This work presents Exaqt : EXplainable Answering of complex Questions with Temporal intent, a system that does not rely on manual rules for question understanding and reasoning.（这项工作提出了 Exaqt：具有时间意图的复杂问题的可解释回答，这是一个不依赖手动规则来理解和推理问题的系统。）
Exaqt is the first end-to-end system for answering complex temporal questions that have multiple entities and predicates, and associated temporal conditions.（Exaqt是第一个用于回答具有多个实体和谓词以及相关时间条件的复杂时间问题的端到端系统。）
Exaqt answers natural language questions over knowledge graphs in two stages, one geared towards high recall, the other towards precision at top ranks. （Exaqt 分两个阶段回答关于知识图谱的自然语言问题，一个面向高召回率，另一个面向顶级精度。）
The first step computes question-relevant compact subgraphs within the knowledge graph that contain all cues required for answering the question, using Group Steiner Trees and fine-tuned BERT models， and judiciously enhances them with pertinent temporal facts.（第一步使用 Group Steiner Trees 和微调的 BERT 模型计算知识图中与问题相关的紧凑子图，其中包含回答问题所需的所有线索，并使用相关的时间事实明智地增强它们。）
The second step constructs relational graph convolutional networks (R-GCNs) from the first step’s output to infer the answer in the graph, and enhances the R-GCNs with time-aware entity embeddings and attention over temporal relations.（第二步从第一步的输出构建关系图卷积网络 (R-GCN)以推断图中的答案，并通过时间感知实体嵌入和对时间关系的关注来增强 R-GCN。）
实验评估
We evaluate Exaqt on TimeQuestions, a large dataset of 16k temporal questions we compiled from a variety of general purpose knowledge graphs (KG-QA) benchmarks. Results show that Exaqt outperforms three state-of-the-art systems (Uniqorn, GRAFT-Net and PullNet) for answering complex questions over KGs, thereby justifying specialized treatment of temporal QA.（我们在 TimeQuestions 上评估 Exaqt，这是一个包含 16k 时间问题的大型数据集，我们从各种通用 KG-QA 基准中编译。结果表明，Exaqt 在回答 KG 上的复杂问题方面优于三个最先进的系统，从而证明对时间 QA 进行专门处理是合理的。）
While using Group Steiner Trees (GST) as a building block in Exaqt, outperforming the Uniqorn method shows that non-terminals (internal nodes) in GSTs, by themselves, are not enough to answer temporal questions.（虽然在 Exaqt 中使用 Group Steiner Trees (GST) 作为构建块，但优于 Uniqorn 方法表明 GST 中的非终结符（内部节点）本身不足以回答时间问题。）
Improving over GRAFT-Netshows that augmenting R-GCNs with time information works well. Questions with implicit expressions are more challenging: we believe that this is where the power of R-GCNs truly shine, as GST-based Uniqorn clearly falls short.（对 GRAFT-Net 的改进表明，使用时间信息增强 R-GCN 效果很好。带有隐式表达式的问题更具挑战性：我们认为，这正是 R-GCN 真正发挥作用的地方，因为基于 GST 的 Uniqorn 显然不足。）
