Partially-Aligned Data-to-Text Generation with Distant Supervision

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 9183–9193, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 9183 Partially-Aligned Data-to-Text Generation with Distant Supervision Zihao Fu1, Bei Shi2, Wai Lam1, Lidong Bing3, Zhiyuan Liu4 1Department of Systems Engineering and Engineering Management, The Chinese University of Hong Kong 2Tencent AI Lab 3DAMO Academy, Alibaba Group 4State Key Lab on Intelligent Technology and Systems Department of Computer Science and Technology Institute for Artificial Intelligence, Tsinghua Univerisity zhfu@se.cuhk.edu.hk; beishi@tencent.com; wlam@se.cuhk.edu.hk; l.bing@alibaba-inc.com; liuzy@tsinghua.edu.cn Abstract The Data-to-Text task aims to generate human- readable text for describing some given struc- tured data enabling more interpretability. How- ever, the typical generation task is confined to a few particular domains since it requires well- aligned data which is difficult and expensive to obtain. Using partially-aligned data is an alternative way of solving the dataset scarcity problem. This kind of data is much easier to obtain since it can be produced automatically. However, using this kind of data induces the over-generation problem posing difficulties for existing models, which tends to add unrelated excerpts during the generation procedure. In order to effectively utilize automatically an- notated partially-aligned datasets, we extend the traditional generation task to a refined task called Partially-Aligned Data-to-Text Genera- tion (PADTG) which is more practical since it utilizes automatically annotated data for train- ing and thus considerably expands the appli- cation domains. To tackle this new task, we propose a novel distant supervision generation framework. It firstly estimates the input data’s supportiveness for each target word with an estimator and then applies a supportiveness adaptor and a rebalanced beam search to har- ness the over-generation problem in the train- ing and generation phases respectively. We also contribute a partially-aligned dataset 1 by sampling sentences from Wikipedia and auto- matically extracting corresponding KB triples for each sentence from Wikidata. The experi- mental results show that our framework outper- forms all baseline models as well as verify the feasibility of utilizing partially-aligned data. 1 Introduction The Data-to-Text generation task focuses on gener- ating human-readable text corresponding to some 1The data and source code of this paper can be obtained from https://github.com/fuzihaofzh/ distant_supervision_nlg <Company of Heroes, genre, strategy video game> Company of Heroes is a strategy video game developed in Canada. <Company of Heroes, genre, strategy video game> Company of Heroes is a strategy video game developed in Canada. Company of Heroes, genre, strategy video game ⟨Company of Heroes is a strategy video game developed in Canada. ⟩ Training Samples Model ⟨Age of Empires, genre, strategy video game ⟩ Age of Empires is a strategy video game developed in Canada. Train } Figure 1: Illustration of the over-generation problem in the partially-aligned data-to-text generation task. In the training set, there is no KB triple corresponding to the text “developed in Canada”. The model is likely to bind the text to existing triples incorrectly. As a re- sult, during the testing or operational stage, the model is likely to overly generate this kind of excerpt for …

Content truncated to 500 words