AI and the future of translation. In search of evidence

This talk is dedicated to the members of Translators and Interpreters Australia, an organization deeply invested in safeguarding the rights, well-being, and compensation of translators and interpreters. Today's focus is artificial intelligence (AI) and its potential ramifications on our professions.

There is a lot of chatter going around—instant experts popping up everywhere predicting catastrophic scenarios for humanity, the environment, and our work in communication and language quality. Amidst this whirlwind of predictions, the critical question we ought to ask is: how much of it is actually true? It seems essential for us to dig deep and search for evidence.

Some might remember the chaos surrounding the millennium bug—the fear that all computers would malfunction because nobody had updated the date. A plethora of self-styled experts emerged, spreading fear. This tendency toward negativity among experts, I believe, stems from a basic risk management strategy: if they’re wrong, nothing bad happens, and if they’re right, it's ‘I told you so.’ It appears that a similar phenomenon is at play here.

Let's kick off this investigation by clarifying the AI technologies we’re discussing. Most of you have likely heard about ChatGPT, which became publicly accessible in late 2022, and GPT-4, which you can play around with as well. I’ve been experimenting with both ChatGPT and Google Bard, which is currently in experimental mode in Australia. ChatGPT has the fastest-growing user base in history for any app, and I hope you’re all having some fun exploring these technologies.

Initially, my foray into this realm was filled with shock and awe. It can translate and provide information across a broad spectrum, which is undeniably impressive. Is this the singularity moment when AI surpasses human brain processing capacity, similar to what we've seen in games like Go and chess? Perhaps it is; perhaps it extends to translation and interpreting too. The truth is, we still do not have all the answers.

However, my subsequent interactions with these technologies reduced my initial fears. An insightful approach is suggested by Sasha Luccioni, who refers to the “iceberg model.” This model recognizes that while there are clear benefits to new technology, lurking beneath the surface are a myriad of threats. The threats include environmental impacts—like significant carbon emissions—and concerns relating to representation, privacy, and copyright infringement.

We must confront the notion that these threats do exist, alongside potential advantages. Lynne Bowker, a specialist in machine translation literacy, echoes this sentiment and adapts it to translation. Since the advent of neural machine translation (NMT) in 2016, we should analyze its available advantages and disadvantages. Yet, the intriguing point is that since 2016, concrete evidence supporting the growth of these negatives is challenging to pinpoint. This will be my focal point as we proceed.

Let's dive into some critical areas of concern:

• Lack of transparency
• Job destruction
• Changing skill sets
• Anchoring effects
• Language marginalization
• Wage dispersal in the industry
• Intellectual property concerns

First and foremost, let’s clarify a recurring problem around us, particularly relevant to the Spanish and Catalan languages. Subtitlers recently raised alarms over the Netflix series "Squid Game," which involved post-editing of machine translations. Translators faced machine-generated translations that required corrections. The head of the Audiovisual Translators Association in Catalonia even spoke out about the massive threat such processes seemed to pose. Going through her talk, one thing became evident: there was a mix-up. She conflated post-editing with raw machine translation, muddling the distinction between the two processes.

Hence, the relevant question we should ask is: is it more advantageous to translate from scratch without electronic aids, an approach that few of us actually adopt, or to pass a text through machine translation for correcting? When I teach translation, I avoid biasing my students towards a method; I encourage them to experiment, measuring results in terms of the time spent and the quality of the final product assessed through peer review.

With tools like GPT and Bard, while they can translate, the quality this process yields is not always superior to offerings from systems like DeepL or Google Translate. It's essential to conduct your own tests to calm any preemptive nerves.

The impressive aspect of new technologies, including GPT, lies not only in direct translation but also in assisting with tasks typically managed via translation memory software. They can do an array of things, from helping refine previous translations to streamlining different machine translation feeds, assisting in terminology management, and enhancing the collaborative aspect of translation work.

Let’s consider some additional functionalities:

• Text simplification and grammar correction.
• Speech-to-text capabilities for more natural-sounding translations.
• Aligning text with old translations to create a translation memory for future use.
• Providing explanations for terms in specific contexts.
• Creating a more interactive translating experience through adaptable prompts.

This leads us to the exciting concept of “augmented translation.” It suggests that rather than being replaced, translators can extend their capabilities through technology. I wholeheartedly support the conclusion drawn by translators across Europe that while automation presents challenges, rejecting the benefits of automation isn't the answer.

Yet, the desire for transparency in the use of AI technologies must be highlighted. For example, if a translator has employed multiple dictionaries during their work, should they disclose this in their published translations? Why not employ this same transparency with machine translation systems? I’ve recently committed to being transparent about my processes in a collaborative project, identifying machine translation systems utilized and clarifying how these tools supported my contributions.

It’s critical to highlight that if you believe your work has been misappropriated due to AI, consulting an attorney could be vital. However, it’s fascinating to note that while instances exist of translators suing companies over copyright infringement, no such cases have emerged directly involving AI.

So, does automation destroy jobs? A recent report highlighted the declining employment of translators in the European Commission, yet the underlying cause is outsourcing, not automation. The budget for in-house translators has actually increased in recent years. Therefore, the overarching narrative that automation leads to job loss in translation remains unsubstantiated.

Moreover, examining the effects of neural machine translation—specifically the changes in the job market from 2016 to 2019—reveals that while there was a noted decline in translation demand, roles requiring non-routine cognitive skills did not see the same decrease. Those equipped with the skills to utilize new technologies and adapt to the evolving demands of the industry fared far better.

Despite fears of job destruction, the increase in globalization indicates that the amount of translated material is on the rise, allowing us to maintain our relevance amid changing landscapes.

However, it's acknowledged that larger languages tend to benefit more from automation than lesser-known languages, a disparity that begs for more resources and support for under-resourced languages in the future. The solution may not be less technology but rather amplifying technological assistance in developing smaller languages.

While there's growing evidence that AI tools perform better with widely spoken languages, it’s necessary to ensure these technologies continue advancing and expanding support for those smaller languages to keep pace.

In conclusion, despite concerns of trending wage disparities and intellectual property problems exacerbated by larger corporations’ practices, it’s crucial to take a measured approach toward emerging AI technologies in our professional landscape. We must focus on collaborative work and transparency within our industry, and instead of succumbing to panic, we ought to foster experimentation and research.

I have yet to find definitive evidence that AI is a dire threat—not to translators, interpreters, or humanity at large. Hence, let's remain vigilant yet calm as we explore and adapt to the changes AI brings to our professional sphere.