ある 翻訳メモリ(TM) is the backbone of a scalable localization strategy. Technically, it is a structured database that pairs your source segments (sentences) with their approved translations. MultiLipi's engine consults this database in real-time before requesting a new AI translation. If we recognize a sentence you have translated before, we retrieve it from memory instantly—saving you quota, time, and ensuring 100% consistency.
This guide details the Storage, Retrieval, and Deduplication mechanics of your TM.
1. The Mechanics of Memory
How the engine thinks.
Ingestion & Segmentation:
When your site loads, MultiLipi breaks the content into distinct "segments" (usually sentences or short phrases).
Hashing & Lookup:
The system generates a unique hash for each segment and queries your TM database.
Match Found:
The stored translation is served immediately (Zero Quota Cost).
No Match:
The segment is sent to the Neural AI for fresh translation.
Commit:
Once a new translation is generated (and optionally approved by a human), it is "Committed" to the TM for future reuse.
2. The Efficiency Layers
MultiLipi's TM operates on two levels of recognition:
100% Matches (Exact Deduplication):
シナリオ:
Your footer says "Copyright 2026" on 500 different product pages.
アクション:
We translate it 一度だけ . The TM applies that single translation to all 500 instances automatically.
Benefit:
💰 Massive quota savings
Fuzzy Matches (Similarity Detection):
シナリオ:
You have a sentence "Free shipping on orders over $50" and a new sentence "Free shipping on orders over $100".
アクション:
The engine recognizes the high similarity and suggests the previous translation as a base, requiring only a minor numerical edit.
Benefit:
⚡ Accelerated human review workflows
3. Strategic Benefits
Why investing in your TM pays dividends:
| Impact Area | 利点 |
|---|---|
Cost Velocity | As your TM grows, your "cost per page" drops because more content is matched from memory rather than generated fresh. |
Brand Consistency | Ensures that specific phrasing (e.g., "Sign Up" vs. "Register") remains identical across your entire digital footprint. |
Speed to Market | Updates to global headers/footers propagate instantly across millions of pages without manual intervention. |
Cost Velocity
Decreasing cost per page as TM grows
↓ 60%
Average Cost Reduction
Brand Consistency
Identical phrasing across all pages
100%
Terminology Match
Speed to Market
Instant propagation of updates
< 1s
Update Deployment
4. Maintenance Best Practices
Your TM is a database; it requires hygiene to remain effective.
Human Validation:
AI is the start, not the end. When a human reviewer edits and approves a string, that "Gold Standard" version is saved to the TM, overriding the AI version for all future occurrences.
Glossary Integration:
While TM remembers sentences, the Glossary enforces terms. Use them together:
- ザ 用語集 ensures "Apple" is never translated as a fruit
- ザ TM ensures the sentence structure remains consistent
Pruning:
Occasionally audit your translations. If you change your brand voice (e.g., from formal to casual), updating the TM entries ensures the old tone doesn't accidentally resurface in future matches.