fbpx

Lexcode Case Study: Japanese-to-English Localization

Service requested: Japanese-to-English localization and editing of 553,331 lines

Result: Completed

In 2020, Lexcode got the opportunity to work on the localization and editing of English documents translated from Japanese for the creation of term bases and training data for a top multilingual machine translation cloud service in Korea.

Client’s background: Top localization company based in Seoul, South Korea

 

Service requested:

  • English Editing
    • Lexcode offers different levels of English editing. For this project, a mix of proofreading and copyediting was carried out for a large volume of files made up of English conversation strings with different levels of complexity and various file sizes. This process included checking of spelling, grammar, punctuation, sentence construction, style, and natural tone.
  • Localization
    • As one of Lexcode’s main services, Japanese-to-English localization was carried out specifically for more complex sentences that required thorough editing to rewriting. The focus of this service is to ensure that idiomatic expressions, specialized terminologies, and other Japanese language subtleties are properly translated to English so that native speakers of the English language would have no trouble grasping the intended message.

 

Project volume and description:

553,331 lines made up of conversation strings (Excel format) that cover topics ranging from medicine, tourism, hospitality, disaster management, and Japanese culture and lifestyle

 

Sample screenshots:

 

Timeline:

  • February–March 2020 (49 days): Total duration
    • February 12, 2020: Project received
    • February 17, 2020: Start of recruitment for editors and linguists
    • February 29, 2020: Dissemination of project guidelines
    • March 9, 2020: Achieved 1st submission milestone
    • March 12, 2020: Achieved 2nd submission milestone
    • March 17, 2020: Achieved 3rd submission milestone
    • March 20, 2020: Achieved 4th submission milestone
    • March 21, 2020: Achieved 5th submission milestone
    • March 23, 2020: Achieved 6th submission milestone
    • March 24, 2020: Achieved 7th submission milestone
    • March 25, 2020: Achieved 8th submission milestone
    • March 27, 2020: Achieved 9th submission milestone
    • March 27, 2020: Achieved final submission milestone
    • March 31, 2020: Received confirmation from the client

 

Work process:

Importing and splitting of source files → Assessment and classification of files based on topic and quality of English MT → Scheduling with linguists → Task allocation to available linguists → Initial editing → Brief assessment of initial edit (for guideline compliance + amount of changes made; for further revision if not compliant) → QA → Exporting files into client’s preferred output format → Submission

 

Challenges:

  • Tight turnaround time to process and finalize 553,331 lines
  • Inconsistent quality in each file, ranging from average to incomprehensible
  • Inconsistent number of words per line, making it difficult to maintain consistency among all files and assign workable volumes of files to translators and editors

 

Solutions:

  • Use of crowdsourced roster made up of 184 editors and Japanese linguists
  • Shorter internal turnaround times set to accommodate final checking and quality standardization
  • Categorized each file based on translation quality and word count of each string
  • Lower-quality files handled by linguists while higher-quality files were managed by editors
  • All editors and linguists underwent strict and comprehensive training to ensure standardized output
  • Creation of a step-by-step project guideline that includes client requirements as well as general reminders for quality management
  • Cloud-based systems shared among all workers involved in the project were utilized as huge amounts of data were handled, therefore allowing real-time monitoring of output status and accomplishment

 

Results:

  • Achieved all milestones and completed the project on time.
  • Revisions required for only 8 out of 553,331 lines, ending with a 0.0014% error rate.