Representing Events Using Fuzzy Temporal Boundaries

This study investigates whether people represent the beginnings and ends of events as fuzzy temporal frames and subsequently construct event temporal relations. The study adopted Allen’s (1984; 1991) seven logical categories of temporal relations. Constructing the seven relations often requires the accurate encoding and (or) retrieval of the beginnings and ends of events. We used a recurrent neural network to simulate the performance of humans in representing event temporal relations. The network was given fuzzy event inputs, generated using Zadeh (1975) fuzzy logic functions, and trained to judge the event temporal relations. We compared the performance of the recurrent neural network to that of humans in a task where they were asked to remember and reconstruct everyday temporal relations. The simulations showed that the recurrent neural network mimicked human judgments in the correct judgments, preferences toward particular temporal relations, and directions of error. The results support that event temporal relations are best thought of as fuzzy analogue representations in humans and the simulated network.