AILS-NTUA at SemEval-2026 Task 12: Graph-Based Retrieval and Reflective Prompting for Abductive Event Reasoning

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, short

Summary

AILS-NTUA developed a winning three-stage system for SemEval 2026 Task 12: Abductive Event Reasoning, achieving an accuracy score of 0.95 and ranking first on the evaluation-phase leaderboard. This system integrates graph-based retrieval, Large Language Model (LLM)-driven abductive reasoning enhanced by reflective prompt evolution, and post-hoc consistency enforcement. Beyond the system's success, a cross-model error analysis was conducted across 14 models from 7 families. This analysis identified three shared inductive biases: causal chain incompleteness, proximate cause preference, and salience bias. The convergence of these biases across different model families, indicated by a 51% cause-count reduction, suggests these are systematic failure modes in multi-label causal reasoning, rather than issues specific to individual models.

Key takeaway

For NLP Engineers developing abductive reasoning systems, you should integrate multi-stage approaches like graph-based retrieval and reflective prompting to enhance LLM performance. Be aware that current models exhibit systematic biases such as causal chain incompleteness and proximate cause preference. You can mitigate these by explicitly designing consistency enforcement mechanisms and considering these biases during prompt engineering to improve multi-label causal reasoning accuracy.

Key insights

The system combines retrieval, LLM reasoning, and consistency for abductive event reasoning, revealing systematic causal reasoning biases across models.

Principles

• Causal chain incompleteness is a shared bias.
• Proximate cause preference affects reasoning.
• Salience bias impacts multi-label causality.

Method

The system employs a three-stage approach: graph-based retrieval, LLM-driven abductive reasoning with reflective prompt evolution, and post-hoc consistency enforcement.

In practice

• Use graph retrieval for context.
• Apply reflective prompting for LLM reasoning.
• Enforce post-hoc consistency in outputs.

Topics

• Abductive Event Reasoning
• Graph-Based Retrieval
• LLM Prompting
• Reflective Prompt Evolution
• Causal Reasoning Biases
• SemEval 2026

Best for: Research Scientist, AI Scientist, NLP Engineer, Prompt Engineer

Related on AIssential

• AILS-NTUA at SemEval-2026 Task 12: Graph-Based Retrieval and Reflective Prompting for Abductive Event Reasoning — AILS-NTUA's three-stage system won SemEval 2026 Task 12 for Abductive Event Reasoning, achieving a first-place rank with an accuracy score of 0.95.
• Text vs. K-Graphs: Why Your Multi-RAG System is Failing — LLMs exhibit severe biases, prioritizing information format over factual accuracy when resolving knowledge conflicts.
• Breaking the Chain: A Causal Analysis of LLM Faithfulness to Intermediate Structures — LLMs' apparent faithfulness to intermediate reasoning structures is fragile; they often fail to update predictions after causal intervention.
• Contrastive Reflection for Iterative Prompt Optimization — Contrastive Reflection iteratively optimizes LLM prompts by debugging errors with targeted, validated edits.
• Analysis of the Neglect-Zero Effect in Large Language Models — LLMs may not share the human "neglect-zero effect" cognitive bias regarding empty sets.
• Implicit Reasoning for LLM-based Generative Recommendation, Why Stronger Encoders Make Weaker SPLADE Models, and More! — Recent research focuses on refining LLM-based recommendation and retrieval systems through diverse architectural and algorithmic improvements.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.