The End of Translation Industry?

?? The article title looks academic but let's face it in a more informal way :

The bad news is that AI is slowly taking over. Whether you like it or not.

After the introduction of the IBM PC in 1981, we had just one year where copy-pasting was our utmost form of automation.

Since then, machine-assisted translation has seen three significant leaps in efficiency:

1. ??? Computer-Assisted Translation (CAT) Tools: Started with ALPS (Sausalito, California) released in 1982 and Trados (Stuttgart, Germany) in 1984 - this was the time when machines purely served us.
2. ??Neural Machine Translation (NMT): Gaining popularity around 2012, this marked a shift from previous purely statistical models. Many translation clients and agencies were happy to save on Post-Machine Translation Editing (PMTE) services, although many translators disliked it. The quality was often poor, and nobody likes editing bad translations unless they are a kind of translation masochist. Typically, you had to post-edit translations ranging from 'fair enough' to 'very very bad', and importantly, it did not improve over time. Only in very rare cases was the argument for a complete retranslation accepted.
3. ??Large Language Models (LLMs): 20 years later, it has become clear that machines can sometimes outperform 'good enough' translators and, with proper input and more or less standard topics, even excellent translators—at least in terms of speed.

Now for the good news. As you can see, there seems to be a 20-year cycle to which the translation industry adheres. We have roughly 20 more years before the next leap in translation technology—unless there is some kind of breakthrough—or before the invention of Artificial General Intelligence (AGI), which will take over all jobs in the world, including those requiring physical presence, as you can imagine from current demos of Atlas by Boston Dynamics, ASIMO by Honda, and other, 'less humanoid' robots like Xiaomi vacuum cleaners.

In translation, progress is slow because, unlike the human approach of taking socialization and learning from kindergarten to university and then to workplace practice, AI tech giants need big solutions fast. And while LLMs are quite good at translating 'generally', they lack logical understanding of what they are translating.

So, let's relax. We have 20 more years. )

Updated prediction is as follows:

• 1982: Introduction of ALPS (Sausalito, California)
• 1984: Launch of Trados (Stuttgart, Germany)
• 2012: Rise of Neural Machine Translation (NMT)
• 2022: Recognition of Large Language Models (LLMs)
• 2024: Start of AI Integration in Translation Workflows
• 2026: Predicted: AI Integration becomes as popular and standard as Computer-Assisted Automation
• 2042: Predicted: Fully Autonomous Translation Systems
• ???: Emergence of Artificial General Intelligence (AGI?)

And we will still proudly continue developing the line of of software products for translators at https://translation3000.com Long Live the Translation Industry!

Part 2 of the article will be published soon. Connect with me (Open profile) and /or follow for more interesting reads.

Updated on 22 June 2024.