Really? A sarcasm detector? Wow. You shouldn't have Researchers from the University of Groningen's Speech Technology Lab say they have created a multimodal algorithm that can detect sarcasm in speech. Because clearly the world's biggest problem was not having computers capable of rolling their eyes and saying, "Yeah, right." Xiyuan Gao, Shekhar Nayak, and Matt Coler are the …
Okay, pitch, no problem. That is purely mathematical and easy to determine.
Speaking rate, no problem there either.
Energy ? How is that qualified ? Someone screaming at the top of their lungs makes a lot of noise, to be sure, but someone looking you in the eye at less than a meter from you with a gun in their hand pointed towards you can softly say "DON'T. MOVE.", and there will be a heck of lot of energy in there that I don't see a computer detecting . . .
It's going to struggle with English sarcasm on energy.
My American and Canadian friends generally you can tell the sarcasm because of pitch and energy, they want you to know they are being sarcastic.
They struggle with sarcasm from Brits, Aussies and so on because we prefer to just deliver it dry, there's been many a time in a bar they didn't pick up on a comment for a good few minutes.
Feed it Peter Cook's summary of the Jeremy Thorpe case and see how it copes with what is probably the finest bit of satirical work ever presented. Delivered in measured tones and without emotion - just as good satire should be. Bet it fails to spot a single joke!
Thank you for the video link. Golden!
Not only dassic Peter Cook but also I remember the case at the time as I was puzzled how sergeant Wilson (Dad's Army) became involved in that unholy mess. (A different le Mesurier.)
"a pink oboe* player " - does depravity know no depths below which it will not sink?
* The Pink Oboe was the title of a 1959 Goon Show S9E11
This is going to work in England where notoriously the more flesh they intend to strip from their victim the more deadpan the delivery?
How on earth does this breakthrough hope to distinguish between irony and sarcasm?
Intonation varies between languages, dialects, and even different generations and various groups within a generation (to distinguish themselves from the others presumably) and even between genders (sexes.) The rising intonation at the end of a sentence which is not a question is now as prevalent as it is irritating. Purposely (or maliciously) placing stress in the middle of words heretofore have remained unstressed seems a more recent innovation. Moreover the misplaced stress appears to maximally disrupt the rhythm and flow of the sentence. I assume it is an linguistic attention seeking behaviour.
One mandatory training set would have to be sir Humphrey's dialogue from Yes Minister and Yes Prime Minister.
Yes, I can see this technology being one of the great breakthroughs of the 21st century and a massive boon to humanity generally relieving future psychotherapists of the tedium their routine consultations to concentrate or more serious cases.
I really do! (Pig's arse.)
The University of Groanagain going to knock over humour next?
We will follow their future endeavours with great interest.
"Acceptable! Not bloody likely mate!"
Look, you and I know that Aussie is not a British accent, but to an American...
I added it as "acceptable" because both of our countries have a fine tradition of taking the piss out of everything, so just assuming that anything said would be dripping sarcasm is the correct response.
As a Londoner living in Dublin, when I brought my American girlfriend to London, she heard an accent which wasn't Irish or English and asked me if it was American. It was Australian.
She had been brought up in the USA thinking of herself as pure Irish. After six months adjusting to Dublin, she was surprised to think of herself for the first time as American. All four grandparents from Ireland and she with pure red hair.
This is going to work in England where notoriously the more flesh they intend to strip from their victim the more deadpan the delivery?
I am sure it will work perfectly. Just ask the children. Ours made an excellent Sunday roast.
How on earth does this breakthrough hope to distinguish between irony and sarcasm?
I'm sure this is an area where the British can be of great assistance given we have never been known to use irony, nor sarcasm. I'm sure our education system will provide help and support, and the UK has a number of leading drama schools teaching our best and brightest to deliver appropriate levels of emotion. Might I suggest training your AI models on the latest series of that most British of family dramas, Dr Who? The most recent episode already contained cutting examples of edge AI developed by the brightest of this sceptered isle. An example of our sceptres was also included in this episode demonstrating how we can combine AI with an organic flash lamp.
:p
/sarc off. Hopefully this just means we Brits may be spared this latest AI offer. The whole thing is just jumping on the AI bandwagon in an effort to obtain more funds. I also feel kinda sorry for non-native English speakers who have to try and learn the idiosyncracies of both our language, and the way we use it. All I can say having worked & lived with assorted non-native English speakers is I'm glad I learned it as a kid.
"How on earth does this breakthrough hope to distinguish between irony and sarcasm?"
I rarely say anything that isn't irony or sarcasm but I live in America and few notice.
Americans will take me literally when I say "brilliant" although sometimes I can't tell if I am being sarcastic.
I hear you. Frenchies, also, can be so damn literal.
A few months back, at work, I mentioned that I used to be a pessimist and since everything I feared would happen did, I decided to switch to being an optimist to see if better things happened.
In the UK, that would get something between a smirk and a gentle giggle. Here in France? I had cow-orkers who weren't even in that conversation congratulating me on a more positive outlook. OMFG. No!
(and my day to day snarkiness? I think they just put it down as "weird Brit frequently says weird things")
Because clearly the world's biggest problem was not having computers capable of rolling their eyes and saying, "Yeah, right."Well, there's your problem right there. Obviously, you've been trying to make your car go by pushing the speedometer needle, haven't you? That's not how any of this works.
Methinks this is akin to equating polygraph tests with 'lie detection'. How will it detect the difference between someone who is demonstrably incorrect, but believes strongly in what they are saying (see, e.g., Liz Truss 'I was right, but brought down by the 'deep state' left wing establishment', Tony Blair: 'I still believe that invading Iraq was justified and the right thing to do', Nick Clegg 'I promise that we will not allow the coalition government to increase student fees')* and those who are genuinely taking the pi*s? (See former President George H Bush, who I strongly suspect was making a little jest when he said: "the French have no word for 'entrepreneur' ".)
A polygraph only records a few real-time changes in physical measurements, heart rate, sweating, etc., and these are/were interpreted as being distressed due to telling an untruth, because, well, everyone has a conscience, don't they? Sarcasm is all in the mind of the speaker and audiences
* Politically required to be balanced.
But, but,... Liz Truss, Tony Blair, Nick Clegg and all "good" politicians are sarcasm incorporated. Politicians have been trained to speak bullshit with a straight face. They are trained to sell you a lie without twitching. How can you tell politicians are lying? Words come out of their mouths.
Therefore, train the sarcasm detecting LLM on politicians' speak and you are sure to be right all the time!
The real prize for AI would be to detect irony, but that's going to be difficult since an awful lot of humans still don't get it.
I think the researchers, like many people, treat irony and sarcasm as equivalent terms. They are not.
Sarcasm is a blunt tool, with a definite intent to ridicule some victim - it sounds embittered and is normally perceived as quite offensive.
Irony is much harder to detect, indeed it is often missed, because it is on the limit of plausibility. The only possible victim is the self.
It is a deliberate risk, if it is not understood then the speaker has conveyed an incorrect impression that he/she is some sort of idiot, or bigot. It's difficult to undo. Tell me about it.
For those who might want a further introduction to irony, I have to recommend Kate Fox in her book "Watching the English": https://edisciplinas.usp.br/pluginfile.php/4434518/mod_resource/content/1/Watching%20the%20English.pdf
For those attempting to acclimatize to this atmosphere, the most important ‘rule’ to remember is that irony is endemic: like humour in general, irony is a constant, a given, a normal element of ordinary, everyday conversation. The English may not always be joking, but they are always in a state of readiness for humour. We do not always say the opposite of what we mean, but we are always alert to the possibility of irony. When we ask someone a straightforward question (e.g. ‘How are the children?’), we are equally prepared for either a straightforward response (‘Fine, thanks.’) or an ironic one (‘Oh, they’re delightful – charming, helpful, tidy, studious . . .’ To which the reply is ‘Oh dear. Been one of those days, has it?’).
I've posted before on the subject, https://forums.theregister.com/forum/all/2022/10/22/chrome_extension_howto/#c_4553678
Researchers from the University of Groningen's Speech Technology Lab say they have created a multimodal algorithm that can detect sarcasm in speech. Because clearly the world's biggest problem was not having computers capable of rolling their eyes and saying, "Yeah, right."
And Gullible isn't in the dictionary. Ja zeker!
Researchers trained a neural network on text, audio and emotional content of video clips from US sitcoms including Friends
Yes, really.
Heard one of the researchers being interviewed on the radio - the training dataset is somewhat small in terms of typical ML training dataset magnitudes.
Multimodal Sarcasm Detection Dataset (MUStARD)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
