Dialogue Rewriting via Skeleton-Guided Generation
Authors: Chunlei Xin, Hongyu Lin, Shan Wu, Xianpei Han, Bo Chen, Wen Dai, Shuai Chen, Bin Wang, Le Sun
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that Real Dia is a much more challenging benchmark for real-world dialogue rewriting, and SGR can effectively resolve the task and outperform previous approaches by a large margin. |
| Researcher Affiliation | Collaboration | 1Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences, Beijing, China 2State Key Laboratory of Computer Science, Institute of Software, Chinese Academy of Sciences, Beijing, China 3University of Chinese Academy of Sciences, Beijing, China 4Xiaomi AI Lab, Xiaomi Inc., Beijing, China 5School of Information Engineering, Minzu University of China, Beijing, China 6National Language Resources Monitoring and Research Center for Minority Languages, Beijing, China |
| Pseudocode | No | The paper describes the architecture and steps of SGR in text and with Figure 2, but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states: "We build Skeleton-Guided Generator based on pre-trained Chinese T5-base1". The footnote 1 links to "https://github.com/ZhuiyiTechnology/t5-pegasus", which is a third-party T5 implementation, not the authors' own source code for SGR. There is no explicit statement or link indicating the release of the SGR's source code. |
| Open Datasets | No | The paper introduces a new dataset: "We construct Real-world Dialogue Rewriting Corpus (Real Dia) from a dialogue corpus provided by a large-scale Chinese Internet company". While it mentions data size and splits, it does not provide concrete access information (link, DOI, specific repository, or formal citation with access details) for this newly constructed dataset. |
| Dataset Splits | Yes | We then randomly sample 700/100/200 dialogues as train/dev/test sets. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for the experiments (e.g., GPU models, CPU types, or cloud computing instance specifications). |
| Software Dependencies | No | Our implementation is based on Py Torch (Paszke et al. 2019) and the Transformers library of Hugging Face (Wolf et al. 2020). We build Skeleton-Guided Generator based on pre-trained Chinese T5-base1, and construct Dialogue Skeleton Extractor and NLI model based on pre-trained Chinese Ro BERTa (Cui et al. 2021). The paper lists software but does not specify exact version numbers for PyTorch, Transformers, T5-base, or Chinese RoBERTa. |
| Experiment Setup | Yes | We optimized our model using label smoothing (Szegedy et al. 2016; M uller, Kornblith, and Hinton 2019) and Adam W (Loshchilov and Hutter 2019) with the learning rate=1e-4. In the experiments, each compared model is trained for 30 epochs with a batch size of 16. |