Want to create an interactive transcript for this episode?
Podcast: Nimdzi LIVE!
Episode: Judge, don’t generate: AI beats MT at QA feat. Marco Baglioni
Description: MT can draft fast, but it often hides meaning errors behind fluent prose.In this session, Marco Baglioni demonstrates how modern LLMs, when used as judges rather than generators, deliver segment-level QA that identifies semantic shifts, enforces terminology (including inflections and multi-word terms), and reduces false positives on source/target inconsistencies.We’ll cover a practical, CAT/TMS-friendly workflow: segment the text, preprocess it for context, run criteria-first prompts, and provide feedback on annotated XLIFF for focused post-editing.Expect a pragmatic playbook: real examples, and what actually moves MTPE time.About Ma...