Summary of Dahrs: Divergence-aware Hallucination-remediated Srl Projection, by Sangpil Youm et al.
DAHRS: Divergence-Aware Hallucination-Remediated SRL Projection
by Sangpil Youm, Brodie Mather, Chathuri Jayaweera, Juliana Prada, Bonnie Dorr
First submitted to arxiv on: 12 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 This paper focuses on improving the accuracy of semantic role labeling (SRL) for multilingual models, particularly in scenarios where large language models (LLMs) generate spurious role labels. The authors identify that hallucinated role labels are related to naturally occurring divergence types that interfere with initial alignments and develop a new approach called Divergence-Aware Hallucination-Remediated SRL projection (DAHRS). DAHRS combines linguistically-informed alignment remediation followed by greedy First-Come First-Assign (FCFA) SRL projection, outperforming state-of-the-art SRL projection based on LLMs in both human and automatic comparisons. The authors achieve a higher word-level F1 score of 87.6% vs. 77.3% for EN-FR and 89.0% vs. 82.7% for EN-ES using the CoNLL-2009 dataset, with human phrase-level assessments yielding scores of 89.1% for EN-FR and 91.0% for EN-ES. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us make computers better at understanding language by improving how they identify roles in sentences. Right now, it’s hard to train computers to do this for many languages because there aren’t enough labeled data sets. The authors show that when computers try to do this, they often get it wrong and produce fake role labels. They develop a new way of doing this called DAHRS, which is better than current methods at identifying roles in sentences. |
Keywords
» Artificial intelligence » Alignment » F1 score » Hallucination
