A Differentiable Relaxation of Graph Segmentation and Alignment for AMR Parsing
This addresses the need for more automated and flexible AMR parsing, reducing reliance on manual preprocessing, though it is incremental as it builds on existing methods.
The paper tackles the problem of segmenting and aligning Abstract Meaning Representation (AMR) graphs to words in sentences, which is typically done with hand-crafted rules, by treating these as latent variables and inducing them through end-to-end training using a differentiable relaxation. The result shows substantial gains over a greedy heuristic and approaches the performance of models using hand-crafted rules.
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 \citet{lyu-titov-2018-amr}, which were hand-crafted to handle individual AMR constructions.