Readable Proofs for the Horn Fragment

Scientific Talk and Mentoring Q&A with Lydia G. Kondylidou (LMU Munich)

2026-11-17

Date: 21 December 2026

Time:

10:00 ‒ 11:00 - Talk

11:15 - 12:15 - Mentoring Q&A. Discussions on AI, automated reasoning, logic, and cybersecurity.

Location: TU Wien, FAV Hörsaal 3 Zemanek (Seminarraum Zemanek) (Favoritenstraße 9-11, Erdgeschoß, 1040 Wien) (HHEG01)

Abstract: Proofs generated by resolution and superposition provers are often hard for humans to read. This is partly because they are proofs by contradiction, and partly because they rely on unification. In this talk, I will present our work on producing readable proofs for the Horn fragment of first-order logic with equality. We introduce a structured proof format that expresses direct proofs. It combines equational rewriting chains with human-style “have/hence” derivations, and it relies on matching instead of full unification, which makes proofs easier to follow. To interface with provers, we introduce an algorithm that translates resolution and superposition proofs into our format, and implement it in a tool called Tälja. We connect Tälja to the Twee equational prover, which supports the Horn fragment through an encoding, and evaluate our approach on TPTP benchmarks.

The talk is based on joint research with Jasmin Blanchette, Nicholas Smallbone, and Uwe Waldmann.

Bio: Lydia G. Kondylidou a researcher in the Department of Theoretical Computer Science and Theorem Proving at LMU Munich. She completed her PhD at LMU Munich in November 2026 (magna cum laude) under the supervision of Prof. Jasmin Blanchette, with a thesis titled Finding, Simplifying and Presenting Proofs Automatically.

Her research focuses on automated reasoning, particularly automated theorem proving. During her PhD, she investigated extending SMT solvers with native support for higher-order reasoning, techniques for proof minimization in first-order logic, and methods for translating machine-generated proofs into readable, direct proofs. Her broader research interests include methods for finding, minimizing, and presenting proofs in both first-order and higher-order logic.