Home / FAQ / False language detection and correction

False language detection and correction

This page was last updated on 2026-05-28

Keep.eu stores project information in many European languages. During the importing and updating of project data, translated project fields may occasionally be assigned the wrong language.

The purpose of this workflow is to identify and correct these situations, helping ensure that project information is searchable, filterable, and correctly displayed in keep.eu. This process contributes to the quality, consistency, and multilingual usability of project information in keep.eu.

Correct language attribution directly affects:

  • Multilingual search;
  • Filtering;
  • Readability;
  • Discoverability of projects and outputs;
  • And the overall consistency of keep.eu data.

This workflow therefore helps improve the quality and usability of multilingual Interreg information across programming periods, programmes, and countries.

How does this work?

Twice a year, the keep.eu data manager checks translated project fields against their assigned languages, with the help of a data check function in the keep.eu system. Whenever a translated field changes, or a new translated field is created, the system compares the content of the text with the language assigned to it. If the system suspects that the text is written in a different language, it creates a ‘false translation object’ in keep.eu’s Django Admin, a high-level Python web framework used by keep.eu as a general back-office. Potential false translations are then reviewed manually.

Currently, the following translated project fields are checked:

  • Name
  • Description
  • Expected results
  • Achievements

Latest check: June 2026.

Operational limitations

Automatic language detection may occasionally:

  • Misclassify short texts;
  • Struggle with multilingual content;
  • Incorrectly identify specialised terminology;
  • Or produce unsupported language identifications.

Because of this, manual validation remains necessary.

The workflow therefore intentionally combines:

  • Automatic detection;
  • Human validation;
  • And semi-automatic correction processing.

Workflow in detail

A detailed operational description of this workflow, including periodicity, exports, validations, review procedures, and exceptional cases, can be consulted here.