ITLC at SemEval-2026 Task 11: Normalization and Deterministic Parsing for Formal Reasoning in LLMs
This addresses content biases in LLMs for multilingual formal reasoning, offering a competitive alternative to more complex methods.
The paper tackles content effects in multilingual reasoning tasks for large language models by introducing a method that transforms syllogisms into canonical logical representations and uses deterministic parsing to determine validity, achieving top-5 rankings across all subtasks on the SemEval-2026 Task 11 benchmark.
Large language models suffer from content effects in reasoning tasks, particularly in multi-lingual contexts. We introduce a novel method that reduces these biases through explicit structural abstraction that transforms syllogisms into canonical logical representations and applies deterministic parsing to determine validity. Evaluated on the SemEval-2026 Task 11 multilingual benchmark, our approach achieves top-5 rankings across all subtasks while substantially reducing content effects and offering a competitive alternative to complex fine-tuning or activation-level interventions.