Is it possible to improve the efficiency of medical translation with ChatGPT? We will introduce real verification results using prompts!

2. How to Improve Efficiency in Medical Translation with ChatGPT? Introducing Some Attempts

Our Medical Group has been conducting verification on effective prompts to utilize ChatGPT for various peripheral tasks that occur in medical translation. Here, we would like to introduce some of them.

2-1. Can you remove unnecessary line breaks within a single sentence? → "Please remove unnecessary line breaks."

There may be unnecessary line breaks in the text when creating PDFs or SRT files (files used for creating subtitles), which can cause one sentence to be split into two, causing problems when using machine translation or translation support tools. However, with the above prompt, line breaks can be removed.

In addition, each sentence will be separated by a period (or comma in Japanese) and a line break with the prompt "Please insert a line break after each sentence" to make it easier to work on translation-related tasks such as subtitle translation.
Although this can also be done with Word's replace function, there are advantages to being able to give instructions in simple language, completing the task with fewer steps, and having intuitive and easy-to-understand results.

2-2. Is it possible to extract the differences between the original text and the back translation (BT) check? → "Please list the "meaningfully" different parts of "Sentence 1" and "Sentence 2" below."

During the translation process, there may be a step called back translation. This involves creating a "translation of the original text" (referred to as back translation: BT text) using the translated text, and comparing it with the original text to check for any mistranslations.
Since the original text and BT text are rarely the same, we have tried to extract only the parts that are "semantically" different by devising prompts to extract only sentences that are different in expression but have the same meaning, rather than extracting sentences that are different in meaning due to mistranslations.
As a result, when we intentionally changed the BT text to the opposite meaning of the original text, "inclusion criteria", and tested it, that part was correctly pointed out.
Of course, there are also parts that are not pointed out by ChatGPT that need to be checked by humans, but it seems to have a certain effect as one of the means of checking.

2-3. Can you extract frequently used words? → "Extract the top 10 terms that appear multiple times in the following sentence."

Not only in the medical field, but also in industrial translation, having a glossary improves work efficiency.
If you create a translation list by picking up important terms for translated documents, it will greatly help with subsequent translations.

After verification in a report related to psychiatry, terms such as BZRA (benzodiazepine receptor agonist), abuse patients, and over-the-counter drugs were picked up. Furthermore, by prompting "how many times each term appears", the number of occurrences of the extracted terms was also shown, and there were no errors in the presented frequency.

On the other hand, when actually using it, it is important to note that words that are not suitable for terms, such as "2018", have been chosen. However, it is easy to create translations in order of appearance in the document, and it can also be said that there is a possibility of extracting only important terms with ingenuity in the prompt.

2-4. Can the translator convert dialects and idiosyncratic speech into standard language for easier understanding? → "Please translate the following text into easy-to-understand standard language with a colloquial tone."

When a native English speaker translates Japanese, there are times when the accuracy of the translation may be affected by the speaker's unique tone, dialect, or use of old words.
While it may be a unique case, when attempting to convert "verlan" Japanese into standard Japanese, it was observed that there was a tendency to shorten the content, but they were able to attempt a rewrite into standard language. While human checking is necessary, it gave the impression that it would save time compared to starting from scratch.

2-5. Can the search time for regular expressions be shortened? → "Please teach me the regular expression for XXXX."

Regular expressions are effective for improving translation efficiency.
For those unfamiliar with programming, regular expressions may seem like a meaningless string of characters such as "^[a-zA-Z0-9]+$". However, by using regular expressions, it is possible to search and replace complex strings, resulting in improved efficiency for processing various types of strings.
For example, in a business setting, regular expressions may be used to extract only strings that appear to be phone numbers from a large amount of text data.
In the case of Japanese-English translation, there is no need to translate already existing English text or English numbers in figures and tables. By extracting only those elements using regular expressions and excluding them from the translation process, the more text there is, the more efficient the process becomes.

However, it can be surprisingly difficult to find information on "regular expressions for extracting half-width alphanumeric characters" by searching the web, and some may even argue that it is better to learn about regular expressions from the basics. In such cases, simply instructing ChatGPT to "teach me how to extract half-width alphanumeric characters using regular expressions" will provide you with the exact regular expression you are looking for: "^[a-zA-Z0-9]+$".

With this regular expression, you can search and extract unnecessary half-width alphanumeric characters from translation targets using translation support tools such as Word or Trados.

2-6. Can translation without word order distortion be generated by brackets?

While a thorough post-edit by a human is essential, current translation cannot be considered without machine translation.
However, if the source text is long, it is prone to dependency errors, and in the worst case, the translated text may fall apart as a sentence if there are many " " or [ ] in the text.
And, it is common for long texts with many parentheses to appear in clinical trial related documents and medical papers.
It is possible to instruct ChatGPT to delete " ", ( ), " ", [ ], etc. in the prompt, but it can also be done with Word's replace function.
Therefore, when we had ChatGPT translate long texts with many parentheses in clinical trial related documents, ChatGPT was less affected by long texts and parentheses, resulting in better translations.
We would like to continue to verify this, including reproducibility.

In addition, we are currently investigating whether it is possible to display the official translation for the UI elements such as Windows that occasionally appear in instruction manuals for medical devices, and whether it is possible to extract only specific languages from multiple languages.

3. Summary

3-1. Efficient Medical Translation with ChatGPT

Above, we introduced various translation-related tasks that can be made easier with clever use of prompts, such as extracting common terms, searching with regular expressions, rewriting into a more translatable style, and removing forced line breaks.
There are concerns about AI taking over human jobs, but as someone who works in the translation industry, I have come to the impression that it is better to actively use and manage AI, rather than fear it, and to use what is available.

3-2. Leave medical translation to Human Science

Human Science Co., Ltd. has a proven track record in streamlining translation work using machine translation and translation support tools. We are constantly exploring ways to improve efficiency in translation and related tasks, such as using AI tools like ChatGPT, and sharing our findings on our blog with everyone.

Human Science Blog Site

We also offer manual translation services and post-edit services. We utilize translation support (CAT) tools such as Trados and Phrase TMS (formerly known as Memsource), as well as automatic translation tools, to improve efficiency and quality of translation, and contribute to the further development of our clients.

Please try Human Science's medical translation, which combines high expertise in document technology and medical science.

If you have any concerns or interests, please feel free to contact us.

Medical Translation Services
Post-editing Services

In addition, we also provide information on medical/medical translation as follows. Please use it.

Related Blogs

What is Medical and Pharmaceutical Translation? Reasons why translation is difficult and points to consider when choosing a translation service provider
How accurate is DeepL in Medical Translation? Comparison results with Google/Microsoft/Amazon
Has the accuracy of Medical Translation improved? Comparison between DeepL and Google Translation in 2020 and 2023
Introduction to books that have helped improve translation quality for Medical Translation professionals!
What is Life Sciences? From the importance of Life Sciences research to methods for obtaining high-quality translations, explained!