Oct 30, 2021 · In this work, we proposed a simple but effective constraint random decoding method for back-translation, which follows an APE framework. First, ...
To alleviate this problem, we propose a simple but effective constraint random decoding method for back-translation. The proposed method is based on an ...
摘要. Back-translation has been proven to be an effective data augmentation method that translates target monolingual data .
We empirically verify this discovery and propose a method to generate diverse translations by ma- nipulating heads.
Jan 10, 2024 · MBR decoding is shown to generate higher quality sentences than random sampling and beam search in directed text generation tasks including ...
Our goal in this work is to find a way to force the output of a model to contain such lexical constraints, while still taking advan- tage of the distribution ...
Dec 6, 2023 · porated in the translations. 2. Random constraint: Random (but presumably correct) word mappings are obtained using a word alignment tool ...
Missing: Diverse | Show results with:Diverse
Sep 19, 2022 · Abstract. When generating text from probabilistic models, the chosen decoding strategy has a profound effect on the resulting text.
Random sampling. This decoding mode randomly samples tokens from the model output distribution. This strategy is frequently used in back-translation techniques ...
Oct 9, 2023 · Random constraint: Random (but presumably correct) word mappings are obtained using a word alignment tool and provided as a pseudo- terminology ...
Missing: Diverse | Show results with:Diverse