News

2026

October 2026: “ArgForge: A Platform for Structured Argumentation Analysis and Explanation” is now publicly available on GitHub.

• Description: ArgForge is an open-source Python framework for computational argumentation based on ASPIC+. It supports the construction and evaluation of structured arguments, attacks, defeats, and acceptability semantics, alongside explainable reasoning and comparative analysis across different reasoning configurations. The framework also provides a bilingual interface in English and Turkish.
• Web Interface: ArgForge
• Code: GitHub

September 2026: “ArgGYM: A Procedural, Engine-Verified Benchmark for Structured Defeasible Reasoning” available on arXiv.

• Description: This work introduces ArgGYM, a procedural benchmark and reinforcement learning environment for structured defeasible reasoning. Every item is generated by construction and verified against an argumentation engine, and its twelve tasks cover both reading what a theory entails and constructing changes that alter it, such as attacks, defences, and preference adjustments.
• Preprint: arXiv:2609.38409
• Code: GitHub

June 2026: “A Diagnostic Framework for Compositional Generalization via NL-to-FOL Translation” published in the Journal of Logic, Language and Information.

• Description: This article presents a systematic diagnostic framework for evaluating compositional generalization in neural semantic parsing models on the NL-to-FOL translation task, introducing the WillowNLtoFOL dataset and comprehensive metrics for assessing both structural correctness and semantic equivalence of model outputs.
• Journal: Journal of Logic, Language and Information
• Article: Enhanced Reading
• Code: GitHub
• Dataset: Hugging Face

March 2026: The WillowNLtoFOL and WillowNLtoFOL_extended datasets are now publicly available on Hugging Face.

• Description: Originally developed in the 2024 master’s thesis and introduced in the ESSLLI 2025 article “Transformer Models for Translating Natural Language Sentences into Formal Logical Expressions,” these datasets provide a structurally diverse and rigorously filtered benchmark for evaluating the compositional generalization capabilities of neural semantic parsing models on the NL-to-FOL translation task.
• Dataset: Hugging Face

March 2026: “Parametric Sound Field Interpolation for Scene-based Navigable Immersive Audio” accepted for publication in the IEEE Transactions on Audio, Speech and Language Processing.

• Description: This article presents a parametric sound field interpolation method using multi-point sparse plane wave decomposition to enable perceptually veridical audio rendering for users navigating within immersive volumetric environments.
• Link: IEEE Xplore

2025

December 2025: Book chapter titled “Yapay zekada sembolik ve sembolik olmayan ayrımı üzerine” (On the distinction between symbolic and non-symbolic in AI) published in Yapay Zeka Felsefesi, edited by Zekiye Göz (Doruk Yayınları).

• Description: The chapter analyzes the operational differences between symbolic expert systems and neural networks within the context of cognitive science and artificial general intelligence.
• Link: Yapay Zeka Felsefesi

November 2025: “The Ouroboros of Benchmarking: Reasoning Evaluation in an Era of Saturation” accepted for poster presentation at the NeurIPS 2025 Workshop on Evaluating the Evolving LLM Lifecycle: Benchmarks, Emergent Abilities, and Scaling.

• Description: This study investigates performance trends across 52 benchmarks for OpenAI, Anthropic, and Google model families to examine how rapid evaluation saturation impacts the measurement of reasoning.
• Poster: NeurIPS 2025
• Preprint: arXiv:2511.01365

October 2025: “Global PIQA: Evaluating Physical Commonsense Reasoning Across 100+ Languages and Cultures” available on arXiv.

• Description: This work introduces a non-parallel benchmark designed to evaluate physical commonsense reasoning across more than 100 different languages and cultural contexts.
• Preprint: arXiv:2510.24081
• Benchmark: Hugging Face Dataset