Guiding Attention in Sequence-to-Sequence Models for Dialogue Act Prediction
Authors: Pierre Colombo, Emile Chapuis, Matteo Manica, Emmanuel Vignon, Giovanna Varni, Chloe Clavel7594-7601
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The proposed approach achieves an unmatched accuracy score of 85% on Sw DA, and state-of-the-art accuracy score of 91.6% on MRDA. |
| Researcher Affiliation | Collaboration | 1LTCI, Telecom Paris, Institut Polytechnique de Paris, 2IBM GBS France, 3IBM Zurich |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an unambiguous statement about releasing code or a link to a source code repository for the described methodology. |
| Open Datasets | Yes | Sw DA: The Switchboard-1 corpus is a telephone speech corpus (Stolcke et al. 1998) ... MRDA: The ICSI Meeting Recorder Dialogue Act corpus (Shriberg et al. 2004) |
| Dataset Splits | Yes | Train/Dev/Test Splits: For both Sw DA and RMDA we follow the official split introduced by Stolcke et al. (2000). Thus, our model can directly be compared to Li et al.; Chen et al.; Kumar et al.; Raheja and Tetreault (2018a; 2018; 2018; 2019). All the hyper-parameters have been optimised on the validation set using accuracy computed on the last tag of the sequence. |
| Hardware Specification | Yes | Models have been implemented in Py Torch and trained on a single NVIDIA P100. |
| Software Dependencies | No | The paper mentions "implemented in Py Torch" and specific optimizers (Adam, AdamW) but does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | Parameters for Sw DA: We used Adam optimizer (Kingma and Ba 2014) with a learning rate of 0.01, which is updated using a scheduler with a patience of 20 epochs and a decrease rate of 0.5. The gradient norm is clipped to 5.0, weight decay is set to 1e-5, and dropout (Le Cun, Bengio, and Hinton 2015) is set to 0.2. The maximum sequence length is set to 20. |