enriched_web_nlg

WebNLG is a valuable resource and benchmark for the Natural Language Generation (NLG) community. However, as other NLG benchmarks, it only consists of a collection of parallel raw representations and their corresponding textual realizations. This work aimed to provide intermediate representations of the data for the development and evaluation of popular tasks in the NLG pipeline architecture (Reiter and Dale, 2000), such as Discourse Ordering, Lexicalization, Aggregation and Referring Expression Generation.

WebNLG 是自然语言生成（NLG）领域一项宝贵的资源与基准。然而，与其他 NLG 基准类似，它仅包含一系列并行原始表征及其对应的文本实现。本研究旨在为数据提供中间表征，以促进 NLG 管道架构中流行任务（如 Reiter 和 Dale 于 2000 年所提出的会话排序、词汇化、聚合以及指称表达式生成）的开发与评估。