Semantic content allows flexible memory-partitioning in hybrid search

In ‘hybrid search,’ people look for one of several memorized targets among irrelevant distractors. Sometimes, only a subset of these targets is relevant to current task demands. Can we flexibly partition our memory into different target subsets and search only for relevant targets? Boettcher et al. (2017) found that participants fail to partition memory into two arbitrary subsets on a trial-by-trial basis: reaction times resembled search for the full target set. These findings were replicated even when using natural categories, perhaps because the categories suffered from large semantic overlap. Here, we tested more semantically distinct object categories. In Experiment 1, the target subsets were additionally dissociated from each other by colored overlays. Results confirmed that extrinsic characteristics like color were not effective for memory partitioning, but participants could limit search to the relevant category (e.g., search only for animals, in blocks containing animal and vehicle targets). However, searching through categorical subsets of 8 out of 16 items was still slower than searching with only 8 items in memory. In Experiment 2, spatial location was used as a retrieval cue to boost memory partition. For example, animal targets might be learned and searched for on the right, vehicles on the left. Partition was fully effective: search for the 8 relevant targets on a trial was the same as search with just 8 items in memory. In Experiments 1 and 2, the distractors could cue the relevant memory subset on each trial. Experiment 3 eliminated that cue, and included task-irrelevant ‘lures’ from the other category in memory. Results still showed successful partition on the basis of semantic category. Thus, people can effectively switch between a memory set used while searching for cookies and chips in the snack aisle, and a set of vegetables that they would search for in the produce aisle.