Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Towards the neurocognitive mechanisms supporting Semantic Feature Generation.
Poster D20 in Poster Session D, Wednesday, October 25, 4:45 - 6:30 pm CEST, Espace Vieux-Port
Alexander M. Swiderski1,2,3, Jason W. Bohland1,2, Jeffrey P. Johnson3, Michael W. Dickey1,2,3, Stephen M. Wilson4, William D. Hula1,3; 1University of Pittsburgh, 2Carnegie Mellon University, 3VA Pittsburgh Healthcare System, 4University of Queensland
Introduction: Semantic Feature Analysis[1] (SFA) is one of the most commonly used treatments for word production deficits in aphasia[2]. The goal of SFA is to improve spoken word production by guiding people with aphasia to produce semantic features related to a target. Generation of semantic features is hypothesized to be a key active ingredient of SFA[1]. The mechanism of action of SFA is hypothesized to be spreading activation between amodal representations[3]. However, it remains unclear whether SFA’s mechanism of action could be better explained by more contemporary theories[4] suggesting that concepts are represented by multimodal co-activation of sensory, motor, affective, and temporal experiences that they are associated with. To begin to address this knowledge gap, we have used representational similarity analysis (RSA) to reveal the similarity structure between publicly-available covert feature generation BOLD data[5] and experiential-model feature vectors accounting for 48 sensory, motor, and affective experiences[6] (henceforth Exp48). Method: We reanalyzed BOLD data generated from seven right-handed adults (4 female, 3 male; age 19-32) who covertly generated semantic features for 60 concrete nouns[5]. Lexical-conceptual representations were operationalized with publicly available experiential feature vectors[6]. Only 30 of the 60 objects included in the fMRI analysis were also included in Exp48, so the analyses were conducted across 30 rather than 60 nouns. RSA with a searchlight approach were used to reveal the similarity structure between the neural data and the Exp48 feature-vectors. One-sample t-tests were performed at each voxel to evaluate whether there was a difference between the RSA-derived correlations across the seven participants compared to zero. Results: The participant-level RSA results indicated substantial subject-to-subject variability in the similarity structure between the neural data and Exp48 across the left and right hemispheres. One-sample t-tests of the RSA correlations across the seven subjects revealed robust similarity estimates within the left IFG, MFG, AG, MTG, STS, MTS and the right AG, STS and STG. Discussion: The group-level findings accord well with the semantic network proposed to support concept knowledge by the reactivation of sensory, motor, and affective modalities [7]. By the time of the conference, we expect to be able to present updated results evaluating the relationship between the neural data elicited by covert semantic feature generation and competing taxonomic, distributional, and experiential semantic models. The proposed analyses may shed light on which semantic model is most consistent with the mechanism of action supporting SFA, which may inform future studies aimed at optimizing this popular treatment. 1. Boyle, M. Topics in Stroke Rehabilitation 17, 411–422 (2010). 2. Tierney-Hendricks et al. American journal of speech-language pathology 1–30 (2021). 3. Collins et al. Psychological review 82, 407 (1975). 4. Barsalou et al. Trends Cogn Sci 7, 84–91 (2003). 5. Mitchell et al. Science 320, 1191–1195 (2008). 6. Binder et al. Cognitive Neuropsychology 33, 130–174 (2016). 7. Binder et al. Cerebral cortex 19, 2767–2796 (2009).
Topic Areas: Meaning: Lexical Semantics, Computational Approaches