RNN-Based Sequence-Preserved Attention for Dependency Parsing
Authors: Yi Zhou, Junying Zhou, Lu Liu, Jiangtao Feng, Haoyuan Peng, Xiaoqing Zheng
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluated the graph-based and transition-based parsing models enhanced with the RNN-based sequence-preserved attention on the both English PTB and Chinese CTB datasets. The experimental results show that the enhanced systems were improved with significant increase in parsing accuracy. |
| Researcher Affiliation | Academia | Yi Zhou, Junying Zhou, Lu Liu, Jiangtao Feng, Haoyuan Peng, Xiaoqing Zheng School of Computer Science, Fudan University, Shanghai, China Shanghai Key Laboratory of Intelligent Information Processing {zhouyi13, junyingzhou14, l_liu15, fengjt16, hypeng15, zhengxq}@fudan.edu.cn |
| Pseudocode | No | The paper describes mathematical formulations and model architectures but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code is available at https://github.com/dugu9sword /spa-dp |
| Open Datasets | Yes | We show the results of the models on the English Penn Tree Bank (PTB) and the Chinese Penn Tree Bank (CTB). |
| Dataset Splits | Yes | We follow the standard split of PTB, with the section 2-21 for training, section 22 for development and section 23 for testing. |
| Hardware Specification | Yes | All experiments were run on a computer equipped with an Intel Core i7 processor, an 8GB RAM and a NVIDIA GTX-1070 GPU. |
| Software Dependencies | No | The paper states 'The model is implemented with the Py Torch2 deep learning framework.' but does not provide a specific version number for PyTorch or any other software dependency. |
| Experiment Setup | Yes | Hyper parameters are fine tuned on the PTB 3.3.0 development set. For graph-based parser, the hidden size of the bi-RNN is set to 400, and the hidden size of RNN for sequence-preserved attention is set to 100; for transition-based parser, the hidden size of both the bi-RNN and the RNN for sequence-preserved attention is set to 700. The dimension of word vectors is 300 and the dimension of tag vectors is 50. We train the model with word embedding and tag embedding drop out set to 30% respectively. The model is trained with a batch size of 32 by an Adam optimizer(Kingma and Ba 2014). |