The crucial importance of context in translation: why AI may yet struggle

When we talk about context in translation, we’re referring to all the information which surrounds a text or discourse. This includes linguistic context (the words and phrases which precede and follow), situational context (where and when the text is used), cultural context (cultural references and norms) and even historical context. All these elements influence the way we understand and interpret language.

Let’s take a simple example: the word “tableau” in French. Without context, it could mean a painting hung on a wall, the blackboard (or whiteboard) in a classroom or a table of data. Only context will provide us with the correct meaning. It’s these kinds of nuances which human translators handle intuitively, but which can pose problems for AI systems.

AI translation systems, such as those used by Google Translate or DeepL, have made remarkable progress in recent years. They can translate entire texts in a matter of seconds, often with impressive accuracy. However, these systems remain fundamentally limited in their understanding of context.

Anecdote: the translation which almost started a war

A famous anecdote perfectly illustrates the importance of context in translation. In 1956, at the height of the Cold War, Soviet leader Nikita Khrushchev gave a speech containing the phrase “Мы вас похороним” (My vas pokhoronim). Literally translated, this phrase means “We will bury you”. The American press, influenced by the tense context of the time, interpreted this as a direct threat: “We will bury you”.

In fact, Khrushchev was referring to a Russian expression meaning “We will outlive you” or “We will attend your funeral”, implying that communism would outlive capitalism. The lack of understanding of the cultural and linguistic context led to escalating tensions between the two superpowers.

This translation error, which could have had catastrophic consequences, would probably not have been avoided by an AI system of the time, and even today’s systems might struggle to grasp all the nuances of this expression.

The specific challenges of AI in translation

1 • Lexical ambiguity

One of the biggest challenges for translation AI is lexical ambiguity. Words can have several meanings and only context can determine the correct one. For example, the English word “bank” can mean a financial institution or the bank of a river. An AI system could easily make a mistake without a thorough understanding of the context.

2 • Idiomatic expressions

Idiomatic expressions are a real headache for AI. Take the French expression “Il pleut des cordes”. A literal translation into English would be “It’s raining ropes”, which makes no sense. A human translator would know to use the English equivalent “It’s raining cats and dogs”. AI systems have made progress in this area, but they can still make embarrassing mistakes.

3 • Cultural references

Cultural references are particularly difficult to translate, even for humans. How do you translate “metro-boulot-dodo” (a common phrase in French which refers to the daily grind of work) into English or Chinese while retaining the same cultural flavour? AI, with no real-life experience of culture, can easily miss these nuances.

4 • Historical background

Historical context can also influence translation. For example, translating the word “comrade” in a text about the Russian Revolution requires an understanding of the historical context that AI may find difficult to grasp fully.

Despite these challenges, translation AI continues to make progress. Neural translation systems, which use artificial neural networks to learn how to translate, are increasingly able to take context into account. They analyse whole sentences, and even paragraphs, rather than isolated words.

However, even these advanced systems have their limitations. They can’t really “understand” a text in the way a human would. They rely on statistical models and patterns learned from vast text corpora, but they lack any genuine understanding of the real world.

Conclusion: the enduring importance of the human translator

Although AI has revolutionised the field of translation, making it possible to rapidly translate large quantities of text, it cannot yet completely replace the human translator. Deep understanding of context, cultural sensitivity and the ability to interpret the subtle nuances of language remain the prerogative of the human mind.

For critical translations, whether of legal documents, scientific papers, literature or diplomatic communications, human translators continue to be paramount. AI can be a valuable tool to assist translators, but it cannot yet capture all the richness and complexity of human language.

Ultimately, translation remains as much an art as a science. It requires not only an in-depth knowledge of languages, but also a nuanced understanding of cultures, contexts and the subtleties of human communication. It is this unique blend of skills which makes translation a field in which humans remain, for the time being, one step ahead of the machine.

Discover AIEdit, our solution that combines the power of AI with the irreplaceable expertise of human translators.