New Python Library CLGC Generates Syllogisms in Knowledge Representation Notations
An innovative open-source Python library named Common Logic Grammar Construction (CLGC) has been launched, designed to automatically produce syllogisms in Knowledge Representation (KR) notations and categorize them according to SEF. This library is part of a research project investigating how various formal KR notations influence syllogistic reasoning in language models (LMs). The study enhances the FOLIO and P-FOLIO datasets, evaluating Small Language Models (SLMs) under Supervised Fine-Tuning (SFT) and Zero-Shot (ZS) conditions. Results suggest that input notation selection can achieve performance levels comparable to natural language while facilitating quicker inference. Additionally, a new syllogistic categorization method (SEF) is introduced to enhance Zero-Shot prompts with logical definitions, improving reasoning in smaller models. The paper can be found on arXiv under the identifier 2608.12374v1, marking the library as the first of its kind.
Key facts
- CLGC is a new Python library for automatically generating syllogisms in KR notations.
- The library also defines SEF categories for the generated syllogisms.
- The study extends the FOLIO and P-FOLIO datasets.
- Experiments were conducted on Small Language Models (SLMs) in Supervised Fine-Tuning (SFT) and Zero-Shot (ZS) settings.
- The choice of input notation can yield performances competitive with natural language while enabling faster inference.
- The SEF method enriches Zero-Shot prompts with logical definitions.
- The paper is available on arXiv with identifier 2608.12374v1.
- The research is focused on syllogistic reasoning in language models.
Entities
Institutions
- arXiv