Smartling’s AI Post-Editing Agent is an innovative AI prompt designed to significantly enhance the overall quality of machine translations by reviewing grammar, tone and semantic accuracy - all in a single, automated step. By enriching machine translations with your linguistic assets, the AI Post-Editing Agent also helps produce a translation output that sounds more like your company's unique voice.
Info: The AI Post-Editing Agent is part of Smartling's AI Toolkit and can be combined with all other AI Toolkit features.
How it works
Smartling’s AI Post-Editing Agent provides a fully automated way of reviewing and enhancing the translation output of any machine translation workflow.
The following steps all happen automatically with a near-instant turnaround.
MT or LLM translation
The AI Post-Editing Agent can be used with any of the machine translation providers or LLMs supported by Smartling, including custom trained MT engines.
Assessment of the translation output
The AI Post-Editing Agent uses a Large Language Model to assess and improve the quality of each translated string, based on four criteria:
- Grammatical correctness
- Fluency
- Semantic coherence
- Lexical accuracy
The machine translation is automatically reviewed and optimized based on these criteria. Any detected inaccuracies or incoherent translations are fixed automatically.
Custom prompt with RAG
With the help of RAG (Retrieval-Augmented Generation) technology, the machine translation output is also enhanced based on your linguistic assets:
- The machine translation output is compared to matches from your translation memory that may be available. Translation memory matches are used as a reference to apply your company's stylistic preferences (e.g. tone of voice, formality register and terminology).
- Based on your translation memory, the AI Post-Editing Agent also extracts locale style rules to optimize translations based on your preferred linguistic conventions.
Locale style rules include information on your preferred tone (e.g. whether colloquial expressions are appropriate to use, how cultural references should be handled, which overall tone should be applied and how your audience should be addressed), as well as information on how specific text elements like abbreviations, acronyms, dates, measurements, currencies and numbers should be handled. Locale style rules also take into account your preferred use of punctuation and capitalization, as well as syntactic elements like pronouns, plurality, verb forms and syntactic order. - Your Quality Check Profile is taken into account, to ensure the quality of the MT output meets your specific quality preferences. Where possible, any detected quality check errors are fixed automatically by the AI Post-Editing Agent.
- If Glossary Compliance quality checks are enabled, the AI Post-Editing Agent verifies if any available glossary terms were applied in the machine translation output. Based on any detected glossary compliance quality check errors and examples of glossary terms used in the context of the available TM matches, the AI Post-Editing Agent inserts glossary terms where possible.
Info: The AI Post-Edit Agent only takes into account the assets that are set up in the Linguistic Package associated with the project(s) where the workflow using the AI Post-Editing Agent is used.
Mitigation
Hallucination detection helps prevent hallucinations and achieve a reliable, high quality translation output.
Final result: High-quality and easy-to-use translations
The final result is a machine translation output that is not only more fluent, coherent and accurate, but also better reflects your company's tone of voice. If desired, the translations can be combined with a step for human post-editing.
Benefits
Smartling’s AI Post-Editing Agent can help reduce the amount of human editing required by a linguist in any post-translation steps before the string can be published.
The AI Post-Editing Agent can also be used for MT workflows without any human in the loop, to achieve a higher level of translation quality without any human involvement. By improving automated translations, you can achieve a faster time-to-market with a cost-efficient solution.
Tip: Use Smartling's Language Quality Estimation Agent to better gauge if a post-editing step is required for machine translations reviewed by the AI Post-Editing Agent.
Prerequisites for using the AI Post-Editing Agent
Supported target locales
For optimal performance, the AI Post-Editing Agent should be used with high-resource languages. On average, the best results are achieved for the following target locales: es-ES, es-LA, de-DE, fr-FR, fr-CA, pt-BR, pt-PT, jp, ko, it, nl, zh-CN and zh-TW.
Reliable linguistic assets
To achieve good results with the AI Post-Editing Agent, your Translation Memory and glossary should be up-to-date and of good quality. Since these assets are used as a reference, the AI Post-Editing Agent can only efficiently review and optimize your machine translations if all assets correctly reflect your preferred translation style, tone and terminology.
How to enable the AI Post-Editing Agent
The AI Post-Editing Agent is part of Smartling’s AI Toolkit, which can be used as an optional add-on for any machine translation workflow or MTPE workflow. If you are currently not using the AI Toolkit, please reach out to your Customer Success Manager for details.
Once the AI Toolkit is enabled for your account, the AI Post-Editing Agent can be activated for any machine translation workflow by following these steps:
- The AI Post-Editing Agent needs to be enabled on a workflow step level. As a first step, locate the machine translation workflow that you would like to use the AI Post-Editing Agent for.
- Click on the Translation step of the workflow. Then select the option “Manage Step”.
- Please ensure that the workflow is already configured with an MT Profile as the translation provider. If the workflow is not yet configured as a machine translation workflow, please follow these steps before proceeding.
- Once your machine translation workflow is fully configured, switch on the toggle “Enable AI Post-Editing Agent”.
- Click Save to confirm.
Note: As part of the AI Toolkit, the AI Post-Editing Agent is a paid feature. If another AI Toolkit feature is already enabled for a workflow, no additional cost will incur by enabling the AI Post-Editing Agent.
How does the AI Post-Editing Agent interact with other AI Toolkit features?
The AI Post-Editing Agent can be combined with all other AI Toolkit features.
The AI Post-Editing Agent functions as an all-in-one, overall translation enhancement that takes many different factors into account. Other, more granular AI Toolkit features can be used to provide further, more targeted translation improvements and analysis.
For example, if a glossary term has not been applied by the AI Post-Editing Agent (e.g. due to the threshold specified for your Glossary Compliance quality checks), the additional use of AI-Enhanced Glossary Term Insertion helps apply the correct terminology.
Tip: To maximize the benefits of the AI Toolkit, we recommend enabling the AI Post-Editing Agent on all workflows that are already using other AI Toolkit features.
In the future, the AI Post-Editing Agent will comprise additional AI Toolkit features.
AI Post-Editing Agent vs. AI Fuzzy Match Repair
If you are using AI Fuzzy Match Repair on the same workflow where the AI Post-Editing Agent is also enabled, please note that strings going through AI Fuzzy Match Repair will not be reviewed by the AI Post-Editing Agent.
- If AI Fuzzy Match Repair is enabled and a TM match above the specified threshold is available, the TM match will be inserted and repaired. This means that no machine translation is applied, and the AI Post-Editing Agent is not used.
- If no TM match meeting your specified threshold is available, a machine translation is applied and the AI Post-Editing Agent is used to enhance the machine translation output.