Proceedings 21st International Symposium on Logical and Semantic Frameworks with Applications

Imagine the world of computer science as a massive, bustling city. In this city, there are two main districts that usually don't talk to each other much. On one side, you have the "Logic District," where the streets are built on perfect, unbreakable rules like math equations and strict grammar. This is where the city's foundation is laid, ensuring that if you build a bridge, it won't collapse because of a silly mistake in the blueprint. On the other side, you have the "Learning District," a chaotic, exciting place where computers try to figure things out by looking at millions of examples, kind of like how a kid learns to recognize a cat by seeing thousands of pictures of cats.

For a long time, these two districts operated separately. The Logic District was worried that the Learning District was too messy and unpredictable, while the Learning District thought the Logic District was too rigid and slow. But recently, the city planners realized that the future of this city depends on getting these two neighbors to work together. They want to build systems that are as smart as the Learning District but as reliable as the Logic District. This is the big question that keeps computer scientists up at night: How do we mix the "perfect rules" with "smart guessing" to create technology that is both powerful and safe?

This paper is essentially a report from a recent city planning meeting called LSFA 2026, which took place in Lisbon, Portugal, over two days in July 2026. The meeting brought together a group of experts who are trying to bridge that gap between the strict rules of logic and the flexible nature of modern machine learning. Think of these experts as architects and engineers who are trying to design a new kind of building material that is both as strong as steel and as adaptable as clay.

The paper doesn't present a single, finished skyscraper or a magic solution that solves everything overnight. Instead, it's a collection of blueprints, sketches, and experiments shared by researchers who are testing different ways to connect these two worlds. Some of the ideas discussed involve taking the old, well-known rules of how computers think (like the rules of "lambda calculus," which is just a fancy way of describing how functions work) and seeing how they can be updated to handle the new, messy data from machine learning.

The main takeaway from this gathering is that the work is still very much in progress. The researchers are actively exploring how to integrate these methods, running simulations and testing theories to see what works and what doesn't. They are suggesting that by combining the "well-established" techniques with "state-of-the-art" machine learning, we might be able to build better tools for the future. However, the paper makes it clear that this is a journey of discovery, not a destination reached. They are gathering feedback on how to actually use these mixed methods in the real world, acknowledging that while the theory looks promising, the practical application is still being figured out. It's a snapshot of a community of curious minds trying to write the rulebook for a new kind of intelligent technology, one that respects the old laws of logic while embracing the new power of learning.