Thinking Clearly, Talking Fast: Concept-Guided Non-Autoregressive Generation for Open-Domain Dialogue Systems

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 175–184 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 175 Logic-level Evidence Retrieval and Graph-based Verification Network for Table-based Fact Verification Qi Shi, Yu Zhang∗, Qingyu Yin, Ting Liu Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, Harbin, China {qshi,zhangyu,qyyin,tliu}@ir.hit.edu.cn Abstract Table-based fact verification task aims to ver- ify whether the given statement is supported by the given semi-structured table. Sym- bolic reasoning with logical operations plays a crucial role in this task. Existing meth- ods leverage programs that contain rich log- ical information to enhance the verification process. However, due to the lack of fully supervised signals in the program generation process, spurious programs can be derived and employed, which leads to the inability of the model to catch helpful logical opera- tions. To address the aforementioned prob- lems, in this work, we formulate the table- based fact verification task as an evidence re- trieval and reasoning framework, proposing the Logic-level Evidence Retrieval and Graph- based Verification network (LERGV). Specifi- cally, we first retrieve logic-level program-like evidence from the given table and statement as supplementary evidence for the table. After that, we construct a logic-level graph to cap- ture the logical relations between entities and functions in the retrieved evidence, and design a graph-based verification network to perform logic-level graph-based reasoning based on the constructed graph to classify the final entail- ment relation. Experimental results on the large-scale benchmark TABFACT show the ef- fectiveness of the proposed approach1. 1 Introduction There are a large number of semi-structured ta- bles on the Internet. How to perform reasoning over semi-structured tables is crucial for people to understand different types of information in the real world. And this direction has spawned many tasks. Among these tasks, table-based fact verifi- cation task has recently received a lot of attention, which is important for many applications, such as ∗Corresponding author. 1Our code is available at: https://github.com/ qshi95/LERGV Chassis Manufacturer Chassis Model Number in Fleet Fleet Numbers man hocl-nl 20 2101 - 2120 mercedes - benz o405nh 2 2520 - 2521 mitsubishi fuso rosa 6 34, 2601 - 2603 scania k280ub 1 3230 scania k320ua 6 2831 - 2836 the smallest number in fleet for chassis manufacturer , scania , with a fleet number 3230 is 1 ENTAILED Table Label Statement l eq { 1 ; min { all_rows ; number in fleet } } = True l eq { 1 ; hop { filter_eq { all_rows ; fleet numbers ; 3230 } ; number in fleet } } = True l eq { scania ; hop { filter_eq { all_rows ; number in fleet ; 1 } ; chassis manufacturer } } = True l … Program-like Evidence Figure 1: Example of TABFACT dataset. Given a ta- ble and a statement, the goal is to verify the correctness of the statement by the table. Program-like evidence contains logical operations and can be used as supple- mentary information …

Content truncated to 500 words