Comparing zero-shot self-explanations with human rationales in multilingual text classification
Brandl, Stephanie, Eberle, Oliver
–arXiv.org Artificial Intelligence
Instruction-tuned LLMs are able to provide an explanation about their output to users by generating self-explanations that do not require gradient computations or the application of possibly complex XAI methods. In this paper, we analyse whether this ability results in a good explanation by evaluating self-explanations in the form of input rationales with respect to their plausibility to humans as well as their faithfulness to models. For this, we apply two text classification tasks: sentiment classification and forced labour detection. Next to English, we further include Danish and Italian translations of the sentiment classification task and compare self-explanations to human annotations for all samples. To allow for direct comparisons, we also compute post-hoc feature attribution, i.e., layer-wise relevance propagation (LRP) and apply this pipeline to 4 LLMs (Llama2, Llama3, Mistral and Mixtral). Our results show that self-explanations align more closely with human annotations compared to LRP, while maintaining a comparable level of faithfulness.
arXiv.org Artificial Intelligence
Oct-4-2024
- Country:
- Asia > Middle East
- UAE (0.14)
- Europe (0.68)
- North America
- Mexico > Mexico City (0.14)
- United States > Washington
- King County > Seattle (0.14)
- Asia > Middle East
- Genre:
- Research Report > New Finding (0.86)
- Industry:
- Law > Labor & Employment Law (0.50)
- Technology: