Does more context change how the brain represents word meaning?

Details

Word meaning in natural language depends heavily on context, but most neuroimaging studies of meaning use isolated words or sentences with little context. Because the brain may process natural language differently from simplified stimuli, the researchers examined whether findings from low-context studies generalize to natural language.

They used functional magnetic resonance imaging (fMRI) to record brain activity in four people, including two women, as they read words in four conditions: narratives, isolated sentences, blocks of semantically similar words, and isolated words. They compared the signal-to-noise ratio (SNR) of the evoked brain responses and used voxelwise encoding modeling to compare semantic information representations across conditions.

The researchers found four consistent effects of context. Stimuli with more context evoked responses with higher SNR than low-context stimuli across visual, temporal, parietal, and prefrontal cortices on both sides of the brain. At the group level, more context increased the representation of semantic information across bilateral temporal, parietal, and prefrontal cortices. In individual participants, only natural language stimuli consistently evoked widespread semantic representations. Context also affected voxel semantic tuning, and models estimated using low-context stimuli did not generalize well to natural language.

The findings indicate that context substantially affects neuroimaging data quality and where and how meaning is represented in the brain. They suggest that findings from studies using stimuli with little context may not generalize well to natural language used in daily life.