Incorporating Graph Information in Transformer-based AMR Parsing
Abstract Meaning Representation (AMR) is a Semantic Parsing formalism that aims at providing a semantic graph abstraction representing a given text. Current approaches are based on autoregressive language models such as BART or T5, fine-tuned through Teacher Forcing to obtain a linearized version of the AMR graph from a sentence. In this paper, we present LeakDistill, a model and method that explores a modification to the Transformer architecture, using structural adapters to explicitly incorporate graph information into the learned representations and improve AMR parsing performance. Our experiments show how, by employing word-to-node alignment to embed graph structural information into the encoder at training time, we can obtain state-of-the-art AMR parsing through self-knowledge distillation, even without the use of additional data. We release the code at \url{http://www.github.com/sapienzanlp/LeakDistill}.
PDF AbstractDatasets
Results from the Paper
Ranked #3 on AMR Parsing on LDC2020T02 (using extra training data)
Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Uses Extra Training Data |
Benchmark |
---|---|---|---|---|---|---|---|
AMR Parsing | LDC2017T10 | LeakDistill (base) | Smatch | 84.7 | # 9 | ||
AMR Parsing | LDC2017T10 | LeakDistill | Smatch | 86.1 | # 3 | ||
AMR Parsing | LDC2020T02 | LeakDistill (base) | Smatch | 83.5 | # 9 | ||
AMR Parsing | LDC2020T02 | LeakDistill | Smatch | 84.6 | # 3 |