# A Differentiable Relaxation of Graph Segmentation and Alignment for AMR Parsing

arXiv: https://arxiv.org/abs/2010.12676 · HTML: https://mlsanity.com/paper/2010.12676/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 23 Oct 2020.

## Built on by (most cited first, 7 of 7)

- [Graph Pre-training for AMR Parsing and Generation](https://mlsanity.com/paper/2203.07836/) (2022)
- [Maximum Bayes Smatch Ensemble Distillation for AMR Parsing](https://mlsanity.com/paper/2112.07790/) (2021)
- [AMR Parsing with Action-Pointer Transformer](https://mlsanity.com/paper/2104.14674/) (2021)
- [Structure-aware Fine-tuning of Sequence-to-sequence Transformers for Transition-based AMR Parsing](https://mlsanity.com/paper/2110.15534/) (2021)
- [Inducing and Using Alignments for Transition-based AMR Parsing](https://mlsanity.com/paper/2205.01464/) (2022)
- [Levi Graph AMR Parser using Heterogeneous Attention](https://mlsanity.com/paper/2107.04152/) (2021)
- [Evaluation of Finetuned LLMs in AMR Parsing](https://mlsanity.com/paper/2508.05028/) (2025)

## Abstract (arXiv)

Abstract Meaning Representations (AMR) are a broad-coverage semantic formalism which represents sentence meaning as a directed acyclic graph. To train most AMR parsers, one needs to segment the graph into subgraphs and align each such subgraph to a word in a sentence; this is normally done at preprocessing, relying on hand-crafted rules. In contrast, we treat both alignment and segmentation as latent variables in our model and induce them as part of end-to-end training. As marginalizing over the structured latent variables is infeasible, we use the variational autoencoding framework. To ensure end-to-end differentiable optimization, we introduce a differentiable relaxation of the segmentation and alignment problems. We observe that inducing segmentation yields substantial gains over using a `greedy' segmentation heuristic. The performance of our method also approaches that of a model that relies on the segmentation rules of citetlyu-titov-2018-amr, which were hand-crafted to handle individual AMR constructions.