There was a certain period when our children were little, when I would be waiting impatiently for Peter to come home so I could run to a meeting, and as I ran out the door I told Peter, “There is an invisible idiot on the second shelf in the kitchen.” This provided Peter with the information he would need about where to find a pacifier when it was time to put the boys to bed, but without reminding them of the pacifier, so they wouldn’t ask for it before bedtime, because we had reached the point where pacifiers were only used for going to sleep.
The code name “invisible idiot” for a pacifier came from an article I had read about advances in machine translations. The article described an example of running the phrase “out of sight, out of mind” through a machine translation, which came out as “invisible idiot”. At the time, I thought it was an impressive achievement, even if not entirely accurate, which also assured me that I would not be made redundant as a human translator any time soon.
That was over twenty years ago, and advances in machine translations are even more impressive since then, but in the six weeks I have been living in Calafou now, as I have been using machine translations quite extensively, I find myself frequently reminded of the invisible idiot. Often I look at the translation of message exchanges, and I imagine that at some level this statement probably makes sense to a machine, and it may create an interesting meaning of its own, but it is probably not exactly what a human meant.
Since I avoid using Google services wherever possible, I have been using primarily the open source alternatives DeepL for Spanish to German translations and Apertium for Catalan to English. Since Apertium can switch between Catalan and Spanish, this is also useful for mixed-language exchanges. If all else fails, an odd Romanian site with a nineties-style interface can usually decipher the most confusing messages. On my phone I couldn’t get the mobile version of DeepL from F-Droid to work, so I ended up with a commercial application, which frequently annoys me when I’m trying to paste a text and accidentally click on an ad. When it is right, this app is very, very good, so it annoys me to have to admit that a commercial app can be that good, but when it is wrong, it is very, very far off, which can result in unpleasant surprises. When it translated a comment last night as a seriously questionable sexual innuendo, I decided to wait for the electricity to come back on and check it on one of the desktop apps, because that is certainly not the way people communicate here. Of course that also raises the question of what is going on in the background that would lead a machine to read a sexual innuendo into a simple remark. If machine translations improve through machine learning, what exactly has this machine been learning?
As I am also working hard on learning Spanish (I have currently reached the communication skills level of a two-year-old), sometimes the mistakes in the machine translations provide me with useful clues about how Spanish works. The most frequent confusion arises from an inability to distinguish between he, she and it, which tells me that it must be the same word in Spanish, when a person’s gender keeps unexpectedly switching from one sentence to the next. I have also learned that machine translations can be confused by proper names. When I see messages with mysterious references to fleas, frames, sweets, etc., it is usually a sign that I need to check the original message for people’s names that I recognize in between the other words. Since it is also often confusing for humans that one of the dogs here is named Muerte, I can’t fault the machine translations for not recognizing this word as someone’s name, but it is still amusing to be informed that “Death will be de-wormed the end of October.” I should probably be more appreciative that the machine translation was able to help me with the word for “de-wormed”, which was the only one in that sentence I didn’t understand. In light of my extensive involvement in the past years with discourses relating to death, however, I still feel that this is simply a wonderful statement: Death will be de-wormed the end of October.
Unsurprisingly, perhaps, and apart from the recurrent problems with proper names and third person singular pronouns, machine translations generally seem to work better for email messages. When confusion arises there, it usually seems to involve references to spaces, practices or groups that are specific to Calafou. As I learn more and more about how things work here, these riddles become easier and easier to solve. I see this as confirming the expectation that people tend to write more formally (or at least in complete sentences using conventional grammar) in email, and I find myself wondering how this might relate to the widespread opinion that email is falling out of favor, especially with young people. I have seen that opinion expressed so often, and I still have reservations about it, but now I’m curious about a possible connection.
Also unsurprisingly, probably, the machine translations seem to find short messages posted to the Calafou Riot channels more challenging. For me, still struggling to keep up and figure out what is going on, sometimes this can be frustrating, but sometimes it is simply hilarious. One thing I have learned, as I am gradually able to identify who the various nicks belong to, is that it is sometimes more useful to simply imagine the person’s voice in my head in order to understand what they are saying. I don’t need machine translations in order to understand voices that have meanwhile become more familiar. And I find myself drawn deeper and deeper into questions about how communication really works.
When someone reached down to pet Ginevra recently, addressing her as “guapa”, my mind “translated” that to the familiar situation of someone reaching down to pet her in Austria and calling her “Du Hübsche” (roughly: “You pretty thing, you”). The same gesture, the same gentle tone told me what “guapa” means better than any dictionary or machine translation ever could. I’m learning that there is a level where humor can still be understood, even when you don’t get the joke because you don’t understand the words. And it is not (only) words that convey kindness, generosity and understanding.
As I was wrestling with machine translations of the Calafou wiki in an attempt to rewrite confusing (not only to me) instructions, somewhere on the periphery of my attention I noticed a friend in Linz posting about special local baked goods at the farmers’ market in Linz, which ended with a picture and the satisfied statement, “Owa an Bauankropfm hob i nu kriagt” (roughly: “But I still got a farmer’s doughnut”, i.e. after everything else was gone). Since it made me smile, the statement caught my attention, so on a whim I decided to see what a machine translation might be able to do with it. Not much. DeepL was at least able to identify the language as German, but the translation “Owa to Bauankropfm lifted i nu kriagt” is hardly useful, since the only two translated words are incorrect.
Somehow I found this exercise strangely comforting. As a human, I was able to recognize that this is simply a phonetic representation of the way people normally speak to one another in Linz. It is an entirely prosaic and unimportant, yet thoroughly familiar expression of satisfaction, and the familiarity of it has an emotional effect. It occurred to me then that this is essentially the same kind of thing I am reading on the Riot channels, which machine translations are unable to explain to me in any even minimally coherent way. It’s not the literal meaning of the words, but the feeling they convey that matters. Realizing this makes it easier to read/listen to the sense rather than the meaning of the words. It makes expressions I don’t understand somehow feel more familiar, because the people who use them – in their own language, in their own personal modes of expression – are becoming familiar to me.
Language has been my life for as long as I can remember. Words, spoken words, written words, playing with words, finding new words, different words, so many words – words in all their many forms and myriad meanings have always fascinated and preoccupied me. It is an unexpected gift in this late phase of my life that I now have a new opportunity to learn to find meaning behind, beyond, between the words.