The Ethics of Using ChatGPT in Translation and Localization

As the field of language technology continues to advance, there's a lot of concern about how it can be used ethically so everyone can truly benefit from it. This question becomes even more relevant with the recent launch of OpenAI's ChatGPT, the latest AI-based translation and communication technology. This model has been a major subject of debate within the translation and localization industry as professionals try to evaluate how it will affect the sector.

Today, we will explore the ethical issues and considerations surrounding the use of ChatGPT. We will dive into the limitations and merits of the tool in translation and localization when it comes to cultural sensitivity, translation accuracy, and other challenges.

Watch the video to know more.

Cultural Sensitivity in Translation and Localization

Before we discuss cultural sensitivity, let’s first discuss what it is. In translation and localization, cultural sensitivity is the ability to convey a text’s cultural nuances and connotations accurately. When linguists translate from one language to another, they must possess in-depth cultural sensitivity to ensure that the context of a text’s message is appropriately captured to prevent damaging cultural inaccuracies and stereotypes.

For the longest time, machine translation has had this issue, which you can see in this research paper. ChatGPT's appearance as the latest AI-based language model has sparked a lot of conversation about whether it can understand a text’s cultural nuance.  Because this program is run by a neural network, it has been noted to provide human-like answers. So, we asked ChatGPT ourselves if it had the cultural sensitivity required to translate text. Here is its response:

“ChatGPT, like any machine learning model, is not inherently culturally sensitive. Its cultural sensitivity is based on the data it has been trained on. The model may produce appropriate translations if the training data is diverse and culturally sensitive. But if the training data is not culturally sensitive, the model may produce translations that perpetuate stereotypes or inaccuracies.”

It’s impressive that ChatGPT has enough "self-awareness" to proclaim that, as an AI language model, it needs to be accurately trained to gain cultural sensitivity. But it presents a very troublesome issue: as a program running its algorithm based on the information it’s fed, it may therefore create content and translate text according to subjective ideas held by its engineers. In fact, it has already been revealed that biases in gender and race are found in machine translation, which are due to AI-models getting information that perpetuates these ideas.

One of the proposed solutions to circumvent this problem is to increase the diversity of the training ChatGPT receives by relying on a wealth of languages and cultures for its input. The approach to how this will be implemented is still a matter of discussion, as there is the further ethical question about the data ChatGPT’s program uses. Fair use and copyright are now at the forefront of this particular debate, with some pointing out that obtaining and using such data is a breach of intellectual property laws.

The best option, for now, is to implement a machine translation post-editing (MTPE) process: after ChatGPT translates the text, human translators post-edit it to check for relevance and accuracy. Since MTPE has been a part of the translation and localization sector for decades, incorporating ChatGPT wouldn’t be much of a problem. Companies  that provide localization and translation services with an MTPE process can get accredited by standardizing institutions, like the International Organization for Standardization (ISO).

If you want to learn more about ISO and the different certifications a localization and translation provider can apply for, you can check out this article: How Do You Achieve the ISO Standard for Translation Services?

 

Accuracy in ChatGPT Translations

Accuracy in translation is crucial for clear and effective communication. ChatGPT has the potential to greatly improve the speed and efficiency of translation, but like previous neural machine translations before, it faces issues of inaccuracies.

This is problematic because inaccurate translations can lead to a domino effect of misunderstandings and misinterpretations. For fields such as law and medicine, there are serious consequences when it comes to mistranslated documents. Individuals' lives can literally be on the line, so it is important to make sure that the models used to assist them can create representations that are exact on both the linguistic and cultural level.

Since ChatGPT is a machine-learning model, it can be taught and tailored for specific fields. However, there are a lot of elements that must be considered when using it for a niche-specific translation. For example, for a legal translation, in addition to mastering the cultural and lexical nuances of the languages, ChatGPT will also have to understand two judicial systems in all their complexity.

As of now, ChatGPT is available for free use under beta testing. This means it can't be modified or tailored to fit a specific domain. But once it is able to, it has the potential to translate more accurately than it currently does.

 

Ethical Implementation of ChatGPT in Translation and Localization

Since we discussed the ethical concerns and challenges of fully utilizing ChatGPT in translation and localization, let's discuss how we can effectively and responsibly implement it.

One of the biggest ethical concerns of ChatGPT is its impact on human translators. Will professional linguists lose their jobs to a machine? One article we published explains that, according to experts, neural machine translation has an accuracy rate of 60-90% and is improving at 3-7% every year. Despite these claims, neural machine translations still obviously lack the cultural and linguistic nuances that a human translator has. The localization and translation sector will therefore always need human translators to post-edit the machine-translated content.

As previously mentioned, ChatGPT is still in its beta testing phase. But once the model is finalized, there's a possibility that it can be used for niche-specific content like law and medicine. However, a study conducted on the use of neural machine translation in legal practice found that even though these technologies have reached "human-like levels" of translation, lawyers who use them exclusively for their work are at higher risk of legal malpractice. Why? Well, any mistranslations made by MT tools can be grounds for negligence on the lawyer's side.

Every word matters in the eyes of the law, and any incorrect translation in a legal document can have dire consequences for both parties involved. For this reason, human translators specialized in the legal domain are favored in court because they have a deep understanding of the legal systems they work with. For now, machine translation post-editing is the most practical way to gain the benefits of both AI and human translators.

It's clear that ChatGPT and other neural network language models will need to be further trained. In the meantime, beyond their use within the MTPE process, these tools can be used as a personal assistant for translation tasks, such as creating emails and proofreading. Since it's a relatively new technology, there is a lot of potential for ChatGPT to be integrated into the localization and translation workplace that can go beyond the post-editing process.

The most probable challenge the language industry will face is how to train people to use ChatGPT and the upcoming AI technology in a way that enhances their skills and expertise to make language-based services more reliably accessible without compromising their quality.

 

FAQs

Are there any specific industries where the ethical use of ChatGPT is more critical?

Yes, certain industries require a more careful ethical approach when using ChatGPT. In the medical and legal fields, for example, accuracy and confidentiality are paramount, as mistranslations can have serious consequences. Similarly, the finance and governmental sectors demand high levels of precision and privacy. If you work in these industries, ensuring the ethical use of AI involves strict adherence to data privacy regulations and rigorous validation processes to guarantee the accuracy and reliability of translations.

How does ChatGPT address potential biases in its translations?

ChatGPT aims to minimize biases through extensive training on diverse datasets. However, biases can still occur. OpenAI, the developer of ChatGPT, continually updates the model to address these issues by refining training data and employing fairness and inclusivity guidelines. As a user, you can help by providing feedback on biased outputs, which contributes to ongoing improvements. By being aware of and actively addressing biases, you can use ChatGPT more ethically and effectively.

How can users ensure the quality of translations provided by ChatGPT?

Ensuring the quality of translations from ChatGPT involves several steps. First, provide clear and detailed instructions to the AI for context-specific translations. Review and edit the output to correct any inaccuracies or cultural nuances. You can also cross-check translations with other tools or consult professional translators for critical documents. Regularly giving feedback to ChatGPT helps improve its performance over time, ensuring better quality in future translations.

Conclusion

There's a lot of potential in ChatGPT in the translation and localization sector. However, because it still lacks cultural sensitivity and accuracy, it's safe to say that the model won't be replacing human translators anytime soon. By understanding the challenges ChatGPT may face in considering cultural nuances and ensuring accurate translations, we can work to mitigate these potential issues. These efforts may include incorporating diverse and culturally sensitive training data, relying on human oversight and feedback, and turning to cultural and domain experts in the training process. With these considerations in mind, we can unlock the full potential of ChatGPT while maintaining ethical standards.

If you want to learn more about how the language industry assesses machine translations, click here: How Is Machine Translation Quality Assessed?