Patterns of English words and their meanings have been mapped into a comprehensive ‘brain dictionary’ model of the semantic system using data-driven functional neuroimaging


Neuroimaging studies have identified a semantic system, or group of regions in the brain that represents information about the meaning of language. Within the semantic system there are domains which are selective for concrete or abstract words, action verbs, and social narratives, as well as related concepts such as living things, tools, and food. However, most imaging studies on the neurobiology of language have not comprehensively examined how information is represented in this system. In an influential report in Nature, a group of researchers have produced a ‘brain dictionary’ model of the semantic system using data-driven neuroimaging techniques.

Seven subjects listened to ten, 10 to 15 minute-long narrative stories that covered a range of topics from The Moth Radio Hour, a live storytelling radio program. During each story, their whole-brain blood-oxygen-level-dependent (BOLD) responses were measured using fMRI. Following the task, the researchers used a voxel-based modelling method to estimate the co-occurrence and relationship of 985 common English words to each voxel, or each unit of a 3D cortical brain space generated from BOLD response analysis of each individual subject. Using this functional mapping technique, the researchers were able to map semantic representation across the cortex.

Their results suggest the existence of a four-dimensional semantic space with 12 categories reflecting frequently co-occurring words and meanings.  A few examples of these categories are ‘numeric’ and ‘visual’. The ‘numeric’ category contained words or meanings associated with numbers, such as the word ‘four’, while the ‘visual’ category contained words such as ‘yellow’. Each of the 12 categories, were visualized in a semantic space model with RGB colour. It was discovered that words in categories related to human and social interaction (social, emotional, violent and communal) were grouped together, while perceptual, quantitative, and setting categories (tactile, locational, numeric, and visual) were also grouped together.

A generalized semantic brain map across subjects was next created that combined word co-occurrence and relationship results with a novel algorithm developed by the researchers. To their surprise, semantic areas were observed to be symmetrical between both cerebral hemispheres, rather than isolated to the left hemisphere, which has been long associated with language from lesion studies. This result suggests that the right hemisphere responds more strongly to narratives, though further research on hemispheric lateralization is needed.

Another interesting discovery was that semantically-selective brain areas were highly consistent across subjects. The lateral prefrontal cortex (LPFC) and medial prefrontal cortex (MPFC) were found to be selective for social concepts, while areas surrounding them were selective for visual, tactile, and numeric concepts. These regions have in the past been associated with introspection and conscious thought, and it is possible that these regions identified in this study represent the same domains active during conscious or internal thought. In contrast, the left temporal cortex (LTC), a brain region which is thought to have a key role in language, had fewer semantically-selective areas than the LPFC, MPFC, and surrounding regions. The researchers note that the quality of fMRI signals recorded in the LTC was poor, and likely contains semantic areas which were not visible with their approach.

The consistency of semantic areas across subjects suggests an inherent brain architecture that is responsible for organizing high-level semantics into functional regions. The researchers also note that this consistent semantic mapping may also be due to common life experiences of the subjects, all of whom were from Western backgrounds. More differences may be apparent if the study included subjects with different experiences and backgrounds to determine if the organizational structure remains the same.

Overall, the ‘brain dictionary’ map presented in this report is highly useful for the study of the neurobiology of language and is a model for future research to build upon.


What to see the brian map for yourself? Follow this link for the Interactive Semantic Atlas:




Written By: Fiona Wong, PhD

Facebook Comments