What is AI Fuzzy Match Repair (AI FMR)?
As part of Smartling’s AI Translation Toolkit, AI Fuzzy Match Repair enhances the functionality of TM Match Insertion in Machine Translations.
AI Fuzzy Match repair helps increase your Translation Memory leverage in Machine Translations by optimizing available matches. As a result, more of your existing translations can be reused, helping you achieve a higher level of consistency, as well as potential cost savings.
How it works
If a Translation Memory Match above a certain threshold is available for a string, TM Match Insertion allows you to use the existing translation from your Translation Memory instead of a Machine Translation. The existing translation from your Translation Memory is inserted and submitted automatically to the next step in the workflow. This means that no Machine Translation is used for that string, helping you achieve cost savings and a higher level of consistency.
By improving the match percentage, AI Fuzzy Match Repair allows you to use Translation Memory matches instead of a Machine Translation for even more strings:
- If a fuzzy match is available for a string, this means that some similar source text was translated before, and the previous translation is stored in your Translation Memory.
- AI Fuzzy Match Repair uses a Large Language Model to detect the differences between the new source string and the source string which was translated previously.
- The translation which is stored in the Translation Memory is then automatically adapted to the new source text.
- This generally results in a translation that is a better fit for the new source string and in a higher fuzzy match percentage.
- The repaired fuzzy match is inserted automatically and submitted to the next step in the workflow.
Example:
The translation stored in your Translation Memory uses the Spanish noun for "photo", as well as demonstrative adjective in the feminine form ("esta foto"). However, the new text to translate uses the noun "chat" instead of "photo":
| Example | |
| Previous source string (English) | Are you sure you want to delete this photo? |
| Spanish translation stored in the TM | ¿Seguro que quieres borrar esta foto? |
| New source string (English) | Are you sure you want to delete this chat? |
If TM Match Insertion is enabled, the previous translation from your TM is automatically reused instead of a Machine Translation. At the same time, AI Fuzzy Match Repair automatically changes the translation of the noun from "photo" to "chat", and adjusts the grammatical structure. In Spanish, the grammatical gender of the word "chat" is masculine, so AI Fuzzy Match Repair also adapts the demonstrative adjective to the masculine form:
|
Example (repaired fuzzy match) |
|
| Repaired Spanish translation | ¿Seguro que quieres borrar este chat? |
The corrected (repaired) translation gets submitted automatically to the next step in the workflow. Due to the adjustment to the new source text, the string is now considered to have a higher match percentage.
Supported workflows
AI Fuzzy Match Repair is available for Machine Translation workflows and can be used with any Machine Translation provider (Smartling Auto Select, a preferred MT engine of your choice or even a custom trained MT engine).
For MT workflows managed by Smartling Language Services and for AI-Powered Human Translation, this feature is enabled automatically.
How to enable AI Fuzzy Match Repair
- AI Fuzzy Match Repair is part of Smartling’s AI Translation Toolkit and cannot be purchased separately. So in a first step, the AI Translation Toolkit needs to be enabled for your account.
For more information or to enable the Toolkit, please reach out to your Smartling Customer Success Manager.
- Once the AI Translation Toolkit has been activated, TM Match Insertion needs to be enabled on the MT Profile which is used for your Machine Translation workflow:
- From your Account Settings, access the Machine Translation Settings > Profiles.
- On the MT Profile you want to enable TM Match Insertion for, click the "Disabled" link under the "TM Match Insertion" column.
- Switch on the toggle to Enable TM Match Insertion.
- Configure the TM Match percentage that should be leveraged by the MT Profile. Any matches above the specified TM Match threshold will be inserted and submitted automatically.
- Click "Save" to confirm these settings.
- AI Fuzzy Match Repair then needs to be enabled on a workflow step level:
- In your Smartling project, navigate to Settings > Workflows.
- On a workflow that uses Machine Translation, click Manage Step on the Translation step.
- Ensure that the toggle to Enable AI Fuzzy Match Repair is switched On.
By default, this setting is disabled.
Tip: If this setting is not visible for your workflow, please ensure to select an MT Profile in the "Translation Provider" dropdown first.
Note: It is generally advisable to include a step for human post-editing in the workflow, so the inserted TM matches can be revised before they reach the Published step. If Translation Memory matches are inserted without further post-editing, there may still be a discrepancy between the existing translation from your TM and the new text to translate.
How to save translation costs with AI Fuzzy Match Repair
AI Fuzzy Match Repair can help skip the Machine Translation step and any associated costs for a larger volume of content.
If your MT workflow uses a step for human post-editing, AI Fuzzy Match Repair can also help cut down on the editing costs, due to more strings having higher fuzzy match scores. Instead of a Machine Translation, the human editor will be able to revise the inserted, repaired Translation Memory match. Since this can significantly reduce the required editing effort, many Language Services Providers grant cost discounts for content with a higher fuzzy match.
To benefit from cost savings based on fuzzy matches, ensure to apply your Fuzzy Match Settings also to the post-translation step. From Account Settings > Fuzzy Match Settings > Workflow Steps, select one of the following options:
- Apply your Fuzzy Match Profile settings to the Translation step and Post-Edit step:
This option will only take effect for MTPE workflows using the workflow step type "Post-Edit". - Apply your Fuzzy Match Profile settings to the Translation step and all post-translation steps:
This option will take effect for workflows using the workflow step types "Post-Edit", "Edit" or "Review", including workflows for human translation.
Word Count Report with repaired fuzzy match scores
Your Word Count Report takes into account the benefits of AI Fuzzy Match Repair. To help you achieve cost savings for content going through a human post-editing step, this report displays the new fuzzy match score after repair.
Based on your vendor agreement with your Language Services Provider, lower post-editing rates may apply for content with a higher fuzzy match score.
Repaired fuzzy matches in the CAT Tool
If your workflow includes a step for human post-editing, the repaired TM match is shown in the translation field of the CAT Tool, as the translation to review.
The Language Resources panel in the CAT Tool shows the original TM match (before repair), as well as the repaired match with the new (higher) fuzzy match percentage.
Important considerations
AI Fuzzy Match Repair yields the best results if the associated Translation Memory is of good quality, and if the content to be machine translated is similar to previously translated content. If you are unsure about the quality of your TM, please reach out to your Customer Success Manager, who can advise on our options for optimizing your TM prior to enabling TM Match Insertion and AI Fuzzy Match Repair.