Summary of Improving Bilingual Capabilities Of Language Models to Support Diverse Linguistic Practices in Education, by Anand Syamkumar et al.
Improving Bilingual Capabilities of Language Models to Support Diverse Linguistic Practices in Education
by Anand Syamkumar, Nora Tseng, Kaycie Barron, Shanglin Yang, Shamya Karumbaiah, Rheeya Uppal, Junjie Hu
First submitted to arxiv on: 6 Nov 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 This paper investigates the effectiveness of multilingual large language models (MLLMs) in assessing student writing, particularly in a bilingual context. Prior studies have focused on LLM-powered learning analytics, but this study explores how well MLLMs perform in grading explanations of Science and Social Science concepts written in English, Spanish, or Spanglish (a mix of both languages). The researchers found that pre-trained models exhibit a significant bias towards bilingual writing, leading to inaccurate assessments. To address this issue, they fine-tuned open-source MLLMs like Llama 3.1 and Mistral NeMo using synthetic datasets in English, Spanish, and Spanglish. The results show improved performance for all three languages after fine-tuning with bilingual data. This study highlights the importance of enhancing MLLM effectiveness to support bilingual learners’ language practices and underscores the value of incorporating non-English languages into language model design. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well computers can grade student writing in different languages. Right now, these computer models are mostly trained on English texts, which can lead to mistakes when grading writing that mixes English and Spanish (called Spanglish). The researchers tested these models and found they make more errors when grading bilingual writing. To fix this problem, they “trained” the models using fake data in English, Spanish, and Spanglish. This helped the models become better at grading all three types of writing. The study shows that we need to improve how these computer models work so they can help bilingual students learn more effectively. |
Keywords
» Artificial intelligence » Fine tuning » Language model » Llama
