(1,80)=10.07, PFDR=0.048, ηp²=0.11) and right DLPFC (CH37: F(1,80)=16.16, PFDR=0.006, ηp²=0.17). Adults with RD showed greater activation than adults without RD when perceiving Spanish syllables highly similar to Chinese (CH31 and CH37: F(1,80)=6.92, p=0.010); no child group differences and no group effects for low-similarity syllables. Developmentally, activation decreased from children to adults in controls but not in RD (greater decrease in controls: CH31 F=6.05, p=0.016; CH37 F=6.66, p=0.012). Production fNIRS: ANCOVA (Group×Age×Language; Raven covariate) showed a significant Language×Group interaction in left MTG (CH22: F(1,74)=11.53, PFDR=0.048, ηp²=0.14). Controls exhibited greater deactivation for Spanish than Chinese (F(1,74)=8.12, p=0.006), whereas RD showed greater deactivation for Chinese than Spanish (F(1,74)=5.64, p=0.020). RD had reduced deactivation relative to controls during Spanish production (F(1,74)=7.10, p=0.002); no group difference for Chinese (F(1,74)=0.17, p=0.68). Brain–behavior correlations: In RD children, right DLPFC activation (CH37) during perception of high-similarity Spanish syllables negatively correlated with pseudoword rhyming (r=−0.629, PFDR=0.084), and this correlation differed significantly from RD adults (Steiger’s z=2.90, p=0.004). Classification (SVM with LOOCV and RFE): Accuracies ranged from 60% to 90% across tasks/groups. Significant models (permutation-tested) included: adults classified using Spanish perception with low similarity (82%, p=0.002), adults using Chinese production (82%, p=0.004), children+adults using Chinese production (73%, p=0.012), and additional combined-task models achieving up to 87% (p=0.003) and 81% (p=0.031). Features with high discriminative power recurrently included left MTG (anterior CH22 and posterior CH24), left premotor cortex, SMA, and left IFG; other contributory regions varied by model (e.g., right IFG, DLPFC, premotor/SMA, postcentral).

The study demonstrates developmental and task-specific abnormalities in speech processing networks in RD. In perception, adults typically show reduced right IFG and right DLPFC engagement relative to children for stimuli similar to the native language, reflecting decreased effort with maturation; this age-related decrease was attenuated in RD, indicating slowed or atypical development of frontal systems supporting phoneme discrimination and executive functions. The negative association between right DLPFC activation and phonological awareness in RD children suggests that better phonological skills reduce executive demands during speech perception. In production, RD readers exhibited reduced deactivation of the anterior left MTG (part of the default mode network) for Spanish relative to controls, suggesting inefficiency in suppressing task-negative processes during the more challenging foreign imitation. Controls showed stronger DMN deactivation for Spanish than Chinese, consistent with higher task demands; RD showed the opposite pattern, aligning with prior evidence of DMN abnormalities in developmental disorders and suggesting that atypical DMN modulation contributes to phonological production difficulties in RD. Behaviorally, children without RD outperformed adults without RD on Spanish vowel pronunciation and outperformed RD children, supporting the notion that children hold an advantage in foreign speech learning that is diminished in RD, potentially due to impoverished phonological representations. Multivariate classification revealed that distributed activation patterns, especially in left MTG, left premotor cortex, SMA, and left IFG, reliably distinguish RD from controls across ages and tasks, converging with models implicating posterior MTG in phonological representations and premotor/SMA/IFG in speech-motor planning, phonological memory, and selection. Together, findings address the research question by linking RD-related phonological deficits to atypical developmental trajectories and altered network engagement in foreign speech perception and production, and by identifying reproducible neural markers of RD.

This study provides converging behavioral and neurofunctional evidence that individuals with RD show impaired foreign speech processing, including poorer vowel production in children and atypical brain activation patterns during perception and production. Developmentally attenuated decreases in right IFG and DLPFC activation from childhood to adulthood in RD suggest slowed maturation of frontal systems. During production, RD readers exhibit reduced DMN deactivation in left MTG for Spanish, indicating inefficiencies in engaging task-positive networks for challenging phonological output. Multivariate analyses identified left MTG, left premotor cortex, SMA, and left IFG activation patterns as reliable markers for classifying RD across age groups and tasks. Future work should employ larger, longitudinal cohorts and test generalizability across languages and orthographies to refine early identification and intervention targeting phonological and speech-motor systems.

The sample size was relatively small, and the cross-sectional design limits inference about developmental trajectories. Adult RD groups had lower Raven scores than controls, though covariate control and matched subgroup analyses mitigated this concern. Generalizability to other languages and populations remains to be established; replication in larger, diverse samples and across language pairs is needed.