Desenvolvimento de uma linguagem natural controlada para programacao de agentes BDI

Abstract

Programming multi-agent systems grounded in the BDI architecture still requires simultaneous knowledge of artificial intelligence concepts, deliberative behavior representation, communication mechanisms, and execution-environment details. Although the MASPY framework reduces part of this complexity by providing abstractions for agent-oriented development in Python, there remains a gap between conceptual modeling and effective implementation. In this context, this work presents CLAG, a controlled natural language designed for the specification of BDI agents through restricted, readable, and semantically predictable linguistic constructions. The project proposes a translation architecture capable of converting textual descriptions written in controlled English into executable code compatible with MASPY. The developed solution is organized into four main stages: lexical and syntactic analysis, semantic interpretation, construction of an intermediate representation, and automatic code generation. The resulting language supports the definition of agents and environments, parameterized beliefs and desires, agent communication, message broadcasting, loss events, perceptual filtering, declarative environmental actions, and scalable instance creation. The project was evaluated through representative usage scenarios designed to verify the expressive power of the language and the consistency of the generated translation. The obtained results indicate that CLAG is a relevant contribution to reducing the complexity inherent to BDI agent development, favoring prototyping, teaching, and experimentation in multi-agent systems.