Urban form is interpreted through the lens of linguistic systems, proposing that the spatial organization of cities can be read as a language structured by syntax, semantics, and pragmatics. Using large language models (LLMs) trained on urban descriptions and spatial data, the study demonstrates that cities exhibit regularities akin to grammatical rules. Urban elements—streets, plots, buildings—are treated as tokens within a structured spatial discourse. The model captures patterns across scales, revealing how design intentions, regulations, and human behavior form a codified urban "text." The LLM approach transcends traditional urban morphology by offering predictive and generative capabilities: it not only deciphers existing forms but can propose plausible future configurations. This suggests a shift in urban theory—from representation to performativity, from map to model. Urban design becomes a computational act of meaning-making, where space is encoded, processed, and reimagined through data-driven language. The research frames the city as an emergent structure, shaped by recursive interactions between form, use, and symbolic value. It opens possibilities for automated design, participatory modeling, and the integration of cultural narratives into planning. The city, like language, is shown to be both historically grounded and generatively open.