Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reassessing Graph Linearization for Sequence-to-sequence AMR Parsing: On the Advantages and Limitations of Triple-Based Encoding

Jeongwoo Kang, Maximin Coavoux, Cédric Lopez, Didier Schwab

TL;DR

The paper reassesses graph linearization for sequence-to-sequence AMR parsing, comparing the conventional Penman encoding with a proposed triple-based encoding. It finds that Penman’s concise, nested representation generally yields higher parsing accuracy, while triple-based encoding, despite aligning more closely with AMR’s graph structure, does not surpass Penman and is more vulnerable to longer graphs. Multi-task learning with mixed encodings offers limited or mixed benefits. The work highlights the practical impact of representation choice on learning dynamics and opens avenues for improving graph-based outputs in AMR parsing.

Abstract

Sequence-to-sequence models are widely used to train Abstract Meaning Representation (Banarescu et al., 2013, AMR) parsers. To train such models, AMR graphs have to be linearized into a one-line text format. While Penman encoding is typically used for this purpose, we argue that it has limitations: (1) for deep graphs, some closely related nodes are located far apart in the linearized text (2) Penman's tree-based encoding necessitates inverse roles to handle node re-entrancy, doubling the number of relation types to predict. To address these issues, we propose a triple-based linearization method and compare its efficiency with Penman linearization. Although triples are well suited to represent a graph, our results suggest room for improvement in triple encoding to better compete with Penman's concise and explicit representation of a nested graph structure.

Reassessing Graph Linearization for Sequence-to-sequence AMR Parsing: On the Advantages and Limitations of Triple-Based Encoding

TL;DR

The paper reassesses graph linearization for sequence-to-sequence AMR parsing, comparing the conventional Penman encoding with a proposed triple-based encoding. It finds that Penman’s concise, nested representation generally yields higher parsing accuracy, while triple-based encoding, despite aligning more closely with AMR’s graph structure, does not surpass Penman and is more vulnerable to longer graphs. Multi-task learning with mixed encodings offers limited or mixed benefits. The work highlights the practical impact of representation choice on learning dynamics and opens avenues for improving graph-based outputs in AMR parsing.

Abstract

Sequence-to-sequence models are widely used to train Abstract Meaning Representation (Banarescu et al., 2013, AMR) parsers. To train such models, AMR graphs have to be linearized into a one-line text format. While Penman encoding is typically used for this purpose, we argue that it has limitations: (1) for deep graphs, some closely related nodes are located far apart in the linearized text (2) Penman's tree-based encoding necessitates inverse roles to handle node re-entrancy, doubling the number of relation types to predict. To address these issues, we propose a triple-based linearization method and compare its efficiency with Penman linearization. Although triples are well suited to represent a graph, our results suggest room for improvement in triple encoding to better compete with Penman's concise and explicit representation of a nested graph structure.
Paper Structure (13 sections, 5 figures, 3 tables)

This paper contains 13 sections, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology: Triple Linearization
  4. Triple linearization strategies
  5. Experimental Setup
  6. Baseline
  7. Multi-task Learning
  8. Results and Insights
  9. Global results.
  10. Triple linearization does not improve learning for deeper graphs.
  11. Triple linearization does not improve learning for longer graphs.
  12. Conclusion
  13. Appendix

Figures (5)

  • Figure 1: AMR graph for "We never seem to see any of the dug-in nutters acknowledge the truth in it." Example from the AMR 3.0 dataset knight-etal-2020-ldc2020.
  • Figure 4: Triple linearization of the graph in Figure \ref{['fig:amr_penman_distance']}.
  • Figure 5: smatch score per graph depth.
  • Figure 6: smatch score per graph length. The length is measured by the number of tokens in a linearized graph and token counts are grouped into buckets of 50 for clarity.
  • Figure 7: AMR graph for "There are too many traitors of China!". This example is drawn from AMR 3.0 dataset knight-etal-2020-ldc2020.