Syntactically Look-Ahead Attention Network for Sentence Compression
Authors: Hidetaka Kamigaito, Manabu Okumura8050-8057
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The results of the automatic evaluation on the Google sentence compression dataset showed that SLAHAN achieved the best kept-token-based F1, ROUGE-1, ROUGE-2 and ROUGE-L scores of 85.5, 79.3, 71.3 and 79.1, respectively. SLAHAN also improved the summarization performance on longer sentences. Furthermore, in the human evaluation, SLAHAN improved informativeness without losing readability. |
| Researcher Affiliation | Academia | Hidetaka Kamigaito, Manabu Okumura Institute of Innovative Research, Tokyo Institute of Technology, kamigaito@lr.pi.titech.ac.jp, oku@pi.titech.ac.jp |
| Pseudocode | No | The paper describes its model and equations but does not include any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | Our code will be available at https://github.com/kamigaito/slahan. |
| Open Datasets | Yes | We used the Google sentence compression dataset (Google dataset) (Filippova and Altun 2013) for our evaluations. ... 1This sentence actually belongs to the test set of the Google sentence compression dataset (https://github.com/google-research-datasets/sentence-compression). |
| Dataset Splits | Yes | We used the first 1,000 sentences of comp-data.eval.json as the test set. We used the last 1,000 sentences of comp-data.eval.json as our development set. Following recent researches (Kamigaito et al. 2018; Zhao, Luo, and Aizawa 2018), we used all 200,000 sentences in sent-comp.train*.json as our training set. |
| Hardware Specification | No | The paper specifies model parameters and training configurations, but it does not provide any specific hardware details such as GPU/CPU models or memory specifications used for running experiments. |
| Software Dependencies | No | The paper mentions using 'Dynet (Neubig et al. 2017) to implement our neural networks' but does not provide a specific version number for Dynet or any other software libraries used. |
| Experiment Setup | Yes | The dimensions of the LSTM layer and the attention layer were set to 200. The depth of the LSTM layer was set to 2. All dropout rates were set to 0.3. We used Adam (Kingma and Ba 2014) with an initial learning rate of 0.001 as our optimizer. The clipping threshold value for the gradients was set to 5.0. The maximum training epoch was set to 20. We used {1, 2, 3, 4} as d in Eq.(16) and Eq.(18). The maximum minibatch size was set to 16. |