StyleLTC: Style-Infused Liquid Neural Network for Automated Claim Identification

ACL ARR 2025 February Submission3062 Authors

Abstract: Claim identification is crucial in NLP for detecting assertive statements, especially with the rise of generative AI and automated fact-checking. Traditional neural networks struggle with the temporal dynamics of language. This paper introduces StyleLTC, which uses liquid neural networks with continuous-time properties to overcome these issues. It also incorporates stylistic features to predict claims. Evaluations show that liquid neural networks outperform static models, offering higher accuracy, robustness, and efficiency. StyleLTC achieves comparable accuracy with only 0.612 MB of memory, far less than traditional models, making it highly scalable and effective for claim detection in combating misinformation.

Paper Type: Short

Research Area: Efficient/Low-Resource Methods for NLP

Research Area Keywords: Linguistic Style, Claim identification, Liquid Neural Network

Contribution Types: Model analysis & interpretability, Approaches low compute settings-efficiency

Languages Studied: English

Submission Number: 3062