With the continuous development of language technology, people begin to worry: how to deal with the ethical issues in language technology so that everyone can really benefit from it? The artificial intelligence company OpenAI recently launched a product called.ChatGPTThe product is the latest attempt of translation and communication technology based on artificial intelligence, which makes it more and more necessary to discuss ethical issues in language technology. The language model has been hotly debated in the translation and localization industry since it came out, and language experts are trying to evaluate its impact on this industry.
This article willChatGPTThis paper discusses the ethical issues and related considerations in the application of. In addition, this paper will also discuss in depthChatGPTThe limitations and advantages in translation and localization involveCultural sensitivity1 andTranslation accuracy and other issues..
(Image source: https://www.tomedes.com/translator-hub/ethics-of-using-chatgpt-in-translation-and-localization)
I. Cultural Sensitivity in Translation and Localization
For a long time, there have been problems in the translation of cultural terms in machine translation, which involve literature and art, historical system, traditional customs, religious beliefs and other related works, as well as social behaviors such as festivals and popular phenomena.ChatGPTIs the latest language model based on artificial intelligence, which has caused a lot of hot discussions since it came out:ChatGPTCan you understand the subtle cultural differences in the text?ChatGPTRunning on neural networks, so people thinkChatGPTA text answer similar to human language is generated.
When the user inputs "ChatGPTDo you have what you need to translate the textCultural sensitivity",ChatGPTThe following answers were generated:
"Like all machine learning models, ChatGPT itself is not culturally sensitive. ChatGPT’s cultural sensitivity is based on the data it has trained. If the training data is diverse and culturally sensitive, ChatGPT can generate appropriate translations. However, if the training data is not culturally sensitive, ChatGPT may follow inaccurate translation or continue people’s stereotypes about other cultures. "
What is impressive is that,ChatGPTHave enough "self-awareness" to claim that,ChatGPTAs an artificial intelligence language model, it needs accurate training to get it.Cultural sensitivity. But it also brings a very difficult problem:ChatGPTRunning an algorithm based on input information may create content and translate text according to the subjective ideas of engineers. In fact, existing studies show that there are sexism and racial prejudice in machine translation, which is attributed to the input of information with sexism and racial prejudice in artificial intelligence model.
(Source: https://www.tomedes.com/translator-hub/goals-for-translators)
One of the proposed solutions to this problem is to inputRich corpus and cultural informationcomeincreaseChatGPTDiversity of training received. However, aboutChatGPTThe data used involves further ethical issues, and how to implement this scheme remains to be discussed. Fair use and copyright issues are at the forefront of this debate. Some people point out that obtaining such data and using it violate intellectual property law.
At present, the best choice is to adoptMachine translation post-translation editing (MTPE)：ChatGPTAfter the translation, the translator edits the machine translation to check its relevance and accuracy. Since post-translation editing of machine translation has been an integral part of the translation and localization industry for decades, it has been put into practice.ChatGPTIt is not a big problem to edit after being integrated into machine translation.
In translation and localization, cultural sensitivity refers to the ability to accurately convey the subtle differences and connotations between different cultures in the text. When language experts translate one language into another, they must have in-depth cultural sensitivity, so as to ensure an accurate understanding of the context in which the text information is located, and prevent translation errors and stereotypes at the cultural level from affecting cross-cultural communication.
to be continued …
Special note: this article is only for learning and communication. If there is any infringement, please contact Xiaobian in the background to delete it.
关于作者