Word Alignment-Based Evaluation of Uniform Meaning Representations

2026-03-27

AI Summary

Existing evaluation metrics for Uniform Meaning Representation (UMR) graphs, such as smatch, struggle to align structures with differing numbers of nodes and fail to distinguish between genuine semantic mismatches and spurious matches, thereby limiting the interpretability of error analysis. This work proposes a node-matching algorithm grounded in word–node alignments, leveraging the inherent correspondence between surface tokens and UMR nodes to circumvent the NP-hard search problem inherent in conventional approaches. By comparing graph-based semantic representations using an F1 scoring scheme derived from these alignments, the method offers substantially improved intuitiveness and semantic interpretability over current evaluation practices. The authors release an open-source implementation to facilitate efficient and explainable comparison of UMR graphs.

Abstract

Comparison and evaluation of graph-based representations of sentence meaning is a challenge because competing representations of the same sentence may have different number of nodes, and it is not obvious which nodes should be compared to each other. Existing approaches favor node mapping that maximizes $F_1$ score over node relations and attributes, regardless whether the similarity is intentional or accidental; consequently, the identified mismatches in values of node attributes are not useful for any detailed error analysis. We propose a node-matching algorithm that allows comparison of multiple Uniform Meaning Representations (UMR) of one sentence and that takes advantage of node-word alignments, inherently available in UMR. We compare it with previously used approaches, in particular smatch (the de-facto standard in AMR evaluation), and argue that sensitivity to word alignment makes the comparison of meaning representations more intuitive and interpretable, while avoiding the NP-hard search problem inherent in smatch. A script implementing the method is freely available.

Problem

• Word Alignment
• Uniform Meaning Representation
• Graph-based Semantic Representation
• Node Matching
• Semantic Evaluation

Innovation

• word alignment
• Uniform Meaning Representation
• node matching
• semantic representation evaluation
• smatch alternative