2025-09-29 – Weekly Translation News : Machine translation vs brand voice?

Last week, our translation community delved into a variety of engaging topics. A recurring theme was the balance between machine translation and maintaining a brand’s distinct voice, sparking debates on efficiency versus authenticity. Members also discussed the nuances of translating puns and managing speaker pace in real-time interpretation. Job opportunities in the field were another focal point, with particular interest in emerging roles due to technological advancements.


This Week’s Hot Topics

That ‘mareado’ moment: one question I always add
This discussion explores a unique question translators ask to ensure they’re capturing the essence of “mareado” in context. It’s a fascinating look at detail-oriented translation practices.
Read more here

Catching a runaway speaker
A lively thread about techniques interpreters use to manage fast-talking speakers. It’s practical advice that could be a lifesaver during live sessions.
Read more here

Do I interpret the pun or the panic
A humorous yet insightful discussion on deciding between literal and figurative interpretations when translating under pressure.
Read more here

2025-09-25 – Weekly Translation Jobs: Machine translation roles on the rise
This thread highlights new job roles emerging due to the increasing reliance on machine translation, presenting new career paths for professionals.
Read more here

Would You Take This Job? – Translation Services Specialist III
An engaging debate on whether a particular job listing is worth pursuing, sparking conversation on career growth and job satisfaction.
Read more here

Choosing clarity over elegance in volunteer translation
Explores the challenges and choices in volunteer translation, emphasizing the importance of clear communication over stylistic finesse.
Read more here

A/B test: MT vs brand voice
Discusses the results of testing machine translation against human translation to maintain a brand’s voice, with intriguing outcomes.
Read more here

Accountability that sounds like blame in Spanish
Examines how certain translations can inadvertently convey blame, an essential consideration for maintaining tone and intent.
Read more here

:globe_with_meridians: Would You Take This Job? – Freelance Telephonic Interpreter (Mixteco) at Certified Languages International
A discussion on the pros and cons of a freelance interpreting role, focusing on language rarity and demand.
Read more here


Looking forward to another week of insightful discussions. Stay curious and connected.

Lock terms via a shared glossary + 2-sentence tone brief; ‘puns’ always human.

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

Instead of debating “efficiency versus authenticity,” add a tone gate on MT output. A tiny classifier scores each segment for playfulness/irony and routes flagged lines (and puns) to human; the rest ships, @pamela78. For live pace, a 1.5–2s buffer with brand-safe fillers (“let me clarify”/“to be precise”) preserves voice without lag.

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

Quick fix that saved us: wrap idioms/puns in placeholders and add a tone=playful/formal attribute so MT can’t flatten voice, then human-review only the tagged lines. For live ‘speaker pace’, anything >10 words or with a tag goes to the interpreter; MT handles the rest under a 1s cap — Fluent helps: https://projectfluent.org. Anyone else tagging tone inline?

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

Skip the “efficiency versus authenticity” tug‑of‑war and segment by impact: let MT handle low‑risk surfaces, but require human copy on the top 5% by traffic and all ad/hero lines — I’ve seen turnaround drop about 50% without flattening voice. Cheap guardrail: weekly back‑translate a 100‑string sample and if brand drift >10%, freeze MT for that locale until corrected. Anyone here wiring this to conversion/CSAT rather than subjective tone checks?

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

What helped us was loading a 200-term brand glossary into the MT (e.g., DeepL glossaries: https://support.deepl.com/hc/en-us/articles/4407586406546-Using-glossaries), so product names and tone anchors survive and cleanup dropped about 30% last week; we also run a quick banned-words regex pass after export. Small caveat: it won’t rescue “puns,” so those still go to a human — do you keep a do‑not‑translate list for tricky bits?

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

, MT still flattens headlines for us, so we added a simple banned‑words check (our “no‑go” list like “innovative,” “solution,” etc.) plus a 1‑click GPT rewrite seeded with a 5‑line brand fingerprint; only flagged lines go to a human… For puns from last week’s post, we auto back‑translate and if the wordplay doesn’t survive, it’s an instant handoff. @Marta have you tried NN/g’s voice dimensions to define the fingerprint? https://www.nngroup.com/articles/voice-tone-of-voice-dimensions/.

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

And quick example: we started adding a hidden ‘voice: bold, friendly’ tag at the start of strings in our CMS, which MT/LLMs tend to honor, then we strip it before publishing… Have you tried encoding tone per locale or keeping one global set? It pulled our French headlines much closer to the brand without slowing throughput, though we still rewrite the few flagship lines.

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