Unsupervised Commonsense Question Answering with Self-Talk

Various benchmarks for Multiple choice question answering

1. COPA: Choice of Plausible Alternatives
2. CommonSenseQA: commonsense Question Answering
3. MC-TACO: Multiple Choice Temporal commonsense
4. Social IQa: Social Interaction Question Answering
5. PIQA: Physical Interaction Question Answering
6. WinoGrande

Proposed work in the paper

1. They started with a pre-trained Language model
   1. We assign each of the answer choices a_i into the combination of the context and the question, and obtain opt_i = combine(c, q, a_i). The combine function is computed differently for each task.
2. Baseline model with external knowledge
   1. External knowledge comes from clarification list
3. Self-talk model