Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Show Me How To Revise: Improving Lexically Constrained Sentence Generation with XLNet
Authors: Xingwei He, Victor O.K. Li12989-12997
AAAI 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results have demonstrated that our proposed model performs much better than the previous work in terms of sentence fluency and diversity. |
| Researcher Affiliation | Academia | Xingwei He, Victor O.K. Li Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 Constrained Sentence Generation with XLNet |
| Open Source Code | Yes | Our code, pre-trained models and Appendix are available at https://github.com/NLPCode/MCMCXLNet. |
| Open Datasets | Yes | We used One-Billion-Word corpus1 to construct the synthetic dataset. 1http://www.statmt.org/lm-benchmark/ |
| Dataset Splits | Yes | We selected 6M, 0.3M and 1K sentences from One-Billion-Word corpus as the training, validation, and test sets, respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running experiments. |
| Software Dependencies | No | The paper mentions 'Hugging Face' and 'GPT-2 small (117M)' but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | Yes | The experiment setups for language models and the classifier are shown in the Appendix. We set N to 20. We set K to 50. All MCMC-based models were run for 200 steps. We ran 20 iterations for Bayesian MCMC, TIGS, L-MCMC, and L-MCMC-C. |