Design and Implementation of CTran V3: A Multilingual Translation Platform for Ctrip International Business

1. Project Background

CTran is the Ctrip International Business Unit (IBU) content translation platform responsible for multilingual processing of hotel, flight and other product information to support Ctrip’s internationalization.

The latest V3 version, launched in early 2018, was completely redesigned and rebuilt to address existing business challenges and establish a new workflow.

This article shares the experience and lessons learned from developing CTran V3, hoping to provide references for similar projects within the department and the industry.

2. Language Challenges in Internationalization

Ctrip’s Chinese system contains massive tourism product data for domestic users. IBU must leverage this resource to provide seamless, language‑free services to overseas users, which is difficult because most data were not originally designed for translation.

Structured information is easier to translate, but a large portion of hotel and flight data is unstructured, requiring extensive cleaning and regular synchronization with source changes.

Hotel data, in particular, demands significant translation resources due to its complexity and variability.

3. Overview of CTran V3

1. Overview

The overall architecture consists of data ingestion, offline or online translation processing, and various output interfaces.

Unautomated data is handled by translators using V3’s translation assistance and content analysis tools; results are stored in multi‑level dictionaries for future batch processing.

For hotels, data is transformed, translated offline, and pushed to IBU’s hotel information database; for flights and other real‑time data, V3 provides an online translation service.

2. Hotel Data Flow

Langs, a sub‑project of CTran V3, synchronizes hotel data, detects changes, and generates translation requests based on table‑driven configuration.

Change detection uses configurable tables to map Chinese and English changes to actions such as updating records or sending translation messages.

3. Translation Tasks

Translation is organized as tasks. Priorities are defined by configurable search criteria rather than static flags, allowing dynamic filtering and flexible task sizing.

Search optimization uses simplified storage, redundant data, indexed databases and ElasticSearch to accelerate data location.

4. Data Analysis and Reporting

CTran V3 integrates with Ctrip’s big‑data platform to perform NLP‑based analysis, generate statistics, and produce dashboards or email reports.

5. Translation Engine Mechanics

The engine combines template‑based, fuzzy‑match, word‑splitting and multi‑level dictionary strategies, exposing intermediate results for translators to fine‑tune translations.

6. Job Management

Over 70 jobs are scheduled on Ctrip’s Qschedule platform to handle data consistency, batch translation, index rebuilding, cache updates and dictionary snapshots.

7. Real‑time Translation

Flit, the real‑time translation service, adds multi‑level caching, achieving sub‑5 ms response for 99 % of flight translation requests.

8. Static File Translation Support

Excel files can be imported/exported for batch translation, with Apache POI utilities wrapped for streaming read/write.

4. Conclusion

CTran V3 is a fully re‑engineered solution that balances engineering efficiency with business needs, providing a maintainable, extensible platform for multilingual content delivery.