Summary of Comparing Explanation Faithfulness Between Multilingual and Monolingual Fine-tuned Language Models, by Zhixue Zhao et al.
Comparing Explanation Faithfulness between Multilingual and Monolingual Fine-tuned Language Models
by Zhixue Zhao, Nikolaos Aletras
First submitted to arxiv on: 19 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper explores the faithfulness of feature attribution methods (FAs) in multilingual and monolingual natural language processing models. FAs provide insights into how input features contribute to predictions, but previous studies have mainly focused on English monolingual models. This study covers five languages and five popular FAs, showing that FA faithfulness varies between multilingual and monolingual models. The results suggest that larger multilingual models are less faithful than their monolingual counterparts, with potential drivers being differences in tokenizers. The study contributes to understanding the limitations of FAs in real-world applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how well certain methods explain why language processing models make predictions. These methods tell us which parts of the input text helped make a prediction. Most studies have only looked at English, but this one explores five different languages and finds that the methods work differently depending on whether the model can understand many languages or just one. The study also shows that larger models that understand many languages are less good at explaining their predictions than smaller ones that only understand one language. This is important because it means we need to be careful when using these methods in real-life applications. |
Keywords
» Artificial intelligence » Natural language processing
