Princeton University Library Catalog

Semantic and episodic prediction reveal long-axis specialization in the hippocampus

Avery, Emily [Browse]
Senior thesis
Turk-Browne, Nicholas B. [Browse]
Princeton Neuroscience Institute [Browse]
Class year:
Summary note:
To engage in goal-directed behavior, we must be able to predict the consequences of our actions. When we drop a glass or strike a match, for example, there is a specific outcome we know to expect. Though the generation of predictions based on such long-term semantic memories remains to be characterized, recent research suggests that predictions based on specific episodes of the past may be facilitated by pattern completion in the hippocampus (Hindy et al., 2016). Hypothesizing that long-term memory prediction continues to be supported by the hippocampus, we collected high-resolution fMRI data while participants selected among familiar actions that changed the state of one object into another. In a ‘novel’ condition representing shorter-term episodic memory, participants learned actions that changed objects in arbitrary ways, requiring predictions to be based on episodes experienced recently, three days before fMRI scan. In a ‘known’ condition representing long-term semantic knowledge, participants could predict the outcomes of selected actions by utilizing their pre-existing knowledge. In both conditions, responses in early visual cortex were attenuated when an outcome could be predicted from an action, and this predictive attenuation was related to background connectivity with the hippocampus. Further segmentation of the hippocampus itself provided evidence for functional distinction along the long axis. Namely, prediction of novel cue-action outcomes in visual cortex was related to the anterior hippocampus, and prediction of known cue-action outcomes was related to posterior hippocampus. These findings indicate that the hippocampus mediates visual prediction based not only on specific episodes, but on long-term semantic knowledge as well. The generation of predictions utilizing these two types of memory, however, may be facilitated by different mechanisms.