Trust But Verify
I’m a child of the Cold War, a less-than-zero Gen X’er. So, please excuse references from the 80’s. When it comes to automation in translation (LLMs, NMT, TMS…the whole alphabet soup of technologies), it’s an arms race. When it comes to using AI technology in translation production, a proverb invoked often during the Cold War applies: “доверяй, но проверяй”–“Trust but verify”.
The reality is that those LSPs who deploy effective, modern technologies the fastest and derive the benefits (increased productivity, faster throughput) will profit. Others may survive but languish.
Fear the Future?
At this year’s Plunet Summit in Berlin, there was a panel discussion on The Future of Technology in the Language Industry. The panel of five eminent individuals in the language industry were asked if they are frightened or excited about the future. Overall, their perspectives were a mixture of both, but what I took away from the discussion is that our industry is slow in adopting these new technologies. Obviously, technology adoption is not evenly distributed. Depending on where a company is positioned, what the primary source of their revenue is, what their technological ability is like, and how many resources they can commit to developing new ways of working, all impact the rate of technology adoption.
If your firm works primarily with hard copy source documents (yes, there are still loads of these in the world) versus digital marketing content, this will inherently drive which technologies may be of importance to you. OCR automation for the former and content management-TMS connectivity for the latter.
Play. Get your hands dirty!
The most common advice to LSPs that I hear from (other) industry pundits is: “Get your hands dirty. Start playing with ChatGPT and other generative AI tools.” This is not bad advice, but I recommend that owners and managers of language service companies use this as an opportunity to assess. Deploying MT into production (either using neural MT or an LLM source) must still be evaluated. Is the potential solution going to bring the benefits you hope for? Is it secure enough for your client’s purposes? Do you have the right humans to put in the loop? (That’s the “trust but verify” part!).
Move strategically
I’m a big fan of approaching this assessment strategically. Do you have customers with a high volume of work and tight turnarounds? Do you have reliable language assets already in place? Translation memory is critical, but what about terminology? Would customizing an NMT engine or Large Language Model get you even better results? Customization may be a step too far for most smaller LSPs. Such work is not something you can throw at capable project managers without training and a significant change in job description. The truth also is that most LSPs don’t have enough data to use for training to make a significant difference in the quality that an engine can provide.
Customers: Follow or lead?
For most smaller LSPs, what draws them to deploying NMT, for example, is price pressure. Customers feel the pinch of inflation like the rest of us and expect the lowest possible cost for their upcoming translation project. For them, their project is big, to us it is just another 150,000-word project. Will using NMT make a difference, can you provide them with the price and quality they want and still be profitable? I am not an enthusiastic fan of deploying a new solution as a “one off”. If I’m going to force my production team to change how they work, it better be for a good strategic reason.
Company owners and Operations Managers are better off adopting innovative technologies deliberately at their own pace and to support their company’s goals. If you are being pushed to adopt a specific technology just to win a given project that is not the optimal way to improve your company’s performance. If customers are pushing us to change, it’s a clear sign that we are behind the curve. The best companies anticipate what their customers will want or lead them to where they need to go.