Abstracts

Rationale:
Cognitive impairments are common among people with epilepsy, and one of the most frequent symptoms is word-finding difficulty. Such symptoms can involve memory and/or semantic processing dysfunction among other causes, and they are particularly notable in temporal lobe epilepsy in which memory and/or language circuits may be affected. The established auditory naming paradigm (Hamberger & Seidel, 2003) invokes elements of language-based semantic memory, though the variable syntax and position of disambiguating (critical) words in sentence stimuli precludes the precise control of distinct semantic processing periods and linguistic memory access. We sought to tailor and validate a newly modified version of this task that improves precision timing of semantic memory windows for in vivo human neurophysiology analysis.
Method:
We adapted the classic auditory naming task into a modified version (m-AN) principally by 1) standardizing stimuli syntax (noun-verb-noun order including an initial category noun) while maintaining natural sentences, and 2) designing the 102 unique stimuli to prompt single word answers that can be determined unambiguously only after the final (disambiguating) noun. The m-AN task thus provides step-wise semantic processing and was run on patients at our center for analysis of parallel neural activity along with behavioral reaction time (RT) as a difficulty index. Since RT also has attentional and other influences beyond language processing, we also presented the stimuli to Amazon Mechanical Turk workers (MTurk; USA-based worker masters, HIT >95%) and quantified the diversity of answers as a metric for semantic ambiguity.
Results:
Eleven patients undergoing ICEEG monitoring performed 50 to 350 trials each, with individual accuracy (omissions considered as errors) ranging between 78-100% correct. Median RT between the final (disambiguating) noun and speech was 1.12 seconds across patients (individual range: 0.54 to 2.01). The semantic ambiguity of each AN sentence stimulus was indexed as the number of unique responses from 100 MTurk workers, ranging from 1 to 26 per stimulus (median=4). Semantic ambiguity correlated with RT (r=0.497, p< 0.001, spearman, unadjusted). We also observed semantic-category-specific activation among subsets of electrodes (e.g. in middle temporal gyrus) which arose most prominently between stimulus offset and the onset of speech responses.
Conclusion:
The m-AN paradigm elicits and isolates semantic memory processing in a tightly controlled yet ecologically valid manner using natural language. As with classic auditory naming, this paradigm is easily implemented in ICEEG monitoring and operating rooms, utilizing only acoustic stimuli (natural sentence prompts) and speech-based answers. As evidence of semantic processing we show behaviorally that RT was partially predicted by semantic ambiguity, and physiologically that category-specific neural activity was increased in certain electrodes just after the final disambiguating word. Our task may timelock the crucial integration of linguistic and mnemonic functions more effectively than previous approaches, and may aid future investigations in semantic memory neurophysiology.
Funding:
:Dr. Kleen was funded by NIH / NINDS grants R25NS070680 and K23NS110920, Dr. Leonard was funded by NIH Grant R01-DC015504 and DARPA contract N66001-17-2-4008. Dr. Chang was funded by NIH Grants (R01-DC012379, R00-NS065120, and DP2-OD00862).