Summary of Argument Mining in Data Scarce Settings: Cross-lingual Transfer and Few-shot Techniques, by Anar Yeginbergen and Maite Oronoz and Rodrigo Agerri
Argument Mining in Data Scarce Settings: Cross-lingual Transfer and Few-shot Techniques
by Anar Yeginbergen, Maite Oronoz, Rodrigo Agerri
First submitted to arxiv on: 4 Jul 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 Recent research in sequence labelling has focused on developing strategies to address the lack of annotated data for most languages. Notable approaches include model-transfer using multilingual pre-trained language models, data translation and label projection, and prompt-based learning exploiting few-shot capabilities. Previous work suggests that model-transfer outperforms data-transfer methods, while few-shot techniques based on prompting surpass fine-tuning. This paper empirically demonstrates that these insights do not apply to Argument Mining, a sequence labelling task requiring the detection of complex discourse structures. Instead, the authors show that data transfer achieves better results than model-transfer and that fine-tuning outperforms few-shot methods. Crucial factors in data transfer include the dataset domain, while few-shot performance is influenced by task length, complexity, and sampling method. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper explores ways to help computers understand language without needing lots of labeled training data for every language. Researchers have tried different approaches like using pre-trained models or translating data from one language to another. Previous studies suggested that these methods work well for some tasks, but this study shows that they don’t always apply. The researchers found that when trying to identify long and complex arguments in text, it’s actually better to use translated training data rather than a pre-trained model. They also discovered that fine-tuning the model works better than using prompts with limited labeled data. |
Keywords
» Artificial intelligence » Discourse » Few shot » Fine tuning » Prompt » Prompting » Translation
