One thing I noticed about this past year of pandemic has been the increased importance of weather forecasts. We had a very small group of friends that we've regularly met with, but only around the fire in our backyard or at the beach. So it became very important to rely on hourly weather forecast apps like weather.com to see whether it was reasonable to expect an hour or two of dry weather between rainfalls (Q: Rainfalls? A: Oregon coast!). Overall, I've been incredibly impressed with the accuracy of the forecasts, and for months I've been planning to once again write something about meteorologists' use of data processing and artificial intelligence in comparison with translators.

But then last week happened. To celebrate our 30th wedding anniversary (👏👏 Thank you! 🙇‍♀‍ Thank you! 🙇‍♂‍), my wife and I planned to take a trip up the coast to places we hadn't visited before. We were really excited about our upcoming break but less excited about the weather forecast, which ominously predicted 100% rain throughout the week (again, Oregon coast!). We each packed several stacks of books, expecting to spend most of our time indoors, but I didn't even finish the first book I started because the weather was glorious.

Clearly my little anecdote contributes diddley-squat to a reasonable discussion about the state of meteorology, but doesn't it remind you of the state of modern machine translation? In my opinion, it's not completely arbitrary to compare the two.

Back in 2012 (in edition 214), I mentioned the similarities between how meteorologists and translators can rely on computer-generated data to form the basis of their work products, which in both cases have to be refined by humans. I cited an article by well-known statistician Nate Silver talking about "literally countless other areas in which weather models fail in more subtle ways and rely on human correction." If that doesn't sound like machine translation, I don't know what could.

Interestingly, just like MT, the computer systems and algorithms used by meteorologists today are complemented by narrow AI. While it was only a massive IBM super computer that crunched the many bits of data to come to reasonable suggestions (ever wonder why The Weather Company/weather.com was purchased by IBM in 2016?), those results are now complemented by AI algorithms developed by companies such as Microsoft and Google -- and of course actual meteorologists. And together they produce amazing results -- until they don't.
Again, there are myriad areas where the parallels to translation don't quite line up, but there are also countless where they do. The most pertinent seems to be the complexity of both language and weather.

Although humans have less complete insight into weather than language, computers ultimately have less in both.