Workshop Description

Deep Learning (DL) and Large Language Models (LLMs) have significantly advanced many Natural Language Processing (NLP) tasks (Zhao et al., 2025). However, these models still have limitations, particularly in complex linguistic scenarios such as multiword expressions or long-context reasoning (Miletić & Walde, 2024; Cheng et al., 2024). Moreover, critical concerns remain about scalability, interpretability, and robustness to adversarial inputs (Barone et al., 2017; Anjum & Krestel, 2024).

This workshop responds to a recent interest in revisiting rule-based and knowledge-based methods that often offer high precision, better explainability, and domain adaptability—features essential in applications like grammar checking, legal document analysis, and medical NLP. Despite the dominance of end-to-end neural models, recent comparative evaluations show that symbolic approaches still perform competitively (or even do better) on specific tasks, especially when training data is scarce or interpretability is critical (Mitkov et al., 2024; Vastl et al., 2024).

In addition to exploring these traditional methodologies, the workshop will examine emerging trends that offer alternatives or complements to LLMs. These include retrieval-augmented generation (RAG) techniques (Su et al., 2025; Wu et al., 2025), neurosymbolic AI that integrates neural and symbolic reasoning (Sheth et al., 2023; Bhuyan et al., 2024), and few-shot or zero-shot learning approaches (Zeng & Xiao, 2024). General-purpose foundation models (Church & Alonso, 2024) and knowledge-graph–powered systems (Chen et al., 2025; Vidal et al., 2025) also represent promising directions.

This workshop aims to encourage a dialogue between advocates of symbolic and statistical models and to critically evaluate how both approaches can complement each other in building robust, efficient, and explainable NLP systems.

Topics

This workshop invites research papers on comparing rule-based and knowledge-based methods across various NLP applications, highlighting the emerging trends and challenges for the current NLP world. Therefore, we expect to cover the following topics:

• The role of rule-based and knowledge-based approaches in modern NLP. Despite the dominance of ML and DL, hybrid models, rule-based approaches, and knowledge-based approaches remain relevant in areas requiring high precision and explainability. Traditional techniques are still valuable for specific domains, especially when data is scarce, interpretability is required, or robustness to adversarial inputs is essential.
• Comparative analysis of rule-based, machine-learning, deep-learning and large language models for different NLP tasks. Evaluating different methodologies across various NLP tasks is crucial for understanding their strengths, limitations, and best applications. We also seek to explore rule-based techniques for modern NLP applications since rules and heuristics remain crucial for preprocessing, domain-specific NLP, and hybrid systems.
• Emerging trends in NLP research beyond deep learning and Large Language Models. General purpose models, neurosymbolic AI, retrieval-augmented generation (RAG), and few-shot learning are some examples of methods for advancing NLP.
• Limitations and performance bottlenecks in scalability and accuracy of deep learning models. Large-scale models require vast training data, extensive compute power, and struggle with long-context reasoning. Overfitting when fine-tuning or lack of interpretability remain key challenges.