Proceedings of the Annual Meeting of the Cognitive Science Society

In the present paper, we introduce a novel computationalapproach for uncovering mental representations underlyinghealthiness judgments for food items. Using semantic vec-tor representations derived from large-scale natural languagedata, we quantify the complex representations that people holdabout foods, and use these representations to predict how bothlay decision makers and experts (trained dietitians) judge thehealthiness of food items. We also successfully predict theimpact of behavioral interventions (e.g. the provision of nutri-ent content information or “traffic-light labels”) on healthinessjudgments for food items. Our models are highly general, andare capable of making predictions for nearly any food item.Finally, these models outperform competing models based onfactual nutritional content, suggesting that health judgmentsdepend more on complex (semantic) knowledge representa-tions than on quantified nutritional information. The results inthis paper illustrate how methods from cognitive science andcomputational linguistics can be combined with existing theo-ries in psychology, to better predict, understand, and influencehealth behavior.