ChatGPT has been increasingly in the spotlight recently, and in May 2023, I published a blog about its application in medical translation.
ChatGPT and Medical Translation
In this article, as the second installment, we will focus on ChatGPT prompts (instructions to ChatGPT).
- Table of Contents
Translate Terms as Specified
In medical translation, accurately translating specialized terminology is fundamental. Can this be addressed by skillfully setting prompts for ChatGPT?
Therefore, using clinical trial consent and explanatory documents as material, we issued instructions as follows.
Human translation example: This clinical trial examines the safety, tolerability, and clinical efficacy of three investigational drugs—Drug A (aaaaaaaaaa), Drug B (bbbbbbbbbb), and Drug C (cccccccccc)—in the treatment of adult patients with advanced liver cancer (hepatocellular carcinoma).
ChatGPT's Translation: This clinical trial is designed to evaluate the safety, tolerability, and clinical efficacy of investigational drugs A (aaaaaaaaaa), B (bbbbbbbbbb), and C (cccccccccc) in adult patients with advanced liver cancer (hepatocellular carcinoma).
Note: The drug names have been changed to pseudonyms.
Compared to the human translation example, it is somewhat unnatural as a clinical trial consent and explanatory document, but the terminology translation seems to be done without problems. Specifying terminology translation appears to be an effective prompt.
Prompts in Subtitle Translation
As introduced in the blog below, medical translation sometimes involves translating subtitles for academic conference lectures, panel discussions, and video content.
In subtitle translation, there are character count restrictions, and it is also necessary to translate using language that sounds natural as if spoken by a doctor. Therefore, translators who usually produce accurate translations in a formal style may sometimes have to summarize the content so boldly that it can be quite confusing.
So, by devising the prompt, we tried to see if ChatGPT could produce English translations suitable for subtitles using the following sentence.
Human Translation Example: Here, for clarity, it is classified into three categories: “lung diseases,” “cancer treatment and prevention,” and “chronic inflammatory diseases and genetic disorders,” (55 characters).
The first translation produced by ChatGPT exceeded the character limit (61 characters), and the next one was too short (33 characters).
In the end, by giving instructions not only on the character count but also to include the translation of "to keep things simple" and pointing out errors in particles, we were able to obtain the following translation.
ChatGPT's translation: To keep it simple, diseases were classified into lung diseases, cancer treatment and prevention, chronic inflammatory and genetic diseases, etc. (41 characters)
Is the human translation example still more natural? Also, "into three" was missing from the translation. For now, it might be possible to have ChatGPT translate first and then use that as a hook to create the actual translation.
In any case, it seems necessary to consider the prompts in advance or through various trials in order to approach a level of translation close to that of a human-translated example.
There Are "Tips" for Prompts Too
It seems that there are also tips for how to issue prompts.
In this example, I felt that breaking down various instructions such as character limits and writing in spoken language into separate lines as bullet points reduces the chance of omissions.
However, as far as the author has tried, there were many cases where the prompt regarding character limits could not be properly handled in the following sentences different from the above (I managed to cope by repeatedly giving instructions without giving up...).
Also, if the instruction is "in spoken language," it may result in overly casual translations like "you know" or "...you see." It might be better to refine it to something like "in the words of a speaker at a lecture" or "in the words of a doctor."
This is a bit off-topic from translation, but websites that guide bloggers and YouTubers on how to create prompts have started to appear, and "prompt engineering," which optimizes instructions to get desirable outputs from AI, seems to be becoming a kind of trend. If you are interested, why not look into it?
Summary
For our clients in the medical field, the prompts of ChatGPT are an interesting topic for both medical translators and translation checkers.
The number of characters in ChatGPT's output per single query seems to be increasing over time, but in practical use, it will likely be used in units of about 3 to 4 paragraphs at most, specifying terminology translations or trying to produce translations suited to the intended purpose.
Therefore, when handling large volumes of medical content, it will be used only as a supplementary tool.
Our company also offers human translation services, post-editing services, and the automatic translation software MTrans for Office equipped with ChatGPT integration features. If you have any concerns or interests, please feel free to contact us.
>Easy Translation Software for Office, MTrans for Office
In addition, we provide information related to medical/healthcare translation, as listed below. We hope you find it helpful.
◆Blog
Return to the top of the page
+81 35-321-3111[Tokyo Headquarters] 
Contact Us
Request for Information