Spatial analysis remains structurally inaccessible to qualitative and mixed-methods researchers, not because the questions are absent, but because the gap between research intent and technical execution has no low-cost crossing point. Existing solutions force a binary choice: delegate spatial decisions to a GIS collaborator and lose methodological agency, or acquire technical fluency and lose research momentum. Neither option treats the researcher as the rightful author of their own spatial methods.This paper introduces the translation-layer paradigm for AI-assisted spatial research, in which large language models serve not as methodological advisors but as execution engines that instantiate researcher intent into auditable spatial evidence. The core design principle is the separation of awareness expansion from selection: rather than recommending a method, the system simultaneously executes multiple spatial-method prototypes from a single natural-language research question, enabling researchers to evaluate competing approaches through direct comparison of outputs, assumptions, and data-quality risks. Selection is conditioned on researcher comprehension, not algorithmic confidence. Built-in validation and bias-disclosure mechanisms ensure that outputs meet the minimum traceability standards required for peer-reviewed methods sections. A pilot deployment is underway with health-geography researchers who have received no prior GIS training. This paper presents the formal specification and system architecture of the translation-layer paradigm, together with the pilot study protocol and preliminary implementation observations. The central hypothesis— that the primary barrier to spatial evidence production is translational rather than conceptual—is operationalized as a testable claim about researcher comprehension and methodological ownership.
Congyuan Zheng (Thu,) studied this question.