The importance of linguistic information in human reinforcement learning

How does the nature of a stimulus affect our ability to learn appropriate response associations? In typical laboratory experiments learning is investigated under somewhat ideal circumstances, where stimuli are easily discriminable visually and linguistically. This is not representative of most real-life learning, where visually or linguistically overlapping "stimuli" can result in different ``rewards'' (e.g., you may learn over time that you can pet one specific dog that is friendly, but that you should avoid a very similar looking one that isn't). With two experiments, we test how humans learn in three stimulus conditions: stimuli with distinct visual representations but overlapping linguistic representations, stimuli with distinct linguistic representations but overlapping visual representations, and stimuli with distinct visual and linguistic representations. We find that decreasing linguistic and visual distinctness both decrease performance, substantially more for the lowered linguistic distinctness condition. We develop computational models to test different hypotheses about how reinforcement learning (RL) and working memory (WM) processes are affected by different stimulus conditions. Interestingly, we find that only RL, and not WM, is affected by stimulus condition: people learn slower and have higher across-stimulus value confusion at decision when linguistic information overlaps relative to when it is distinct. These results demonstrate strong effects of stimulus type on learning, and highlight the importance of considering the parallel contributions of different cognitive processes when studying behavior.