Machine translation (MT) of Chinese-English journalistic texts frequently suffers from discourse-level coherence errors that are undetectable by sentence-level metrics. This study proposes a theory-driven method to identify and revise such errors using thematic progression (TP) as a diagnostic framework. Taking the machine-translated version of Ideas of Journalism in Contemporary China produced by ChatGPT as an example, we extracted 156 clauses and identified two dominant TP patterns in the source text (constant theme and derived theme), and documented pattern-specific MT errors as follows: ambiguous reference and improper thematic shift under the constant pattern, and cohesion loss and information focus shift under the derived pattern. Based on these error types, we develop three post-editing strategies specification, combination, and amplification--each designed to repair a specific coherence failure while preserving the original TP structure. The findings demonstrate that TP theory provides an objective, replicable heuristic for detecting coherence errors in MT output and guiding targeted revision. This study contributes a practical framework for post-editing training and offers implications for discourse-aware MT evaluation.
Chao Tang (2026) studied this question.