In pursuit of precision: Can ChatGPT rise to the challenge?

New technologies are rapidly changing the landscape of many industries, including translation. OpenAI’s ChatGPT is a remarkable addition to the technological landscape, quickly becoming one of the most popular and successful applications of large language models (LLM). Its ability to drive innovation and improve efficiency across various fields has earned OpenAI, its developer, a $29 billion valuation

However, ChatGPT has a number of limitations that make it unsuitable for many translation needs, particularly those required by businesses. Not only does it struggle with appropriately translating syntax errors, slang, and misspellings, but it is also not zero-trace. This means businesses are putting sensitive internal data and customers’ personal information at risk if they try to translate content via ChatGPT.

So, can ChatGPT truly address the business world’s real-time translation requirements? The short answer is no. Even if the security issues weren’t a blocker, ChatGPT is still under development and is just not there yet from a translation perspective — especially for businesses that need to translate large amounts of text or require high-accuracy translations.

Instead, businesses should consider equipping tools that enhance translation quality through contextualization technology. While ChatGPT showcases promising capabilities in various domains, its limits become apparent when applied to complex real-time translation scenarios.

Contextualizing ChatGPT’s role and limitations

ChatGPT’s popularity immediately skyrocketed following its launch in 2022 because of its ability to generate realistic, coherent responses. It even has the capacity to produce accurate translations — in some cases. 

Despite its innovations, ChatGPT’s limitations render it unsuitable for businesses aiming to effectively translate substantial volumes of text. One noteworthy challenge is its difficulty comprehending text encompassing multiple languages. This hurdle becomes particularly pronounced in scenarios involving code-switching, a linguistic phenomenon where individuals fluidly transition between different languages within a conversation. Such code-switching is exceedingly common in some populations, especially in minority communities where multilingualism is a norm rather than an exception. This nuanced interaction of languages poses a significant obstacle for many machine translation tools, including ChatGPT, undermining the ability to produce valid and intelligible translations. 

ChatGPT’s weaknesses are further accentuated by the complexity of context and cultural subtleties. While the technology performs well in translating straightforward and clear content, it struggles with linguistic intricacies, like idiomatic expressions, culturally-specific terms and slang, often resulting in confusing, imprecise and inauthentic translations.

Though ChatGPT has access to an expansive dataset, it performs the strongest in English because that is the language that the bulk of the data used to train it is written in. While it can understand other widely-used languages, it struggles with less commonly published dialects, especially those originating from regions with little-to-no internet access. 

Misspellings add to ChatGPT’s mistranslations, leading to incorrect or convoluted interpretations and breakdowns in the intended meanings of the messages. This obstacle is especially problematic in real-time communication, such as customer service interactions, where misspellings are common and can impede effective translation.   

The promise and peril of generative AI in customer service and machine translation

Generative AI is rapidly changing the landscape of customer service and machine translation. Chatbots like ChatGPT can provide 24/7 customer support in multiple languages, seemingly saving businesses time and money. 

However, ChatGPT and other tools used for translation wrestle with complex customer requests. Below are some examples of issues with other popular translation or LLM-based tools:

  • Google Translate is one of the most popular machine translation engines, known for its speed and wide breadth of languages. However, Google Translate doesn’t perform equally in all of its many offered languages, and can sometimes produce poor or unnatural-sounding translations.
  • DeepL is a newer machine translation engine gaining popularity for its accuracy and fluency. While some linguists believe DeepL’s translations are of superior quality to Google Translate, it similarly struggles to accurately translate colloquialisms and slang. 
  • LaMDA is designed to generate accurate and factually correct text based on the input it receives. It is a language model from Google AI that can generate text, translate languages, write creative content and answer questions using a dataset of accurate information. Yet it can sometimes produce biased or faulty results, especially when asked to construct text about sensitive topics.

Using generative AI tools like ChatGPT in customer service and machine translation also has ethical implications. The massive dataset of text and code that these technologies are trained on is from the internet, meaning they can inherit biases and inaccuracies directly from the source. 

Such a tool can only access publicly available data, so gated content, which requires specific permissions or actions to access, is typically beyond its scope and may not be part of the knowledge base. This restriction poses a challenge when dealing with queries or texts referencing information hidden behind access barriers, as the LLM upon which tools like ChatGPT are trained may lack the necessary context to provide accurate translations or responses.

And the security concerns associated with ChatGPT can’t be emphasized enough. Even with secure versions, ChatGPT retains stored content. While this might facilitate continuity, it also raises valid privacy concerns, particularly when dealing with sensitive or confidential information.

Enhancing translation excellence through contextualization

Translation quality alone makes ChatGPT an unreliable translator for business or customer service purposes. But adding contextualizing technology can provide much-needed background information to produce better, more relevant translation results.

Some instances of implementing contextualizing technology include using: 

  • Industry-specific glossaries to help ChatGPT understand industry terms and jargon.
  • Machine learning to aid ChatGPT in grasping the context of a conversation and producing more precise translations.
  • Customizable language models that can be fine-tuned to align with specific industries or business domains, enhancing ChatGPT’s ability to translate specialized terminology and maintain consistency in communication correctly.

By layering contextualizing technology on top of ChatGPT, businesses can improve the accuracy and relevance of their translations, leading to improved multilingual support and heightened customer satisfaction.

ChatGPT is a powerful tool, but it is not a silver bullet for translation. Organizations cannot fully replace the human ability to contextualize, decipher and truly understand the intricacies of human language, especially in complex business interactions. By combining the strengths of technology with human expertise, organizations can ensure their translations are accurate, culturally sensitive and contextually appropriate.


Heather Shoemaker
Heather is the co-founder and CEO of Language I/O, a Cheyenne, Wyoming-based neural machine translation company. Language I/O’s technology dynamically selects the NMT engine that will best translate a given piece of content and imposing company-specific terminology onto any of the many NMT engines integrated into the Language I/O cloud solution.

Weekly Digest

Subscribe to stay updated

MultiLingual Media LLC