Summary of G-sciedbert: a Contextualized Llm For Science Assessment Tasks in German, by Ehsan Latif et al.
G-SciEdBERT: A Contextualized LLM for Science Assessment Tasks in German
by Ehsan Latif, Gyeong-Geon Lee, Knut Neumann, Tamara Kastorff, Xiaoming Zhai
First submitted to arxiv on: 9 Feb 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 presents a large language model called G-SciEdBERT, designed specifically for evaluating German-written responses to science questions. The model is tailored to address the challenges of standard BERT-based approaches, which lack contextual knowledge in the science domain and may not align with student writing styles. To train G-SciEdBERT, the authors used a corpus of 30K German written science responses from PISA 2018, pre-training it on this dataset before fine-tuning it on an additional 20K student-written responses. The results show a significant improvement in scoring accuracy with G-SciEdBERT, outperforming standard BERT-based approaches by 10.2%. This research highlights the importance of specialized language models like G-SciEdBERT for enhancing the accuracy of contextualized automated scoring, contributing to AI applications in education. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates a special computer model called G-SciEdBERT that helps score student answers about science questions written in German. The old way of using BERT didn’t work well because it doesn’t know much about science and students don’t write like the model does. To fix this, the authors used a big collection of science answers to teach G-SciEdBERT. They tested it on more answers and found that it did a lot better than before! This means that G-SciEdBERT is very good at helping teachers grade student work accurately. |
Keywords
» Artificial intelligence » Bert » Fine tuning » Large language model
