To address the mixed uncertainty arising from concurrent numerical errors and cognitive vagueness in multi-sensor track association for unmanned surface vessels (USVs), we propose an association method that jointly exploits interval numbers and interval two-dimensional two-tuple (2D2T) linguistic semantics, termed the “Hybrid Uncertainty Modeling” method. Measurement and system errors are modeled as interval numbers, while expert confidence is encoded via interval 2D2T linguistic semantics to construct a hybrid numerical–linguistic support measure. This support is then mapped to an interval basic probability assignment (IBPA), and Dirichlet–Monte Carlo sampling is used to derive adaptive weights for a dynamically weighted interval Dempster–Shafer (D–S) fusion. We conduct 100 Monte Carlo simulations in a two-sensor, twenty-target maritime scenario, benchmarking against nearest-neighbor, numerical-only, linguistic-only, and equal-weight D–S baselines. The proposed approach markedly improves pairwise target separability and decision stability, achieving an average correct association rate of 96.15%, while simultaneously reducing uncertainty and variance and enhancing robustness and interpretability. The principal contributions are: (i) a unified representation and fusion framework for numerical error and linguistic uncertainty, and (ii) an interval “support → IBPA” mapping coupled with a dynamically weighted interval D–S fusion tailored to mixed uncertainty.