At the 26th Annual Conference of the European Association for Machine Translation, held in June 2026 in Tilburg, insights from Smartling were brought to the forefront, examining the potential of automated prompts in localization. The research, spearheaded by Smartling's linguistic and data science team, including Senior Linguistic Engineer Marina Sanchez-Torrón and Data Scientists Daria Akselrod and Jason Rauchwerk, presented intriguing results. Their findings suggest that while automated prompts can achieve compliance rates in terminology insertion comparable to expert-written prompts, they excel in certain areas where human expertise is unbeatable (via Slator).

The study concentrated on three pivotal localization tasks: terminology insertion, translation, and language quality assessment (LQA). Smartling's researchers highlighted how automatically optimized prompts showed their strength by improving terminology match rates across several models, thus matching glossary compliance rates typically achieved by localization experts. This indicates that in repetitive tasks like terminology insertion, automation can successfully shoulder some of the traditional burdens carried by human localization experts.

However, when it came to translation quality and language error detection, the dynamics shifted. Localization expert prompts showed superior performance in identifying translation errors. Yet, once these errors were detected, automated prompts demonstrated their edge in effectively categorizing and describing those errors, offering a nuanced understanding that can support subsequent correction processes. This division of task strengths emphasizes the complementary roles of human and machine, rather than a strictly competitive one.

Ultimately, the researchers at Smartling argue that no single method dominates across all localization scenarios. They propose that the process of developing prompts—whether manual or automated—must be assessed through thorough cost-benefit analyses throughout the product lifecycle. By recognizing each method's strengths and constraints, Smartling offers a nuanced framework for integrating automated and expert inputs in a way that enhances overall localization efficiency and effectiveness.