Slashdot Mirror
Tech Companies Looking Into Sarcasm Detection
Nerval's Lobster writes "Now here's the greatest thing ever: French tech firm Spotter has apparently devised an analytics platform capable of identifying sarcastic comments, according to the BBC. Spotter's platform scans social media and other sources to create reputation reports for clients such as the EU Commission and Air France. As with most analytics packages that determine popular sentiment, the software parses semantics, heuristics and linguistics. However, automated data-analytics systems often have a difficult time with some of the more nuanced elements of human speech, such as sarcasm and irony — an issue that Spotter has apparently overcome to some degree, although company executives admit that their solution isn't perfect. (Duh.) Spotter isn't alone: IBM, Salesforce, and other IT vendors are hard at work on analytics software that can more perfectly determine when you're mouthing off, you little punks. In theory, sarcasm detection can help with customer service, and judging how well products are doing on the open market... and we all know it's going to work perfectly, right? Nothing could possibly go wrong with automated platforms built to assess the nuances of human speech."
I hope they get 75% of it right. My personal guess is that around 25% of humans are unable to detect any sort of sarcasm, perhaps not quite as bad as Sheldon, but quite bad.
Like that's going to work.
A feeling of having made the same mistake before: Deja Foobar
Sarcasm is nice, but what about detecting the effect of Poe's law?
If they make it work and ever point that at slashdot, the readings are gonna be flying off the charts!
(My turn)
Oh! I'm not being sar-cas-tic.
http://youtu.be/ziH9St7ajuw
crazy dynamite monkey
Gee, that's useful.
They're geniuses.
For all intensive purposes, "whom" is no longer a word. That begs the question, "who cares"?
Sarcasm is very frequently indicated by nuances that aren't transmitted through text. If humans have trouble getting sarcasm out of text, why should an algorithm do any better with the same set of data?
simpsons did it
There's no way they'll just leave it to languish once it's working. Corporations just love bad news and never ignore negative feedback.
That's a really useful invention.
Current best sarcasm-detection algorithms are barely 62% effective even with text that includes smiley faces or other obvious sarcasm tags.
this is where skynet decides to kill all humans, isn't it?
the preceding comment is my own and in no way reflects the opinion of the Joint Chiefs of Staff
Do we have a sarcasm sign?
What we really need is a lie detector, a spam detector, and a troll detector
Extra points for the spam detector, THAT is what is most sorely needed, and what is so inadequately provided thus far.
Elementary, my dear General, we just scan for the use of the French language!
Unfortunately, there's been a setback in the schedule. They tested it on Slashdot and it exploded.
to make sure I get my snark tuned exactly right.
if this is supposed to be a new economy, how come they still want my old fashioned money?
Right.
I've calculated my velocity with such exquisite precision that I have no idea where I am.
Often the detection of sarcasm relies on understanding of popular opinion on a topic. I don't think we'll have any magic bullet algorithm to detect sarcasm until we have hard AI with a far-reaching corpus of current knowledge. Take these two sentences: "DRM is the best. It makes everything so much easier!" and "The iPhone is the best! It makes everything so much easier!" Ok, algorithm. Pick the one containing sarcasm...
sarcasm = (company.attributes.include?([:big_and_evil]) && comment.classification == "complimentary")
http://science.slashdot.org/story/10/05/17/1541236/software-recognizes-sarcastic-tweets
Is not like you won't end in jail for a sarcastic comment, or get expelled over a joke, it will work in the other way, seeing sarcams where they aren't and getting you anyway. And getting this mess in your private mail, where you usually joke and don't care a lot about potential readings of what you say, because, well, you don't have anything to hide, will make life interesting in the next years.
'Elysium' Trailer, 33 seconds in.
" In theory, sarcasm detection can help with customer service, and judging how well products are doing on the open market... "
Or, just perhaps, marketing could read (listen?) for themselves to see how things turned out...
Next up....
Sarcasm in 3D !!!
And brought to justice! How dare they making legal and ethical NSA interception and interpretation of all communication harder! That amounts to terrorism! Time to find all these thought-criminals and lock them away for good. All clear speaking and thinking citizens will live in a better world for that.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
It's an arms race. The better the detector, the better the spammer, the better the detector...
But, can you imagine how good sarcasm would get if they started treating it the same way? Words so powerful they could melt your screen!
--- Need web hosting?
You have to have a clear notion of what's expected to identify irony, and that's a function of the topic, the venue, and the history of the writer.
Fortunately, the utter brilliance the designers have shown by thinking of the idea in the first place will carry them beyond such minor details and bring them complete success.
I've often thought slashdot would benefit from a -1 Woosh mod option.
What we really need is a punctuation mark for sarcasm.
Of course, the illegible drivel that sits atop most /.-pages defies classification even by humans, so some margin for error is reasonable.
...sarcasm detection...
What could go wrong?
Python: 'And then suddenly you have a language which says "we're all stuck with whatever the whiniest coder wants".'
Will they be able to compile a parenthesis?!
Download SpamAssassin and rename it SarcasmAssassin.
Detect for "right." alone in a sentence, add bonus point when it separate the main text with a cariage return. "Yeah, right." make the detector explode.
C. Sagan : A demon haunted world:
http://www.amazon.com/gp/product/0345409469/
visit randi.org
It's a function of intent and inflection.
Trying to divine it from raw text is going to fail. Simply because such systems will be deprived of the necessary information to make such a call properly.
Sure, old chestnuts like "Nothing could POSSIBLY go wrong!" might trip it. But sarcasm extends beyond the basics and into some fairly obscure, arcane and downright subtle usage.
It's going to be like handing a blind person a ball and asking them to divine the color.
Chas - The one, the only.
THANK GOD!!!
Though it would be far from flawless, such a feature would be useful due to this incident http://www.gamespot.com/news/texas-teen-jailed-after-league-of-legends-argument-6410871
You Suck...No you really do suck.
hey, a.i. -this,- cha-cha.
This project will achieve the first 66% outright, then fanny about ad in-fanny-itum on government subsidies to ivory towers near you, workin' hard, strivin' to get the other 9% nailed down.
Beside the white collar welfare, of course, this will help keep those academics and the stunnedents whose minds they poison voting correctly.
Sweet, sweet descent into oblivion!
Get thee glass eyes, and, like a scurvy politician, seem to see things thou dost not.--King Lear
Another stupid smartphone app coming to a screen near you.
"French tech firm Spotter has apparently devised an analytics platform capable of identifying sarcastic comments" Hence forth referred to as Sheldon Cooper.
Troll is not a replacement for I disagree.
When I was in university, I heard a story about a professor in a university lecture theatre filled with hundreds of students lamenting the fact that there are many logical constructions in English where two negative statement combine to make a positive statement, but none where two positive statements combine to make a negative statement. Some smart alec in the back yelled "Yeah, right".
I told this story to a CS prof. I had. He started trying to reaffirm the assertion, then I mentioned that "Yeah" is positive, and "right" is positive. He thought for a second, squinted and pressed his lips together, then blurted that sarcasm doesn't count.
As for the software developers creating the sarcasm filter:
Good luck with that, ya bunch of winners!
Sarcasm detector pointed at Slashdot, promptly exploded.
Oliver's law of assumed responsibility: If you're seen fixing it, you will be blamed for breaking it.
Oliver's law of assumed responsibility: If you're seen fixing it, you will be blamed for breaking it.
No, really.
EU Commission could save some money here. The algorithm to detect sarcasm when speaking of EU Commission is simple, as nobody ever tells anything good about the EU Commission : If a sentence has a positive word, then it is sarcasm.
http://www.theonion.com/video/report-70-percent-of-all-praise-sarcastic,14134/
Remember kids, if you're not paying for the service, YOU ARE THE PRODUCT THAT IS BEING SOLD.
I think I'd probably overload it...
So we can rest.
That's such a great idea. /me waits for the sarcasm detector to melt down due to this, and other posts in this thread...
In theory, sarcasm detection can help with customer service,
So, if you know that your product/service is crappy and customers tell you that "you guys are just the greatest" the red sarcasm light will go on.
Ingenious! Nobody could have figured that out without software. You guys are just the greatest!
</sarcasm>
(That tag was just for /. moderators, because they're such sensitive creatures aware of the smallest subtleties.)
</sarcasm>
Yes, one can have nested sarcasm, you genius.
Implicitly we all realize that they want to filter sarcastic remarks out of online posting. Sarcasm is a very effective way to combine criticism and humor, and the result can be a very effective critique. This makes it very troublesome to those with power and money. They don't want anyone rocking the boat or getting uppity.
So instead of addressing potentially meaningful critical responses, or accepting the reality that people enjoy making bad jokes, they seek to automate the process of self serving censorship.
The intent is bad. I'm sure that organizations considering using this technology don't care about false positives. What they want is for you to STFU, unless you say what they want you to say.
So while Slashdot posters make the truly obvious jokes, or argue about technology and false positive/negative rates, this reveals the ugly truth about the intent of big online organizations. They want to enforce a one way channel where users are censored. Considering that Slashdot considers itself to be an elite corner of the internet, I find it pathetic that no one has a clue about what this means.
Why is Snark Required?
1) Is the internet connected?
2) Well there you go.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
Really
The guy who wrote the summary is really good at sarcasm.
For sarcasm to effectively be used in a text media, the writer will in many cases leave clues to its sarcastic nature. You shouldn't dismiss this so quickly, though it is a difficult problem to solve, it isn't as difficult as a blind person divining the color of a ball.
And can easily be achieved without any sarcasm detection.
Social media mentions have a strong positive bias, so simply guessing positive all the time will get you
pretty close to the 80% mark. and with a simple list of negative phrases you can pass the 80% mark in identifying if a social media mention(twitter, facebook, etc.) is positive or negative.
Linus is a fun guy.
This whole idea that sarcasm doesn't come through in text needs to be revisited.
I have a reputation in my work environment for being perceptive, thoughtful, and lucid. I also have a reputation for having near perfect recall of anything previously discussed that could possibly go wrong, and for sometimes becoming extremely intense and hard to deter from constantly injecting these unhappy reminiscences into self-satisfied negotiations until everyone else glasses over. Others might characterize this as a geek loss of control thing. I prefer to characterize this as an obnoxious streak where I constantly remind people of just how lazy they are (cognitively).
Drucker says that if there's no conflict around a decision, you should cancel the meeting and come back better prepared. I have this weird capacity to internalize long lists of reasons why anything might possibly not work, and recall much of this years later, the way some people memorize lists of baseball players. Let's put it this way: on the first round of viewing, no one in the room was surprised that I could beat Sheldon to many of his lines. Nor would people have been surprised for me to comment (at an appropriate juncture) that K4 is the smallest complete graph with no Hamiltonian path. So for the purposes of linear screen-writing (essential to the joke), Spock (the punch line) needs his lizard. But why "lizard"? Well, some combination of phonology, meter, and a viable coin-flip vagary of superiorities and frailties, with a subtle invocation of cheesy Gorn action figures (universally possessed among the group, though left unstated in the script through the use of Hemingway silence).
First of all, dead-straight sarcasm mainly plays against personal memes established in the group. If you're widely known as the guy most likely to mention Hamiltonians and Hemingway in the same sentence, your remarks will be taken in a certain light when your sentence is uncharacteristically plain, and you can play off that. No single-utterance algorithm ever devised will detect this (though it might succeed in raising a flag that an otherwise bare statement likely plays off in-group dynamics).
Second, much intended sarcasm is simply bad writing, typically perpetrated by the participants who are better at jostling for attention than presenting a sustained perspective. These people want their successes to be more memorable than their failures and tend to be completely content receiving credit for a remark construed opposite to the spin attempted. These people are happiest when no particular valence sticks to their persona. Misconstrual is half the payload. That's the closest they ever get to a date. (Note that this verbal pun on "Miss Construal" leaves the word "menstrual" partially activated in the subconscious with nowhere to go, which is half of the humour but none of the joke. The other half of the humour, but not the joke, is the unsticky half payload of no particular valence. "There was some scattered clapping, but most of them were trying to work it out to see if it impugns virility." That's also funny on a second level: mocking with a formal nuance of the word "impugn" this entire business of establishing one's virility via assertional discourse. Oops, I did it again.) Summarizing: bad sarcasm is half-assed and, sadly, good sarcasm often lands to scattered applause (someone disaccord me a Kleenex).
Finally, sarcasm is not a quality of an utterance as a whole. The same paragraph can wend through sarcasm, mockery (sometimes self-ish), memes, cues, call-backs, verbal and visual double meanings with all the obviousness of Sheldon's didactic Hamiltonian. Koothrappali gets it, on the first take.
That's a joke itself (from the original script) in how it emphasizes one kind of cognition over another. More typical in human discourse is a litany of ten ambiguous statements. The people who get sarcasm most reliably are the ones who can maintain a larger cloud of ambiguity for longer durations. Rushing to accept
Call me when they develop something that detects sarcomas. That would be useful.
Star Trek transporters are just 3d printers.
A bullshit detector was considered too destablizing, and a criminal-intent-o-meter a very dangerous two-edged sword. Best to ease into these things, slip them under the radar, so to speak. A working sarcasm detector fills the bill. Nobody likes a smartass, right?
Delusional AI reductionists will never actually be able to build any such thing, of course. That's why the law is an ass, same principle. Why do they think an algorithm by any other name will be any different? Because people are, not stupid, just lazy and willing to surrender their own judgment to authority. Especially when it has guns attached, or rather, men with guns. Ample incentive for rationalization abd equivocation.
Course, I could just be waxing sarcastic. What does the meter say?
It's a function of intent and inflection.
Trying to divine it from raw text is going to fail. Simply because such systems will be deprived of the necessary information to make such a call properly.
It's more a function of context, which includes familiarity with the speaker/writer, or their ilk (as others have noted above) — as well as one's own predilections and (obviously warped! :)) viewpoint
[...] sarcasm extends beyond the basics and into some fairly obscure, arcane and downright subtle usage.
It's going to be like handing a blind person a ball and asking them to divine the color.
(Forgive my smirk: But I think it's more like handing a blind person a seeing-eye cat... :))
Sorry! I had no intention, to be anonymous... So: Now that I'm a "member" of this what-ever-it-is, can we talk? (We'll see.) I had to return, and comment, because of epine's comment... (I could argue. But only for fun. S/he nailed it – or close enough, for government work.) The Turing test will continue to fail, until a practical joke played by a computer causes the death of a human... Nyuk, nyuk! But maybe not: If the defendant offers a reason for why Jerome Howard was called Curley, and riffs, who knows...
"Humor is emotional chaos remembered in tranquility." - James Thurber