This work investigates a neuro-symbolic agentic framework to enhance the reasoning capabilities of SLMs, specifically Gemma 3 (1B, 4B) and Llama 3.2 (3B), using the CLUTRR kinship benchmark.
Abstract
Although large language models (LLMs) have set benchmarks for zero-shot reasoning, their deployment remains cost-prohibitive and environmentally taxing. Small Language Models (SLMs) offer a sustainable alternative, but prone to errors, on tasks requiring complex, multi-hop logical grounding. We investigate a neuro-symbolic agentic framework to enhance the reasoning capabilities of SLMs, specifically Gemma 3 (1B, 4B) and Llama 3.2 (3B), using the CLUTRR kinship benchmark. Our approach transforms the SLM into a minimalist agent utilizing two specialized tool calls: extract_facts for symbolic triplet extraction and get_hint for expert reasoning via a Relational Graph Convolutional Network (RGCN). We evaluate these models across two configurations, both in an Oracle scenario with ground-truth triplets and a Realistic scenario relying on self-extracted knowledge. Our results reveal that while RGCN-derived hints provide a 1.5 - 2x performance gain over story-only baselines, the system is constrained by the extraction bottleneck and sequential deductive fragility, where early extraction errors compound over multi-hop chains. Furthermore, we identify a"distraction effect"in specific architectures where noisy, self-generated facts degrade performance despite the presence of expert hints. This work characterizes the challenges of symbolic grounding in low-resource agentic systems and provides a roadmap for iterative verification in neuro-symbolic agentic pipelines.
Lite-CoNER is proposed, a lightweight NER framework that achieves an effective balance between recognition accuracy and inference efficiency and provides a transparent view of the decision-making process, proving that lightweight models can effectively inherit complex logic through structured distillation.
Yang Wang, Lushuang Gao· International Conference on...· 0 citations
Debate-on-Graph (DoG) is proposed, a new framework that enables LLMs and UKGs to collaborate adaptively for reliable reasoning and introduces a Multi-Agent Debate mechanism, which yields reliable answers through adaptive adversarial debates, aiming to fully exploit the knowledge in UKGs while preserving the reliability of retrieved evidence.
Relational databases (RDBs) play an essential role in real-world scenarios, such as e-commerce, social media, and industry. Recently, with the rapid development of Large Language Models (LLMs), using LLMs to exploit RDBs has become a significant trend. Plenty of works have been proposed to use natural language to describe RDBs or use graph neural networks~(GNNs) to embed relations in RDBs before applying LLMs to them. Despite the achieved progress, existing works still suffer from inevitable weaknesses. For one thing, using natural language to describe RDBs not only causes excessive context length, but also leads to the loss of critical structural information. For another, using GNNs to capture complex structural dependencies requires extensive human-labeled data for supervised fine-tuning, limiting their scalability. Therefore, one important question remains unsolved: '' How to leverage the capability of LLMs to realize robust relational reasoning in RDBs? '' In response, we propose a novel self-supervised framework (ZeroRel) for relational reasoning over RDBs. ZeroRel treats context sparsity as a controllable curriculum variable and leverages it to induce a progressive shift from semantic-dominant inference to structure-aware relational reasoning. Specifically, ZeroRel contains two key modules: Graph-guided Prompt Alignment (GrPA) and Progressive Sparsity-based Context Refinement (PSCR). GrPA uses a heterogeneous GNN to encode multi-table relational structures and projects the resulting structural embeddings into the semantic space of LLMs. PSCR gradually reduces visible attribute context and acts as an information bottleneck, forcing the model to internalize cross-table dependencies rather than relying on superficial semantic shortcuts. Finally, extensive experiments over 7 datasets and 12 downstream tasks demonstrate the superiority of ZeroRel. Furthermore, ZeroRel trained without any task-specific labels achieves an average improvement of 6.24% over models trained with supervised labels.
Yujie Tian, Kun Zhang, Qiuyuan Li et al.· Proceedings of the 32nd ACM...· 0 citations
Treating supervision format as a first-class hyperparameter for multi-task reasoning SFT in large language models—at least in this benchmark-and-model setting—rather than a mere rendering detail is supported.
Nhat Thanh Vu, M. Rashid, Fariza Sabrina· Electronics· 0 citations
The results demonstrate that suppressing entity-linked supervision during pretraining induces a shift in epistemic behavior: KLLMs rely less on parametric knowledge and more on external evidence, leading to improved reliability under realistic conditions, suggesting that pretraining-time control over knowledge acquisition can complement retrieval-augmented and tool-based systems by providing a more evidence-sensitive base model.
Roi Cohen, Yvan Carr'e, Nick Lechtenborger et al.· 0 citations
This framework performs LLM knowledge elicitation to extract factual knowledge from the model’s internal representations and transforms sentence-level representations into entity-level representations and aligns them within a unified space.
Deyu Chen, Qiyuan Li, Jinguang Gu et al.· 0 citations