This project is from SemEval 2020 Task 4 - Commonsense Validation and Explanation (ComVE) Subtask C. The dataset for this project is from 1.
Approaches:
- NMT
- GPT-2
- KaLM(Bart-Large)
- GRF
data_dir
: experiment results with GPT-2:subtaskC_answers_CONFIG.csv
are the files of the generated dev resultssubtaskC_answers_test_CONFIG.csv
are the files of the generated test results
GPT-2
: please refer to the readme under this directoryNMT
: please refer to the readme under this directoryKaLM
: please refer to KaLMgraph_method
: GRF is implemented under this directory, please refer to the readme under this directory
System | Dev(BLEU) | Test(BLEU) | Human | PT LM Params |
---|---|---|---|---|
Copy Source | 16.53 | 17.23 | N.A. | N.A. |
NMT | around 1.0 | around 1.0 | 0.59 | N.A. |
GPT2-small | 14.01 | 13.66 | 1.80 | 117 M |
GPT2-medium | 16.79 | 16.08 | 1.94 | 345 M |
KaLM (BART-Large) | 18.98 | 18.5 | 2.08 | 406 M |
GRF (GPT2-small) | 17.83 | 17.56 | 1.91 | 117 M |
Muti-Task (GPT2-small) | 17.11 | 17.82 | 1.97 | 117 M |
GRF (GPT2-medium) | 18.36 | 18.45 | 2.21 | 345 M |
Muti-Task (GPT2-medium) | 19.15 | 18.31 | 2.26 | 345 M |
- Copy Source - from the original data