Translation Quality Index (TQI) is our approach for a comprehensive translation quality scoring index. Integrated with Transifex AI, TQI leverages machine learning algorithms to provide a comprehensive framework for real-time translation quality assessment.
It is calculated for each AI translation and evaluates its accuracy, grammatical and syntactical quality, and adherence to linguistic style, ensuring it aligns with your brand's tone of voice. This robust system enhances focused post-edits, automates quality assurance, and facilitates rapid content deployment across multiple languages, streamlining your localization process.
TQI calculation
TQI is being calculated in real-time upon generating translations either through AI Tasks or AI Fillups; it is then displayed right next to each translation in the Editor:
TQI Ranges
The TQI ranges help assess the quality of the translations, giving you a quick insight into how accurate and reliable the AI-generated content is. Here's a breakdown of the TQI ranges:
0.85–1.00: Good
0.70–0.85: Fair
0.50–0.70: Poor
<0.50: Unusable
Not Available: AI Translation have not been saved or reviewed yet
📝 Note: Please keep in mind that if you edit an AI translation and save your changes, the TQI will no longer be displayed for that translation.
Using TQI
To locate AI translations within a specific TQI range, utilize the available filters (under 'More'). You can specify a threshold to filter TQIs that are either lower or higher for further review.
📝 Note: Upcoming features will allow you to set up advanced automation to enhance the localization workflow, such as automatically approving translations that meet or exceed a designated TQI score threshold.
TQI Insights and Breakdown
To provide better transparency and actionable insights, you can now view a detailed breakdown of how each TQI score is calculated, including recommendations on improving translations.
Accessing Detailed Analysis
When checking a translation generated by Transifex AI, you can view a detailed breakdown by clicking the TQI score in the editor. Alternatively, use the shortcut Ctrl+T for quick access.
The modal window that opens displays the score for each TQI component:
Translation Consistency: Measures similarity between translations generated by multiple LLMs, indicating alignment, context clarity, and manageable complexity.
Structural Integrity: Verifies adherence to formatting, glossary use, and structural elements like HTML tags to ensure technical correctness.
Semantic Quality: Assesses the translation’s quality using the MQM framework, evaluating how closely it resembles human-level accuracy.
Further down in this view, you’ll see the issues identified in the Semantic Quality component, with each issue color-coded by severity level (critical, major, minor). This layout allows you to focus on areas needing improvement without any guesswork.
Here are a few examples:
If you find an issue helpful and informative (or not), you can provide immediate feedback by clicking the respective 👍 or 👎 icon. This feedback will help us continuously improve Transifex AI and the algorithm for calculating TQI.
⚠️Warning: TQI analysis is available for AI-generated translations created after November 5th, 2024.
TQI in Workflow Automation
In addition to assessing translation quality, TQI can be used to automate key parts of your localization workflow. These automation options help reduce manual steps and accelerate the review and delivery process when translations meet your quality standards.
Available automation options
Beyond evaluating AI-generated translations, the Translation Quality Index (TQI) can also be used to automate parts of your localization workflow. These automation options allow you to reduce manual effort while maintaining quality standards. To enable these options:
Go to the Project Settings of your desired project.
Open the Workflow tab.
Scroll down to the Quality Assurance section.
Under Review, enable the available TQI-based automation options and set the desired TQI threshold for each.
You can configure the following actions based on a TQI threshold:
Automatically review translations
AI-generated translations that meet or exceed the defined TQI threshold are automatically marked as reviewed and added to the Translation Memory (TM), without requiring manual input.
Automatically add unreviewed AI translations to Translation Memory
AI translations that meet the TQI threshold can be added to TM even if they haven’t gone through manual review.
💡Tip
Looking for more help? Get support from our Transifex Community Forum!
Find answers or post to get help from Transifex Support and our Community.