Content Generation · Read + Write

Translation Thai ↔ EN

Augment Rep · High Reason · Medium

The problem

Marketing produces in English; sales operates in Thai. Generic Google Translate butchers product names, kills tone. Agency translation costs ฿2-5 per word.

The AI approach

Glossary-locked LLM translation. Brand glossary protects specific terms; tone prompt matches your brand voice; integrity check confirms every glossary term survived.

The outcome

~10× cheaper

Translation cost ~1/10 of agency, with brand glossary terms preserved. Translator becomes editor, not first writer.

Try itInput → Process → Output

Input — source text + glossary
Source · Englishblog post · ~480 words
"Our new Plus Plan unlocks unlimited workflows, advanced reporting, and priority support. Built for teams that have outgrown spreadsheets but aren't ready for enterprise complexity."
Brand glossary · 6 locked terms
Plus PlanPlus Plan (do not translate)
workflowsเวิร์กโฟลว์
priority supportการสนับสนุนแบบเร่งด่วน
Process — AI pipeline
1Detect glossary terms in sourceReadPredictive
2Apply brand glossary substitutionsRulesSymbolic
3LLM translate with tone promptWriteGenerative
4Glossary integrity checkRulesSymbolic
Output — translated output
Click Run demo to translate while preserving brand glossary.
Translated · Thai glossary ✓ · 480 words
"Plus Plan ใหม่ของเราให้คุณใช้ เวิร์กโฟลว์ ได้ไม่จำกัด รายงานขั้นสูง และ การสนับสนุนแบบเร่งด่วน ออกแบบมาสำหรับทีมที่เริ่มจะเกินขีดจำกัดของสเปรดชีต — แต่ยังไม่พร้อมจะเข้าสู่ความซับซ้อนระดับองค์กร"
Glossary check: ✓ "Plus Plan" preserved as-is · ✓ "เวิร์กโฟลว์" used (not "ขั้นตอน") · ✓ "การสนับสนุนแบบเร่งด่วน" used. Tone: warm, direct, brand-voice match. Native editor reviews next.

Three AI types in this use case

SymbolicBrand glossary (locked translations); tone rules; banned-phrase list per language; integrity check that all glossary terms survived translation.
PredictiveGlossary term detection (NER-style); language detection; sentence segmentation.
GenerativeLLM translates Thai ↔ EN preserving tone, idiom, and brand voice.

The stack

  • LLM · Claude with tone prompt
  • Glossary · YAML/JSON store
  • Memory · translation memory DB
  • Hooks · CMS / Notion

When this works

  • Brand glossary is documented
  • Native-speaker review remains in the loop
  • Volume justifies the setup (≥ 50k words / quarter)
  • Tone is documented

When it fails

  • Idioms over-translated word-by-word
  • Cultural references (Thai humour, English puns) lose meaning
  • Legal text needs certified translation
  • Auto-publish without native review