- Anson Miu
- Cheryl Chen
- Clara Gila
Sentence-level quality estimation (QE) has emerged as a challenging task in natural langauge processing: given a source sentence and its machine translation (MT), the model should predict a score that expresses the quality of the translation (e.g. higher values indicate accurate translations).
We introduce four different methods for performing sentence-level QE for pairs of source sentences and MT in English and Chinese respectively.
Open Sentence_Level_QE_2020.ipynb on Google Colab and run the cells in order.
Tested on Google Colab
All required module/package dependencies, language models and data files will be downloaded in the notebook cells as required.
- Make sure GPU is selected under Runtime > Change runtime type
- Restart the runtime and run the cells in order - the utility section contains functions used globally throughout the notebook