Controllable Text Generation with Neurally-Decomposed Oracle

Authors: Tao Meng, Sidi Lu, Nanyun Peng, Kai-Wei Chang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments conducted on two tasks: (1) text generation with lexical constraints and (2) machine translation with formality control demonstrate that our framework efficiently guides the base model towards the given control factors while maintaining high generation quality.
Researcher Affiliation Academia Tao Meng University of California, Los Angeles tmeng@cs.ucla.edu Sidi Lu University of California, Los Angeles sidilu@cs.ucla.edu Nanyun Peng University of California, Los Angeles violetpeng@cs.ucla.edu Kai-Wei Chang University of California, Los Angeles kwchang@cs.ucla.edu
Pseudocode No The paper does not contain explicit pseudocode or algorithm blocks.
Open Source Code Yes Our code can be found at https://github.com/Mt Some Three/constrDecoding.
Open Datasets Yes For unsupervised LCG, we follow the settings in POINTER (Zhang et al., 2020) and conduct our experiments on Yelp! Review and News dataset. [...] For supervised LCG, we evaluate the proposed method on Common Gen (Lin et al., 2020). [...] We conduct our experiments on Fisher and CALLHOME Spanish-English Speech Translation Corpus (Post et al., 2013)
Dataset Splits Yes For training, it contains 32,651 unique key concepts (i.e. the constraints) with 67,389 completed sequences in total. It also contains a validation set with 993 concepts and 4018 reference sequences.
Hardware Specification No The provided text does not contain specific hardware specifications (GPU/CPU models, memory, etc.) used for running experiments. Although the author checklist states it's in the appendix, the appendix content is not provided.
Software Dependencies No The paper does not explicitly list software dependencies with specific version numbers.
Experiment Setup Yes For each pseudo key, we sample 32 target text with top-p (p = 0.8) random sampling from base model p. [...] We put all details about hyper-parameter settings in the appendix.