Optimization of Translation Decisions Under Lexical and Semantic Uncertainty A Decision Theoretic Framework With An Illustrative Case Study

Translation entails choices of meanings, lexical equivalents, and valid expressions in the target language. The choices become complex when multiple interpretations are possible for the source text or when the information provided by the translation tools is not certain. In this paper, we consider a decision theory framework for choosing translations in the face of uncertainty about lexis and semantics. Interpretations of the source are modeled as a finite number of semantic states, whereas choices of target language are modeled as possible actions. Translation choice starts by minimizing expected loss, followed by distributionally robust optimization when probabilities of interpretations are uncertain. Bounded set of ambiguous probabilities enables the decision-maker to manage the degree of conservatism. The model also encompasses a clarification procedure and document-level consistency constraint. A constructed two-sense example demonstrates how a direct equivalent, ambiguity-conserving translation, and clarification by humans can be optimal under varying degrees of uncertainty. These figures are indicative only, and no corpus-based experiments are made. The article ends with the presentation of a real-world test plan and consideration of the drawbacks of the semantic state inventory and probability calibration and automated measures of quality.

Hali tarjima qilinmagan