The Sensitivity of Language Models and Humans to Winograd Schema Perturbations

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 732–741 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 732 Line Graph Enhanced AMR-to-Text Generation with Mix-Order Graph Attention Networks Yanbin Zhao, Lu Chen, Zhi Chen, Ruisheng Cao, Su Zhu, Kai Yu∗ MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University SpeechLab, Department of Computer Science and Engineering Shanghai Jiao Tong University, Shanghai, China {zhaoyb, chenlusz, zhenchi713}@sjtu.edu.cn {211314, paul2204, kai.yu}@sjtu.edu.cn Abstract Efficient structure encoding for graphs with labeled edges is an important yet challeng- ing point in many graph-based models. This work focuses on AMR-to-text generation – A graph-to-sequence task aiming to recover nat- ural language from Abstract Meaning Repre- sentations (AMR). Existing graph-to-sequence approaches generally utilize graph neural net- works as their encoders, which have two limi- tations: 1) The message propagation process in AMR graphs is only guided by the first- order adjacency information. 2) The relation- ships between labeled edges are not fully con- sidered. In this work, we propose a novel graph encoding framework which can effec- tively explore the edge relations. We also adopt graph attention networks with higher- order neighborhood information to encode the rich structure in AMR graphs. Experi- ment results show that our approach obtains new state-of-the-art performance on English AMR benchmark datasets. The ablation anal- yses also demonstrate that both edge relations and higher-order information are beneficial to graph-to-sequence modeling. 1 Introduction Abstract Meaning Representation (Banarescu et al., 2013) is a sentence-level semantic representation formalized by a rooted directed graph, where nodes are concepts and edges are semantic relations. Since AMR is a highly structured meaning repre- sentation, it can promote many semantic related tasks such as machine translation (Song et al., 2019) and summarization (Liao et al., 2018). How- ever, the usage of AMR graphs can be challenging, since it is non-trivial to completely capture the rich structural information in the graph-based data, es- pecially when the graph has labeled edges. ∗Kai Yu is the corresponding author. run-02 he fast-02 equal wind degree compare ARG1-0f ARG0 run-02 he fast-02 equal wind ARG1-of ARG0 degree compare original graph concept graph line (relation) graph Figure 1: An AMR graph (left) for sentence ”He runs as fast as the wind.” and its concept graph and relation graph (line graph). Two graphs are aligned with each other based on the node-edge relations in the original graph. Generation from AMR aims to translate the AMR semantics into the surface form (natural lan- guage). It is a basic Graph-to-sequence task that directly takes AMR as input. Figure 1 (left) gives a standard AMR graph and its corresponding surface form. Early works utilize sequence-to-sequence framework by linearizing the entire graph (Konstas et al., 2017; Cao and Clark, 2019). Such repre- sentation may lose useful structural information. In recent studies, graph neural networks (GNNs) have been in a dominant position on this task and achieved state-of-the-art performance (Beck et al., 2018; Song et al., 2018; Guo et al., 2019; Damonte …

Content truncated to 500 words