Wednesday, July 19, 2017

Reading the Research: Translating Sarcasm

Welcome back to Reading the Research, where I trawl the Internet to find noteworthy research on autism and related subjects, then discuss it in brief with bits from my own life, research, and observations. 

Today's article details an achievement in translating social language, specifically sarcasm, into "plainspeak" as I'd call it.  System detects, translates sarcasm on social media

This is notable, according to the article, in the main because it's the first system to directly translate, rather than simply mark or note when a sentence is sarcastic.  So it would translate something being "the best" sarcastically into "the worst" unsarcastically, leaving the rest of the sentence untouched.  This is promising, but also somewhat worrisome to me. 

It's promising, because it's a possible starting point for autistic people to learn sarcasm.  Getting a start on flipping that mental switch from one extreme to the other can be difficult for people who automatically take things literally.  But it's worrisome, because it does not, in fact, allow for any learning possibilities. 

I do not recall precisely how my first real friend, a British citizen, tutored me in sarcasm.  I suspect it was probably more via excessive amounts of examples than actual direct teaching, but whatever the method, I am now quite fluent in speaking and translating it (unless I don't know the speaker at all). 

If I had been equipped with this translation technology, it would merely have translated every sarcastic sentence he typed, leaving me entirely unaware that he was being sarcastic at all.  At which point, I would never have learned this particular style of communication and would been poorly equipped to handle the real world, where sarcasm is readily and often used in verbal communication. 

I recognize that not everyone can simply pick up sarcasm via repeated exposure, but it worries me that there isn't even the possibility of learning with this technology.  I would favor an increasing difficulty model, starting with a direct translation like this provides.  After that, you could move to marking the sarcastic sentence, perhaps side-by-side with the original sentence, and then simply marking the sarcasm and letting the autistic person do their own "inversion of meaning" and try to understand it without help.  Eventually, you could leave off the sarcastic markers and hopefully people like me could simply get along by ourselves. 

Unrelatedly, the article calls the translated sarcastic sentences "honest" sentences.  I hope that's a translation eccentricity, because I rather resent the implication that I'm being dishonest by being sarcastic.  When I say things sarcastically, I do not intend to mislead people, but instead express my sentiments in a humorous, boomerang-like fashion.  So I hope the translation for the non-sarcastic sentences could also be "direct." 

It occurred to me, as I was reading the article, that this technology is still very much a proof-of-concept.  Their system would not, for instance, pick up my sarcastic use of the word "charming" to describe behaviors in people that are distinctively anything but charming.  For instance, any rendition of archaic viewpoints on women, including the irksome "make me a sandwich" or any reference to women being confined to the kitchen or being silent, tends to elicit that particular comment.  Particularly if I don't feel comfortable calling the person on their offensively repressive crap.  This translation system would miss that, as it seems to focus on superlatives like "best" "worst" "awesome" "awful" etc. 

Lastly, the article comments that this translation technology could be useful for communicating with computers.  I find this quite appropriate, since computers presently lack a sense of humor and the ability to take things figuratively.  That may change in the future, but for the time being it seems wise to research such applications for tech. 

No comments:

Post a Comment