The elicitation of naturalistic linguistic data depends on the cultural and ecological suitability of the stimuli employed to prompt speaker responses. Standardized visual tools in language documentation are typically based on Western contexts, creating a mismatch in linguistically diverse settings such as India. This mismatch results in cognitive load and can affect elicited data due to increased code-mixing, reduced deictic accuracy, and interrupted speech. This paper puts forth a methodological proposal in the form of a Human-in-the-Loop (HITL) workflow for generating culturally grounded elicitation stimuli using generative AI. The approach formalizes iterative prompt engineering, ethnographic constraint specification, and pre-field validation to produce stimuli aligned with specific linguistic targets. Rather than treating generative models as autonomous systems, the framework positions the linguist as an active agent who constrains and evaluates outputs. The paper does not present an empirical evaluation; instead, it demonstrates the operational logic of the HITL methodology and documents the prompt design process in supplementary materials. The contribution is a replicable and transparent workflow for integrating generative AI into linguistic fieldwork.
Mohan et al. (Sun,) studied this question.