Learning compositional structures for semantic graph parsing
This reduces manual effort for semantic graph parsing, making it easier to apply to new datasets, though it is incremental as it builds on existing AM dependency parsing frameworks.
The paper tackles the problem of requiring expert-written heuristics for training AM dependency parsers by introducing a neural latent-variable model that learns compositional structures directly from graphs, achieving comparable accuracy to supervised methods.
AM dependency parsing is a method for neural semantic graph parsing that exploits the principle of compositionality. While AM dependency parsers have been shown to be fast and accurate across several graphbanks, they require explicit annotations of the compositional tree structures for training. In the past, these were obtained using complex graphbank-specific heuristics written by experts. Here we show how they can instead be trained directly on the graphs with a neural latent-variable model, drastically reducing the amount and complexity of manual heuristics. We demonstrate that our model picks up on several linguistic phenomena on its own and achieves comparable accuracy to supervised training, greatly facilitating the use of AM dependency parsing for new sembanks.