Source Text Complexity and Machine Translation Post-Editing Effort: Evidence from Professional Arabic–English Translators across Technical and Legal Domains

Abstract

Machine Translation Post-Editing (MTPE) has become an integral component of professional translation workflows, yet the influence of source text complexity on post-editing effort remains insufficiently understood across specialized domains. This study investigates the relationship between source text complexity and MTPE effort by examining the post-editing behaviour of 24 professional Arabic–English translators working with automotive, pharmaceutical, and legal texts. Drawing on translation process research and machine translation quality estimation, the study operationalizes source text complexity through four linguistic dimensions: Flesch–Kincaid readability, syntactic complexity measured by mean sentence length and clausal embedding depth, domain-specific terminology density, and text type, categorized as procedural, declarative, or argumentative. Post-editing effort was measured using multiple complementary indicators, including keystroke logging captured through Translog-II, time-on-task, and edit distance between machine-translated output and the final post-edited version. Statistical analyses revealed significant positive relationships between all source text complexity measures and MTPE effort, indicating that increasingly complex texts require greater cognitive and physical editing effort. Among the variables examined, syntactic complexity, particularly clausal embedding depth, and terminology density emerged as the strongest predictors of post-editing effort across all text types. Although legal texts did not consistently exhibit the highest readability or syntactic complexity scores, they generated disproportionately greater post-editing effort than automotive and pharmaceutical texts. Detailed error analysis suggested that this increased effort resulted from machine translation systems' recurring difficulties in accurately rendering legal definiteness patterns, terminological precision, and formulaic legal expressions. Keystroke logging data further indicated longer pause durations, increased revision frequency, and more extensive editing operations when translators processed legally complex segments. The findings demonstrate that traditional readability measures alone are insufficient predictors of post-editing effort and that domain-specific linguistic characteristics play a critical role in determining translation productivity. This study contributes to machine translation post-editing research by providing empirical evidence linking source text complexity with translator effort and offers practical implications for machine translation quality estimation, workflow optimization, translator training, and the allocation of post-editing resources in professional translation environments.