uninterpretable

Uninterpretable is an adjective meaning not capable of being interpreted or understood. In everyday use it describes something obscure or lacking a clear meaning. In technical contexts, interpretations become more specialized, with interpretable and uninterpretable taking on domain-specific senses.

In machine learning and data science, uninterpretable models are those whose internal decision-making processes are not readily understandable by humans. Examples include deep neural networks and large ensembles. The existence of uninterpretable models raises questions about transparency, accountability, and trust, especially in high-stakes domains such as healthcare or criminal justice. Efforts to address this fall into two broad camps: interpretable-by-design models (such as simple linear models or decision trees) and post hoc explanations (feature importance, surrogate models, SHAP, LIME). The field stresses that interpretability is not binary and that explanations must be accurate, useful, and contestable.

In linguistics, uninterpretable features are formal features that carry no lexical semantic content and must be valued or checked during syntactic computation (for example, in mechanisms like Agree). They enable certain derivations and patterns by being interpreted only after valuation, contributing to cross-linguistic variation and the mechanics of syntax.

Etymologically, uninterpretable is formed from un- + interpretable, with interpret from Latin interpretari meaning to explain. In discourse, the term signals that clarification, transformation, or methodological tools are needed to reach meaningful interpretation.