I Tried Translating with These AI Chatbots, Here's What I Found
How are language comprehension and translation challenging for AI chatbots? Check out their LLMs.
One thing I can say about translation is that it used to be a human’s job, and still, the translation process can’t be completed without humans.
So far we’ve seen artificial intelligence (AI) from chatbots to app integrations perfecting the language effortlessly in little to no time if provided a prompt.
Behind those auto-generated pieces of writing, large language models (LLMs) are developed to form those words in linguistically correct order, sometimes surprisingly creative and beautifully put, may I add.
That said, it improves technological capability on a broader scale while enhancing customer and user experience for efficient communication and task execution.
Chatbot’s Language Recognition
I think ChatGPT misunderstood my intention when I asked how it would respond with sarcasm to people who love pineapple on their pizza. But at least ChatGPT, among many other chatbots, can devise a deliberate joke or a pun in a few seconds.
Not every chatbot has the same sense of humor, nor do they have the same language model.
Large language models are engineering-based
They are not polyglot…technically. At least they don’t learn it the way we usually do.
These multilingual AI chatbots though provide an output with a supposedly practical language-based structure, it is, in fact, a result of sequence predictors and embedded templates with little to no connection to linguistics.
Large language models behind these chatbots are a type of AI algorithm trained to recognize, translate, and generate output in different languages and are available in multiple forms.
Their massive datasets are designed to perform Natural Language Processing (NLP) tasks and excel in language comprehension and generation.
The text is divided into individual units called “tokens” and is later transformed into an easy-to-digest numerical representation for a computer system called “vectors”.
Fed by different groups of developers, offered by different organizations.
Translation is a form of art
Translation requires cultural awareness and localization, most of which still need to be established and frequently updated.
You can’t deliver a joke to one person and expect another to react the same way. While an Australian may consider it offensive, another British may interpret it as a casual dad joke you’d make at dinner.
Then here go the questions—How formal should it be? How can this particular phrase be translated to cover both literal and metaphorical meanings? How much of the content can be omitted?
You can add in details when creating a prompt to ensure an appropriate response. I did try specifying an audience and implying what kind of platform would this excerpt be used. All of them still fail to adjust.
However, so far I could say that these five chatbots are surprisingly impressive but not yet decent.
I'll give it a nod
Writesonic seems to stand out the most, followed by ASAP, while Bard directly admits the excerpt is pulled from the website, with the source attached.
Each one has its forte, obviously. But translation-wise, some of these need room for improvement.
Since these multilingual chatbots are integrated to enhance customer service accessibility and efficiency, they serve users and customers of small businesses or world-leading companies with human-like conversations via chatbot interface.
They still might not be translation expert yet, but for general and corporate use, they sure are internationally preferable.