Graphine: A Dataset for Graph-aware Terminology Definition Generation

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3453–3463 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 3453 Graphine: A Dataset for Graph-aware Terminology Definition Generation Zequn Liu1, Shukai Wang1, Yiyang Gu1, Ruiyi Zhang1, Ming Zhang1∗, Sheng Wang2∗ 1Department of Computer Science, School of EECS, Peking University, Beijing, China 2Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, WA zequnliu,shukaiwang,yiyanggu,zhangruiyi,mzhang_cs@pku.edu.cn swang@cs.washington.edu Abstract Precisely defining the terminology is the first step in scientific communication. Develop- ing neural text generation models for defi- nition generation can circumvent the labor- intensity curation, further accelerating scien- tific discovery. Unfortunately, the lack of large-scale terminology definition dataset hin- ders the process toward definition generation. In this paper, we present a large-scale termi- nology definition dataset Graphine covering 2,010,648 terminology definition pairs, span- ning 227 biomedical subdisciplines. Termi- nologies in each subdiscipline further form a directed acyclic graph, opening up new av- enues for developing graph-aware text genera- tion models. We then proposed a novel graph- aware definition generation model Graphex that integrates transformer with graph neural network. Our model outperforms existing text generation models by exploiting the graph structure of terminologies. We further demon- strated how Graphine can be used to evaluate pretrained language models, compare graph representation learning methods and predict sentence granularity. We envision Graphine to be a unique resource for definition generation and many other NLP tasks in biomedicine.1 1 Introduction Obtaining the definition is the first step toward un- derstanding a new terminology. The lack of precise terminology definition poses great challenges in scientific communication and collaboration (Oke, 2006; Cimino et al., 1994), which further hinders new discovery. This problem becomes even more severe in emerging research topics (Baig, 2020; Baines et al., 2020), such as COVID-19, where curated definitions could be imprecise and do not scale to rapidly proposed terminologies. ∗Corresponding author 1Our Dataset is available at https://zenodo.org/ record/5320310#.YSxHgI77Q2w. Our code is avail- able at https://github.com/zequnl/Graphex Terminology: Enhancer of variegation Definition: Genotype g1 is an enhancer of variegation if, and only if, some genotype g2 has a variegated phenotype and the degree of variegation caused by g1, g2 is greater than that caused by g2 alone. Terminology: Modifier of variegation Definition: Phenotype that is reduction in or loss of a stereotypical behavioral response to touch. Terminology: Phenotype Definition: A defect in or loss of some anatomical structure or biological process compared to wild-type. 227 directed acyclic graphs, 2,010,648 terminology definition pairs Figure 1: Graphine dataset contains 2,010,648 ter- minology definition pairs organized in 227 directed acyclic graphs. Each node in the graph is associated with a terminology and its definition. Terminologies are organized from coarse-grained ones to fine-grained ones in each graph. Neural text generation (Bowman et al., 2016; Vaswani et al., 2017; Sutskever et al., 2014; Song et al., 2020b) could be a plausible solution to this problem by generating definition text based on the terminology text. Encouraging results by neural text generation have been observed on related tasks, such …

Content truncated to 500 words