Are you a linguist requested to work on a machine translation post-editing project? Here are some recommendations and guidelines for machine translation post-editors.
Make sure that you have established clear communication with the company or organization you will be doing post-editing work for, and establish clear expectations and pricing before you begin any work on the project. It is important that both parties agree on the quality goal of the completed target translated document so that you know how much time and effort to spend on the project.
In most cases, if a translated document will be published, full post-editing is required. If an internal document is being translated, a light post-editing may be sufficient. Quality of the completed translation is dependent on several factors. The most essential measure of quality depends on the source document and whether or not it is well written and well suited for MT with post-editing. A well-written source document with MT into the target language with light post-editing may be sufficient for the “good enough” standard meaning that it is useful in its current form and further editing is not required to make the document useful for its intended purpose. As a general rule, whether light or full post-editing approaches are adopted, post-editors are expected to use as much of the raw MT output as possible.
Light MT Post Editing
Here are guidelines for achieving “good enough” quality with light post-editing. First of all, as you do your editing work, a good goal is making sure that the target document is semantically correct. As you process the document, delete any additional suggestions the MT has included. When you compare the source and target documents, make sure that the translation is complete. An important guideline is to modify the translation to include only culturally appropriate language and idioms that will be understood by native speakers of the target language. Since your goal is light post editing, try to preserve the MT as much as possible so you are not doing unnecessary work. Another important guideline is to correct spelling errors as you work. To maintain the light post-editing standards, you should not be concerned about style or editing sentences to improve the flow. These additional steps are unnecessary for light post-editing and you will not be paid sufficiently to be so detail-oriented.
Full MT Post-Editing
On the other hand, if you are tasked with a full post-editing project, you will be processing the MT document with a fine attention to detail and your goal is to produce a translated document that could be mistaken for a fully human translation, or even considered on par with writing done by native speakers of the target language. You must produce a translation that meets all of the standards mentioned above for light post-editing. Then for full post-editing, the first additional guideline is to ensure the translation is syntactically correct, but also grammatically and semantically correct. Pay special attention to technical, industry-specific terminology to make sure it has been correctly translated throughout the document. Try to improve the style, whenever possible. Finally, ensure that the formatting is correct and that you have followed any specific requirements set by the employer.
Light post-editing = Readability
Full post-editing = Readability + Style
Post-editing Quality Check
Upon the request of the client, or for the purpose of testing new freelance post-editors, an extra QC phase might be applied.
Tools Used for MT Post-Editing
There are two recommended methods on currying out Machine Translation Post-editing based on the following cases:
- If the analysis report shows several repetitions and internal fuzzy matches: regular Translation Environment Tools (e.g. Trados, etc. or as required by the client) can be used.
- If the analysis report does not show several internal matches: it is optional to use Translation Environment Tools (e.g. Trados, etc.) unless a specific tool or platform is required by the client. Otherwise, the post-editing document can be in the form of table (e.g. Excel) with columns: Source, Translation, QC Score, QC Comment, Corrections (if any), as well as a final row for Average QC Score.