Temporal Reasoning

Temporal reasoning asks systems to represent and infer event order, duration, change, persistence, and constraints over sequences or time-stamped data. Researchers use formal methods such as temporal logic, Datalog, answer-set programming, interval graphs, and constraint solvers, alongside neural methods for video, language, knowledge graphs, and conversational memory; hybrid systems often use symbolic checks to verify model outputs. Evaluation spans date puzzles, clinical timelines, temporal knowledge-graph question answering, long videos, event streams, and multi-session agents, testing ordering, multi-step constraints, grounding, and stale information. Current systems handle explicit or local cues better than iterative, long-horizon reasoning, and benchmark scores may reflect shortcuts such as timestamps or positional encodings rather than use of temporal evidence.