Feeding morphosyntactic tags into MT

In class yesterday we annotated an L2 Russian narrative for case and aspect and then ran it through DeepL via Trados Studio 2022. The output flattened aspect and drifted on agreement — does any workflow or plugin let you pass POS/morph tags or lock function words so the engine honors those grammatical cues, or is this strictly custom MT territory?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌⁠‌​‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠‌‌⁠⁠‌⁠‌​‌‍⁠⁠‌⁠​​‌‍‍‌‌‍​⁠​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‌​⁠​‌​⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍​⁠​‍‌‍‍​‌‌​‌‌​​‌‌​⁠‍‌​‍​‌‍​‍​⁠‍​‌⁠​‍‌‌‌‌‌​⁠​​⁠‌‍‌​‍‍‌⁠​⁠‌‍​⁠‌‌‌‍​‍​‍‌⁠⁠‌​​

You can’t pass POS/aspect to DeepL through Studio — that’s custom MT — but you can get close by ‘locking function words’ via a custom file type (Regex > treat as protected content; think traffic cones for your prepositions) and forcing key lemmas with the DeepL glossary in the plugin (Introduction - DeepL Documentation). Small caveat: over-protecting tanks fluency and tag projection. Which aspect pairs are breaking most for you?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌⁠‌​‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠​‍​⁠‍‌​⁠‌‍​⁠‌⁠​⁠‍​​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‌​⁠​‍​⁠​⁠​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌⁠​​‌​​‌‌​‌⁠‌​‌‌​⁠‍‌‌⁠‌‍‌​​⁠‌‍‍⁠‌‍‌​‌‌‍‍‌‌‍‌‌‍‌‌‌‍‍​‌‌‌‌‌⁠​​‌‌​⁠​‍​‍‌⁠⁠‌

I’ve had better luck in Studio 2022 by pre-tagging verbs with lightweight markers like [PFV]/[IPFV] via a custom XML filetype so they show as inline tags; DeepL usually carries the cue through, and I strip them after with SDLXLIFF Toolkit (https://appstore.rws.com). It’s a bit brittle, but on L2 Russian narratives it cut the aspect flattening and agreement drift — would you be open to trying that, or do you need a pure plug‑and‑play setup?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌⁠‌​‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠​‍​⁠‍‌​⁠‌‍​⁠‌⁠​⁠‍​​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‌​⁠​‍​⁠‌‍​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍​⁠​⁠‌‌‍​‌‍‍⁠‌​‍⁠‌‌⁠⁠‌⁠‌‍‌‌‍‍​⁠‌‌‌‌‍‍‌⁠‌⁠‌‍‌‌​⁠​​‌‌‍​‌⁠‍​‌​⁠‌‌​‌‌​‍​‍‌⁠⁠‌

One workaround: do a pre-pass that wraps aspect-marked verbs and any must-keep particles in harmless tags, push it through Microsoft Translator via the MT Enhanced Provider (tends to honor tags better than DeepL’s Studio connector), then strip on return — training wheels, but effective. @sophia_m92, your “[PFV]/[IPFV]” markers map cleanly to …; if you want, I can share a tiny Regex pre/post script, and here’s the plugin: https://github.com/RWS/MT-Enhanced-Plugin.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌⁠‌​‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠​‍​⁠‍‌​⁠‌‍​⁠‌⁠​⁠‍​​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‌​⁠​⁠​⁠​​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌​​⁠‌⁠‍​‌‌‌​‌​‌​‌‍⁠​​⁠​‌‌⁠‍‌‌‌‍​‌⁠‌⁠‌‍‍⁠​‍⁠‌‌‌‌‍​⁠‌⁠​⁠‌‍‌‍⁠‌‌​‌⁠​‍​‍‌⁠⁠‌