Words-In-Sequence Memory Formed by Eye Movement Sequences During Reading: A Network Model Based on Theta Phase Coding

摘要

Revealing the neuronal mechanisms enabling the hippocampus to maintain episodic memory (i.e., memory for personal events) is a fundamental issue for our understanding of human intelligence. A temporal compression mechanism based on theta phase coding has been shown to be essential for the encoding of episodic memory occurring on a behavioral timescale (>a few seconds) and capable of operating sequential and spatial memory. However, episodic memory models should be developed to handle more semantic-rich contents. Here, memory from the reading of a passage of literature was evaluated using a network model based on theta phase coding. Two types of input were simultaneously applied: word inputs, defined by the eye movement sequence obtained during the reading, with each fixated word encoded by a vector computed from a statistical language model with a large text corpus; and sequence inputs, defined by random activations and representing a gradually changing input pattern, independent of the eye movement. The results successfully demonstrated a memory generated by the word sequence from the 6-min reading session, and the memory network for the sequence input was shown to be essential for the retrieval. This result was characterized by a words-in-sequence structure in the network formed by theta phase coding, of which, the unidirectional connections among sequence units dominantly evoked the sequential activation, which further conveyed word unit activations through the unidirectional connections from the sequence to the word units. This suggests a general role for theta phase coding in the formation of episodic memory.