Systems and methods of automatic post-editing of machine translated content
Automatic post-editing of machine translated content is disclosed herein. An example method includes presenting machine translated segments of a document and their associated quality estimation scores, invoking an automated post-editing system for segments with unsatisfactory translation quality, inputting the segments into a generative AI model alongside contextual information, producing a revised translation of the segment using the generative AI model and iterating the generative AI process with varying prompts until a quality estimation score of the machine translated segment meets or exceeds a quality estimation score threshold, or a predetermined number of attempts are reached.
1. A method for automated post-editing of machine translated content, the method comprising:
presenting machine translated segments of a document and associated quality estimation scores for each of the machine translated segments;
invoking an automated post-editing system for machine translated segments with unsatisfactory quality estimation scores;
inputting the machine translated segments with unsatisfactory quality estimation scores into a generative AI model, the generative AI model using contextual information for the document;
producing a revised translation of a machine translated segment using the generative AI model; and
iterating with the generative AI model with varying input until:
(a) a quality estimation score of the machine translated segment meets or exceeds a quality estimation score threshold; or,
(b) a predetermined number of attempts are reached.
2. The method of claim 1 , wherein iterating includes:
generating an updated quality estimation score for the revised translation; and
generating an additional revised translation when the updated quality estimation score is unsatisfactory.
3. The method of claim 1 , wherein quality estimation scores associated with the machine translated segments of the document are generated using a machine translation quality estimation model that is configured to identify machine translated segments with unsatisfactory translation quality.
4. The method of claim 1 , wherein the contextual information is provided to the generative AI model, the contextual information comprising a variable window of text adjacent to each of the machine translated segments with unsatisfactory quality estimation scores, each of the machine translated segments with unsatisfactory quality estimation scores to be processed by the automated post-editing system.
5. The method of claim 1 , wherein the generative AI model is employed for revising the machine translated segment and is based on a transformer architecture.
6. The method of claim 1 , wherein the automated post-editing system utilizes a machine translation quality estimation model to determine which machine translated segments require automated post-editing based on quality estimation scores.
7. The method of claim 1 , wherein the generative AI model is iteratively prompted with the varying input to enhance a quality of the revised translation, with the varying input being generated based on unsatisfactory aspects of previous translation attempts.
8. The method of claim 1 , further comprising presenting an option for manual intervention and input to the generative AI model to further improve a quality of the revised translation.
9. The method of claim 1 , wherein the predetermined number of attempts is for achieving a satisfactory translation and is configurable.
10. The method of claim 1 , wherein the automated post-editing system maintains a log of iterations, including prompts and responses, to facilitate analysis and quality control of the automated post-editing system.
11. The method of claim 1 , wherein the automated post-editing system incorporates feedback from a database to override a top candidate generated by the generative AI model.
12. The method of claim 1 , further comprising providing an option to override a choice made by the automated post-editing system.
13. The method of claim 1 , further comprising calculating a before machine translation quality estimation (MTQE) score and calculating an after MTQE score, for a whole paragraph or the document.
14. The method of claim 1 , further comprising accepting human translation as input instead of machine translation and applying automated post-editing to enhance the human translation.
15. The method of claim 1 , further comprising incorporating an additional agent to further enhance output of the automated post-editing system.
16. A method for enhancing machine translated content using contextual information, the method comprising:
translating a source segment into a target segment using a machine translation engine;
evaluating a quality of the target segment with a machine translation quality estimation model;
extracting contextual information from surrounding content and available metadata of the target segment;
inputting the target segment, the quality, and the contextual information into a generative AI model;
automatically generating an improved translation of the target segment based on the target segment; and
iterating the improved translation with the generative AI model until:
(a) a quality estimation score of the target segment meets or exceeds a quality estimation score threshold; or,
(b) a predetermined number of attempts are reached.
17. The method of claim 16 , wherein iterating includes:
generating a quality estimation score for the improved translation; and
generating an additional improved translation when the quality estimation score for the improved translation is unsatisfactory.
18. The method of claim 16 , wherein the predetermined number of attempts is for achieving a satisfactory translation and is configurable.
19. A system for automated improvement of machine translated content, the system comprising:
a memory for storing executable instructions; and
a processor coupled to the memory, the processor for executing the executable instructions to perform a method, the method comprising:
translating a source segment into a target segment using a machine translation engine;
evaluating a quality of the target segment with a machine translation quality estimation model;
extracting contextual information from surrounding content and available metadata of the target segment;
inputting the target segment, the quality, and the contextual information into a generative AI model; and
automatically generating an improved translation of the target segment based on the target segment.
20. The system of claim 19 , wherein the method further comprises iterating the improved translation with the generative AI model until:
(a) a quality estimation score of the target segment meets or exceeds a quality estimation score threshold; or,
(b) a predetermined number of attempts are reached.
21. The system of claim 20 , wherein the predetermined number of attempts is for achieving a satisfactory translation and is configurable.