Abstracts

Rationale: Accessing stored long-term fact-based knowledge and associations can be affected in a variety of neurological disorders (temporal lobe epilepsy, Alzheimer’s dementia, semantic dementia, and traumatic brain injury) due to disruption of semantic networks, which have been commonly investigated using non-invasive methods (fMRI, MEG). Intracranial neurophysiology could more directly evaluate the neural basis of semantic processing, though there are few existing behavioral paradigms serving this purpose. We recently adapted the classic auditory naming task into an adapted version (AAN) to elicit and time-lock neural activity from intracranial recordings related to semantic memory processing. Here we describe an index for AAN that quantifies semantic processing load can be used to assist intracranial neurophysiology studies of semantic memory.

Methods: The AAN features standardized syntax (Noun1 -- > Verb -- > Noun2) while maintaining a natural sentence structure, allowing accurate time-locking of neurophysiological data across progressive steps of disambiguation. We extracted successive phrases from the AAN task to provide stepwise indices of ambiguity for each stimulus: N1 -- > N1V -- > N1VN2. For example: What is the person? -- > What is the person that fixes? -- > What is the person that fixes teeth? We delivered these stimuli to 100 Amazon Mechanical Turk workers (U.S.-based worker masters, HIT > 95%) for each successive step to obtain crowdsourced insights into the stepwise ambiguity of each prompt.

Results: The number of unique answers ranged from 7 to 68 for N1 phrases, 3 to 60 for N1V, and 1 to 26 for N1VN2. This provided a quantifiable metric we termed the semantic ambiguity index (SAI). SAI and reaction time correlated well (r = 0.526, p < 0.001, Spearman). SAI for the N1VN2 step did not correlate with a lexical property (word frequency) of the most common response (p > 0.05, Spearman). Comparing trial SAIs between successive steps of AAN stimuli (Figure 1), SAI significantly decreased (semantic narrowing) between N1 --> N1V and especially between N1V --> N1VN2 (p < 0.05 and p < 0.001, paired t-test) or when looking across all steps (N1 --> N1VN2; p < 0.001). Neural high gamma (HG) signals during the sequential steps correlated with the sequential SAI values for certain electrode sites across patients. Preliminary results revealed positive correlations (> 95% CI of shuffled Spearman Rho distributions) were relatively more common in the left middle temporal gyrus while inverse correlations were more common in the left superior temporal gyrus and left lateral frontal lobe.

Conclusions: SAI provides insight into the neural processing of ambiguity during retrieval of long-term factual knowledge, beyond simple behavioral metrics such as reaction time. The lack of correlation with word frequency suggests SAI is more relevant to semantic processing than lexical. Given the correlations with behavior and preliminary correlations with HG activity in intracranial recordings, the SAI is a promising alternative for investigating the anatomy and timing associated with neural semantic processing in the brain.

Funding: NIH/NINDS grant K23NS110920