Backtracking in the 'not A until B' Construction and Translation Strategies

The current paper examines the structural properties of the English construction not A until B, the backtracking phenomenon it induces, and its implications for machine translation. It highlights the necessity of reversing word order in Korean translations—typically rendered as “~하고 나서야”, “~되 어서야”, or “비로소”—to ensure naturalness and semantic clarity. A preference survey of not A until B, not until B did A, and only when B did A reveals the dominance of the first construction among Korean speakers. In not A until B, the informational focus lies in the latter part of the sentence, and the full interpretation of A’s negation depends on reaching B. This disrupts linear processing and triggers structural reanalysis—an instance of backtracking. In contrast, machine translation systems generate output in a left-to-right fashion, which makes post-hoc restructuring difficult and often leads to rigid or unnatural translations. This study further compares translation patterns for not A until B and not A before B through a survey among university students, empirically distinguishing between backtracking and non-backtracking structures.<br/>ChatGPT-4.0, a generative AI, is also employed to support the analysis.