Schedule of Events | Symposia

Concept Feature Diagnosticity: a new metric to quantify conceptual access

Poster Session E - Monday, March 31, 2025, 2:30 – 4:30 pm EDT, Back Bay Ballroom/Republic Ballroom
Also presenting in Data Blitz Session 1 - Saturday, March 29, 2025, 10:30 am – 12:00 pm EDT, Grand Ballroom.

Anna M. Keresztesy^1,2 (anna_keresztesy@urmc.rochester.edu), Muzi Li¹, Jessica M. Smith¹, Frank E. Garcea², Bradford Z. Mahon¹; ¹Carnegie Mellon University, ²University of Rochester

Semantic representations are commonly modeled in terms of elementary features (e.g. CAT = {<fur>, <eats mice>, <purrs>}). Concept-feature models have traditionally studied how features are generated from concepts. Here we define and study a new way to quantify concept-feature relationships, which we term diagnosticity. Diagnosticity refers to the proportion of target responses (e.g., ‘cat’) that are spontaneously generated when naive participants are given a feature in isolation (e.g., ‘eats mice’). Across four Experiments we demonstrate the validity and power of diagnosticity to account for variance in the speed with which healthy participants complete semantic verification tasks. In the core paradigm, participants were presented with a written item name and a semantic feature that was either true of the concept or not. Participants indicated whether the feature was associated with the item by pressing a button (Exp 1) or responding verbally (Exp 2). In Exp 3, a separate group of participants were presented with isolated features and were tasked with generating the corresponding concept name. We found that target production in the free-response task of Exp 3 predicted 70% of the variance in decision times in Exps 1 and 2: highly diagnostic features (those that elicited the target concept most consistently) were associated with shorter verification latencies in Exps 1 and 2. Exp 4 replicated the full pattern with a new and large set of items and new group of participants. These findings suggest that the dynamics of information access is from semantic features to concepts during semantic verification.

Topic Area: LANGUAGE: Semantic