Last week, I spent about twenty minutes identifying cars, buses, roads, and storefronts from pixelated images to log in to an important online account. Now, why on Earth would someone do that? Well, because reCAPTCHA, the free service that offers protection against intrusive, automated queries thought I wasn’t human! As the lines between human and artificial intelligence blur, it’s time to take a step back and gauge where we are headed.
In a world powered by technological advancements, it is only natural for artificial intelligence (AI) to have taken the front seat. While it is often more cost-effective, easily accessible and convenient, there are some areas where AI just doesn’t cut it, and Language Translation is one of them. Now, I don’t say this with any deeply rooted opposition towards AI or technology in general. AI has been helpful in achieving mammoth tasks like missions to outer space, automation in various industries and it’s also the technology behind the computer or mobile device you’re currently using to read this. There is no denying that AI is indeed an integral part of our daily life, and rightfully so as long as it is a means to an end.
When it comes to language translation, the most important criterion in testing accuracy is contextual relevance. When something is being translated from one language to another, the context must be preserved, and the meaning shouldn’t be lost. That’s just the tip of the iceberg. Language is very closely linked to the culture and region where it is used, and those nuances are as important as the meaning of the individual words being translated. If that weren’t the case, most of us would still be speaking only Latin and Greek. Machine language translation gets even trickier when we bring in languages that can be both formal and informal or include nouns that are masculine or feminine.
Let’s talk about the two languages I know very well: English and Gujarati (my mother tongue, spoken in the state of Gujarat, India). Here are some simple sentences that have been translated from English to Gujarati and their interpretations:
|English||Machine Transaltion||Interpretation||Human Translation|
|Where is my spoon?||મારા ચમચી ક્યાં છે||The word for ‘spoon’ in Gujarati is feminine, and so as a rule, the possessive pronoun ‘my’ should also be feminine. In the machine translation, the pronoun has been made masculine, which is incorrect.||મારા ચમચી ક્યાં છે|
|Close all the doors.||બધા દરવાજા બંધ||The translated text reads ‘all doors close.’ The verb ‘close’ is present, but the imperative form which is used to give instructions is omitted.||બધા દરવાજા બંધ કરો.|
|Please close all the doors.||કૃપા કરીને બધા દરવાજા બંધ કરો||When ‘please’ is added, the sentence translation is correct. ‘Please’ is just an additional word added to soften the imperative form and is not essential to the sentence formation, however, the sentence translation changes drastically.||કૃપા કરીને બધા દરવાજા બંધ કરો|
|There are three boxes in the kitchen.||રસોડામાં ત્રણ બૉક્સ છે||Here, the word ‘boxes’ is not translated even though there is a specific word in Gujarati for box/boxes.||રસોડામાં ત્રણ ડબ્બા છે.|
|You’re my younger brother.||તમે મારા નાના ભાઈ છો||In Indian culture, language may be formal or informal based on the relationship shared and the age of the individuals communicating. As a rule, the informal case will be used with someone who is either very close to you or younger than you. The sentence here satisfies both those conditions, but the translated text is written in formal Gujarati.||તુ મારો નાનો ભાઈ છે.|
As can be seen, machine translation usually cannot process all the exceptions and irregularities present in a language. There are a couple of reasons for that:
Machines follow the same process every time – AI first analyses the structure of the sentence. The sentence is broken down, each word is translated, and then the sentence is composed again, retaining the same syntax or structure. AI fails to take into consideration the fact that sentence structures vary in different languages.
Machines carry out word-to-word translation – AI performs literal, word-to-word translations. This can result in serious consequences since the meaning may be comprehensible but the translated content won’t sound natural. Additionally, there may be grammatical errors owing to which the sentence wouldn’t retain its original meaning. Moreover, some words can have several possible meanings based on the context in which they are used.
Benefits of Human Translators – Human translators triumph over machine translation because of their ability to differentiate and process all the possible sentence structures and usable words to find the combination that is best suited in a particular context. Like in the example described earlier, my knowledge of Gujarati as a native speaker helped when I needed to choose between the formal and informal case. A machine, unless explicitly told, cannot understand the closeness of two individuals who share a dialogue. AI may be able to translate content on the surface, but it still cannot localise it accurately.
The power of AI ensures tasks that humans cannot perform, become a reality. However, that takes me back to my original argument that AI should be a means to an end, not an intrusive form of technology that quickly replaces humans, especially in an area as personal and culturally significant as language and translation. A smarter, more accurate way to use AI in translation is to allow it to assist native, human translators. One popular way is through the use of translation memory which is a database that contains frequently translated words. The translator only has to translate unique words and many common words like articles and pronouns that have been translated earlier can be easily available. Several professionals successfully achieve absolute accuracy in translating an array of documents at very competitive prices. They may not provide you with instant results, but if you can wait for 24 hours, they’ll definitely save you the embarrassment of referring to someone senior with less respect than is due or, even worse, referring to a lady as a gentleman!