Resolving sentence ambiguity with strategic planning and working memory resources: Behavioral, imaging and functional connectivity analyses

We used behavioral, functional magnetic resonance imaging (fMRI) and functional connectivity MRI (fcMRI) analyses to test competing strategies for disambiguation of a sentence featuring a temporary structural ambiguity (TSA), where working memory (WM) demands were also manipulated. Written sentences with a direct object (DO) structure or a sentential complement (SC) structure were shown to 19 healthy, right-handed, young adults in a phraseby-phrase manner. These sentences contained a main verb that is statistically more likely to be associated with a DO structure or an SC structure. Half of each type of sentence also contained an extra phrase strategically located to stress WM prior to disambiguating the sentence. We found that sentences featuring a less consistent verb-structure mapping recruit greater dorsolateral prefrontal cortex (dlPFC) activation than sentences with a more consistent verb-structure mapping, implicating strategic on-line planning during resolution of a TSA. By comparison, we observed left inferior parietal cortex (IPC) activation in sentences with an increased WM demand compared to sentences with a low WM load. Our fcMRI analysis showed that, although the same set of brain regions are often activated during different sentence processing conditions, the correlations between these areas change depending on the specific sentence processing demands. Less ambiguous sentences do not elicit a positive correlation between inferior frontal cortex (IFC) and dlPFC, but the positive correlation between IFC and dlPFC during more ambiguous stimuli suggests that these are co-activated and collaborate in order to resolve the ambiguity. We also found that the correlations between left IPC and both left IFC and left dlPFC remain constant during the course of sentence presentation. However, the correlation between left IPC and left IFC increases only during processing of the last phrase of "high WM" stimuli. Taken together, our observations are compatible with a serial processing model, where a large-scale neural network for sentence processing recruits distinct strategic planning and WM processing resources as needed to support the comprehension of sentences. Our results underline the value of complementing functional neuroimaging analyses with functional connectivity studies, in order to improve our understanding of neurocognitive function.