Top-Down RST Parsing Utilizing Granularity Levels in Documents
Authors: Naoki Kobayashi, Tsutomu Hirao, Hidetaka Kamigaito, Manabu Okumura, Masaaki Nagata8099-8106
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The results on the RST-DT corpus show that our method achieved the state-of-the-art results, 87.0 unlabeled span score, 74.6 nuclearity labeled span score, and the comparable result with the state-of-the-art, 60.0 relation labeled span score. Furthermore, discourse dependency trees converted from our RST trees also achieved the state-of-the-art results, 64.9 unlabeled attachment score and 48.5 labeled attachment score. |
| Researcher Affiliation | Collaboration | 1Institute of Innovative Research, Tokyo Institute of Technology, 2NTT Communication Science Laboratories, NTT Corporation |
| Pseudocode | Yes | Algorithm 1 Top-down parsing |
| Open Source Code | No | The paper does not provide an explicit statement about the availability of its source code for the described methodology. |
| Open Datasets | Yes | We evaluated our method by using the standard benchmark dataset RST-DT (Carlson, Marcu, and Okurowski 2001). |
| Dataset Splits | Yes | RST-DT was officially divided into 347 documents as the training dataset and 38 documents as the test dataset, which indicates that there is no development dataset available. Thus, we used 40 documents in the training dataset as the development data by following the study by Heilman and Sagae (2015). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions the use of Adam optimizer but does not specify version numbers for any software dependencies like programming languages, libraries, or frameworks. |
| Experiment Setup | Yes | The dimension size of the hidden layers was set to 250, dropout layers were incorporated with the ratio 0.4 at the training step, and the maximum training epoch was set to 50 for each model. All weight parameters were updated using Adam (Kingma and Ba 2014) with an initial learning rate being 0.001. The learning rate was decayed for each epoch with the ratio of 0.99. The gradientclipping threshold was set to 5.0 and the weight-decay was set to 1e 4. |