Temporal Knowledge Graph Question Answering (TKGQA) commonly relies on retrieval-aggregation-reasoning pipelines, which suffer from two critical challenges: error propagation, where early retrieval mistakes accumulate along reasoning paths, and temporal conflicts caused by post-hoc temporal filtering, leaving temporally inconsistent paths in the search space and undermining multi-hop reasoning reliability. To address these limitations, we propose ConTExT (Contractive Temporal Exploration), a contractive subgraph search framework over structural and temporal spaces for TKGQA. It dynamically anchors question-relevant entities, performs selective subgraph exploration guided by path semantics and sub-question consistency, and applies Temporal Feasible Region Propagation (TFRP), which actively propagates temporal constraints during search and maintains a temporal feasible region. Experiments on MultiTQ show that ConTExT is particularly effective for complex temporal reasoning, yielding a 27.1% relative improvement in Hits@1 over the strongest baseline on Multiple questions. Its gains are smaller on Single and Time-answer questions that depend more on direct evidence matching or exact temporal endpoints.

We analyze GPT-4o’s ability to represent numeric information in texts for elementary school children and assess it with respect to the human baseline. We show that both humans and GPT-4o reduce the amount of numeric information when adapting informational texts for children but GPT-4o retains more complex numeric types than humans do.