Modeling Source Syntax and Semantics for Neural AMR Parsing
Authors: DongLai Ge, Junhui Li, Muhua Zhu, Shoushan Li
IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on an English benchmark dataset show that our two approaches achieve significant improvement of 3.1% and 3.4% F1 scores over a strong seq2seq baseline. |
| Researcher Affiliation | Collaboration | Donglai Ge1 , Junhui Li1 , Muhua Zhu2 and Shoushan Li1 1School of Computer Science and Technology, Soochow University, Suzhou, China 2Alibaba Group, Hangzhou, China |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper refers to 'https://github.com/Rik VN/AMR' for pre-processing and post-processing scripts provided by [van Noord and Bos, 2017], but does not state that the source code for their own proposed methodology is open-source or available. |
| Open Datasets | Yes | For evaluation of our approach, we use the sentences annotated with AMRs from the LDC release LDC2017T10. |
| Dataset Splits | Yes | The dataset consists of 36,521 training AMRs, 1,368 development AMRs and 1,371 testing AMRs. |
| Hardware Specification | Yes | In all experiments, we train the models for 250K steps on a single K40 GPU. |
| Software Dependencies | No | The paper mentions 'Allen NLP' and 'tensor2tensor' as tools used, but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | In parameter setting, we set the number of layers in both the encoder and decoder to 6. For optimization we use Adam with β1 = 0.1 [Kingma and Ba, 2015]. The number of heads is set to 8. In addition, we set the hidden size to 512 and the batch token-size to 4096. In beam searching, we increase the extra length as 100 from default 50. We also set Google NMT length penalty parameter α = 1.0 to encourage longer generation. In all experiments, we train the models for 250K steps on a single K40 GPU. |