Solving arithmetic word problems is difficult, as it requires the integration of arithmetic processing and language comprehension. It is not yet fully understood how specific neural processes subserve the difficulty of word problem-solving. Therefore, the objective of this study was to investigate the neural correlates of word problem-solving in adults ( $N = 28$ ) while considering arithmetic and linguistic complexity. The sources of difficulty were orthogonally manipulated by arithmetic operation (addition/subtraction) and lexical consistency (consistent/inconsistent form). Functional near-infrared spectroscopy (fNIRS) was applied to assess brain activation during word problem solving. The results showed that solving word problems engages the fronto-parietal network underlying arithmetic processing extending to temporal regions. Crucially, when the language was inconsistent to the arithmetic operation, increased activation was found in left-hemispheric language processing areas associated with working memory and cognitive control (middle frontal gyrus and precentral gyrus), as well as in the perisylvian language areas associated with text comprehension (supramarginal gyrus and superior temporal gyrus). This reflects the increased demands for language comprehension to disentangle the arithmetic operation. Regarding the arithmetic operation, activation for subtraction was not increased but rather prolonged compared to addition, suggesting later recognition and execution of the more difficult arithmetic operation. Uncovering the network underlying word problem solving revealed that the sources of difficulty in word problems are not only attributed to arithmetic but also to language and cognitive control – with implications for math education.
