Text translation services have broken barriers in accelerating content consumption and communication around the world. With the advent and rapid improvement of natural language models and translation applications, translation is now an easy task. Unsurprisingly then, there’s a constant comparison of the various translation applications out there, the most popular being ChatGPT vs. Google Translate.
However, ever since ChatGPT gained popularity, people have also tested its language translation capabilities. Using ChatGPT for translation has been an interesting application of AI-based language models. Let’s investigate the nuances of artificial intelligence translators to understand who wins between ChatGPT vs. Google Translate, and how we can use ChatGPT for translation to its best capability.
Firstly, What Does a Good Language Translation Look Like?
Well translated text does not just imply word-to-word translation from one language to another. It should also account for the context, lexical semantics (i.e. word relationships) and the intended meaning of phrases. A piece of well-translated text will identify the use of idioms and metaphors that are native to a particular language and accurately convey the meaning of the same in the translated language rather than just providing a literal translation for these phrases.
In other words, text translation should convey the meaning of the entire body of text as a whole.
However, this is not an easy task.
Language translation has its unique challenges, especially when it comes to machine translation. Machine translation is tough because all languages aren’t the same in their approach to constructing sentences.
For instance, in languages such as French, pronouns exhibit gender distinctions, whereas in Japanese, pronouns are mostly left out.
Likewise, in Chinese, distinctions between singular and plural nouns are minimal.
However, in English there are noticeable distinctions between singular and plural words.
Moreover, there’s the matter of context and informal language usage. Machine translation tools struggle to accurately interpret the context of a statement. A statement that holds a particular meaning in one language might convey a different meaning in another. The challenges continue with colloquial expressions that incorporate proverbs, idioms, and linguistic creativity. These aspects can pose difficulties when attempting translation.
ChatGPT vs. Google Translate: Where does ChatGPT win?
Let’s take an example of a commonly used English idiom “through thick and thin” which means “under all circumstances, irrespective of the difficulty”. You can use it in a sentence by saying – “My friends have been by my side through thick and thin.”. Let’s ask Google Translate to convert this phrase into the Spanish language.
(Source: Google Translate | Using Google Translate for language translation gives us a word-to-word mapping of the phrase from one language to another)
In the above image, we see that Google Translate directly translated every word of the phrase into the corresponding word in Spanish and gave the feminine and masculine words for the adjectives “thick” and “thin”. Google Translate did not really understand the context and the meaning of the phrase while translating it. Let’s now try using ChatGPT for translation of the same phrase.
(Source: ChatGPT | Using ChatGPT for translation shows a better quality of machine translation results by understanding the context and the meaning of the phrases.)
In the above example, we see that ChatGPT can not only understand the meaning of the original input phrase in English, but also translate it accurately in Spanish. It also allows the user the flexibility to ask for the meaning of the phrase in the same prompt. This is a great example of an artificial intelligence translator understanding natural language well and providing accurate results for machine translation.
Let’s look at another example of using these AI translators to give us the English translation for a common Chinese phrase. This phrase literally translates to “In this hypocritical society, true personality is a rare and precious asset.”
(Source: Google Translate | In this case, using Google Translate for language translation of a complex Chinese sentence gives us an accurate and comprehensive result.)
When using ChatGPT for translation of this phrase, the following results were obtained.
(Source: ChatGPT | The ChatGPT translation quality of this Chinese phrase was also accurate and retained the meaning of the sentence.)
Both artificial intelligence translators gave similar results.
Let’s take another example for investigating the ChatGPT vs. Google Translate debate. This time, we consider Creoles, which is the language spoken and used by an ethnic group in the Caribbean. It is a form of Pidgin English i.e. a mixture of English and local languages which enables people who do not share a common language to communicate. This language has words from the English language along with other languages.
(Source: Google Translate | In this case of machine translation from Pidgin English to English, Google translate detects the input language as English and returns the translated text as is.)
Checking on ChatGPT’s translation quality – the AI translator did a good job in understanding the text and translating it accurately. ChatGPT also detected the language accurately and explained the origin of the same, when prompted.
(Source: ChatGPT | Using ChatGPT for translation of Creole English text gave better accuracy as compared to Google Translate)
What Contributes to ChatGPT Translation Quality?
These examples show that using ChatGPT for translation not only gives better results in most cases but also provides additional benefits of explaining the phrase or the input sentence. This is because the artificial intelligence translator allows the users to design their prompts according to their needs. Google Translate is a dedicated service for language translation and has limited capabilities.
Let’s elaborate on ChatGPT’s translation abilities:
- ChatGPT is an artificial intelligence translator with natural language capabilities beyond language translation.
- This means that ChatGPT comprehends context better than Google Translate and can produce answers that integrate the provided context.
- This enables it to manage intricate language situations and produce consistent translations, while Google Translate primarily emphasizes isolated sentences or phrases.
This gives ChatGPT the advantage in the ChatGPT vs. Google Translate debate in terms of understanding context and the lexical semantics.
The special ingredient that makes ChatGPT unique is RLHF – Reinforcement Learning with Human Feedback.
In this process, the artificial intelligence model gathers information about human inclinations towards model-generated responses, considering aspects like accuracy, lack of harm, usefulness, and so on. By taking human preferences into account, the system can choose responses that are better suited to different cultural contexts, particularly beneficial for those who are not native speakers.
ChatGPT also has this ability to customize the nature of the translated text on the basis of specific word-mappings or domains (for example: medical, legal) that the user can mention in their prompts. For instance, in the below image, we provided certain words that we wanted the translated text to have, to make it sound more professional in the legal jargon.
(Source: ChatGPT | USing ChatGPT for translation allows you to enter specific word mappings to customize the tone of the language translation.)
So you can enter custom term lists and translation memories in ChatGPT for this customization. This flexibility is not offered by Google Translate.
Moreover, Google Translate and other conventional machine translation systems cannot account for variations in language based on regions. For example, they don’t distinguish between Canadian and European French, Brazilian and European Portuguese, or Mandarin and Cantonese Chinese. However, ChatGPT is capable of doing this. Here’s an example:
(Source: ChatGPT | You can use ChatGPT for translation of text into different regional versions of the same language.)
ChatGPT vs. Google Translate: Where does ChatGPT lack?
Despite the impressive ChatGPT translation quality, there are a few caveats in how it operates.
The first and the easiest factor is the ease of access of Google Translate. You don’t need to login to a portal or create a specific account for accessing Google Translate for language translation.
Google Translate provides a specialized user interface and seamlessly connects with other Google services, enhancing its user-friendliness and convenient accessibility. However, to use ChatGPT for translation, you have to create an account. However, this account is free to create and maintain.
Moreover, since ChatGPT has been trained on a huge corpus of historical data, there is a chance of errors and inconsistencies creeping in. It is not a fail-proof tool. Moreover, if there isn’t enough content on the web on a particular language, say the Cambodian Khmer language, the artificial intelligence translator will never be able to learn and to pre-train itself on this language. This would lead to inaccurate or incomplete translations.
Additionally, Google Translate functions as a specialized language model designed specifically for machine translation purposes. Given its specialization, Google Translate proves effective for rapid and uncomplicated translation needs, particularly in cases where utmost precision isn’t the central requirement.
The user-friendly nature of Google Translate, coupled with its seamless integration with various Google services, renders it a practical choice for day-to-day translation tasks.
Also, since it is a simpler machine learning model, it is faster and more computationally efficient to run when you need a framework for machine translation of languages.
ChatGPT vs. Google Translate: Will ChatGPT replace Google Translate?
Our final verdict on ChatGPT vs. Google Translate is that ChatGPT cannot fully replace Google Translate as a machine translation tool, owing to the ease, speed and convenience of the latter. However, it definitely has better performance in terms of understanding context and language nuances which Google Translate lacks. This would require Google Translate to work on its current language model to incorporate these cases and language nuances.
With Google’s current work in developing its large language models, it should improve the performance of its machine translation tool to match, or even surpass the current ChatGPT translation accuracy. Perhaps then the whole ChatGPT vs. Google Translate debate will have different results.
