An Overview of Neuro-Symbolic Frameworks for Software Development

Neuro-symbolic artificial intelligence aims at combining the learning capabilities of neural models with the structured reasoning provided by symbolic representations. Despite a growing number of frameworks supporting this integration, from a developer’s perspective, the design space of neuro-symbolic systems remains fragmented, making it difficult to compare approaches systematically and to understand their underlying trade-offs. In this article, we present a systematic literature review and an original taxonomy-driven analysis of open-source development frameworks for building neuro-symbolic software systems. We consider four orthogonal dimensions: (i) linguistic aspects, capturing the underlying logic formalism and expressiveness, (ii) inference and reasoning mechanisms, including semantics, solvers, and differentiability, (iii) neuro-symbolic integration pipelines, interpreted through Kautz’s architectural patterns, and (iv) usability and engineering considerations. The analysis identifies open challenges related to balancing expressiveness and scalability, improving engineering maturity, and performing comparisons over standard benchmarks.