Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 50 original podcasts with new episodes released every day.
Introducing: The Tone Analyzer
IBM Watson has sent out a press release, herewith dutifully reproduced by Gizmodo:
Taking a break from treating cancer and making cocktails, IBM’s Watson is now turning its attention to how people write. The supercomputer has been trained to judge the tone in people’s written messages—and can even give feedback about how to change it.
The IBM Watson Tone Analyzer sifts through passages of text to identify the emotional and social tone contained within, as well as the writing style. It can work out if the passages is cheerful, negative or angry, and then classify it depending on the openness, agreeableness, and conscientious of the message. Finally, it’ll describe how analytical, confident or tentative your writing style is.
That all allows you to tweak the text, with Watson making suggestions about which words can be changed to tweak the tone. The idea is to help guide people to write content that better suits their needs. Think of it, I suppose, like a spell check for tone.
I’m of several minds. On the one hand, I’m the one who said, “I bet we can automate the ‘cordial’ part of ‘cordial conservative conversation.'” Remember that? If Americans can buzz Pluto, they can automate the tone. So I feel vindicated at the thought that someone, somewhere, agrees with me.
On the other hand, since the beneficiary won’t be me, but IBM Watson, I’m completely against it. This thing will put me out of a job if it works.
On the third hand (that’s just my rhetorical forelimb, don’t stare), I’m not sure they’ve quite achieved what they’re hoping for:
What’s your verdict: Successful tone analysis? You think this gizmo’s a threat to my job?
Published in Culture, Entertainment, General, Science & Technology
From now on, I will picture confetti when politcal parties are mentioned. So cheerful!
Get rid of the damn thing. It’s bad enough with you correcting us, let alone a machine. We can argue with you and on occasion you might be wrong. But a machine telling me I’m hostile? Ftttt
Somehow I’m guessing this thing is extremely bad at recognizing sarcasm.
I’m detecting some negativity. ;)
Claire,
If you notice all such successful text analysis devices, including spell checkers and grammar checkers, work the same way. They highlight the “questionable” text and offer alternatives. One of the alternatives is always to make no change at all. This lets you in on the confidence level of the automation of human language. It can only guess your intent and a how to do it a better way part of the time. Human language is so filled with nuance that all such software is never used as a final judge. Translation software, for instance, has become useful as a first pass translation. However, a real human expert in the foreign language must do the final pass and check the machine’s work.
Now you can rest assured that you are not out of a job. Besides, if Watson tried to push it too far with me I’d sock the d@#med thing right in the mouth..er..speaker. Watson could call my tone angry all the way to the emergency room..er..geek squad.
Regards,
Jim
It’s interesting how “Social Tone” seems to be the bucket that “meaningful” words get thrown in when they’re neither emotional words (“Emotion Tone”) or structural cues (“Writing Tone”).
For example, “On Friday Evening last” is marked as “Social Tone” in your example. Running an analysis on my latest post, “spring” and “sat on” were also marked “Social”. Now, spring isn’t particularly antisocial, and sitting on something can be done sociably (or not). But why did the analyzer mark these words but ignore the word “complement” (an analytical word) entirely? Why does the word “risk” not fit into any categories? (Isn’t risk another analytical word? Even if it weren’t, does “risk” fail to convey social tone or emotion?) And why is driving somewhere not considered “Social” while sitting on something is?
Their demo is also set up to compare the writing sample with a business e-mail template. Useful if a business e-mail is what you’re writing! Less clearly useful otherwise.
Even so, as easy as it is to poke holes in gadgets like this one when they’re in their infancy, I’m glad someone is curious enough to develop these gadgets and try to make something of them! Even in its current form, I can picture this gadget helping people to become more aware of their word choices in… business e-mails.
This must be how the NSA evaluates my email. :)
Note to self: be careful!
I’m with you. This is clearly in its infancy, but it’s fascinating. It’s clearly incapable of doing anything like what I do when I read, but that doesn’t mean it couldn’t be put to use.
Oh, that’s a fun idea — anyone else want to run his or her latest posts and see what it says? What else did it say about yours?
It is. It marks text on a word-by-word basis, as far as I can tell, ignoring context and semantic structure. So, for example,
gets marked as 100% angry, 100% negative, and only 94% cheerful (as compared to their business e-mail template), because “horribly” and “dangerous” are marked as negative, angry words. That these “negative”, “angry” words are arranged in a manner that conveys gratitude and relief is overlooked. On the other hand,
clearly meant as a sarcastic phrase when directed at an adult, is marked 100% cheerful and agreeable, 0% negative, angry, or hostile.
Automated focus groups. Good thing ad blockers work. Political speechifying? They’re already generally untrustworthy; this just would add to the cynicism.
Too, I have to wonder who wrote the rules that were programmed into IBMW. What Leftist slant (or Rightist) were written into the definitions for this or that…tone? With what weights?
Eliza was boring after a bit. This one will be more interesting.
Eric Hines
It gave the following breakdown on word count:
A post I wrote on my blog about the Wisconsin Supreme Court’s decision regarding John Doe persecutions got me 87% Social Tone (whatever that is) that was spread across Agreeableness, Conscientiousness, and Openness at a roughly 1:1:2 ratio.
I quoted the Court’s opinion extensively, so maybe that ratio comes from that. I can’t recall the last time I got accused of being Agreeable. On the other hand, the thing found a lot of emotion in the Court’s remarks….
Eric Hines
It’s interesting in a ‘you’ve got to crawl before you can walk’ sense, but I don’t think it’s terribly useful at this stage, other than to highlight the differences between machines and humans.
I just ran a rather abrupt email that I sent to my ISP yesterday (concerning an ongoing issue that’s lasted since just before Christmas) through it, and it tells me that I’m only 2% emotional, 91% social and 6% writing tone. Last time I looked, that only added up to 99%, but who’s counting, right?
I only had four ‘anger words,’ although I don’t think that’s actually fair, since I used the word “stupid” twice. The other two were “insulting” and “wrong.”
The word “care,” in the phrase, “I don’t care [who caused the problem, I would just like it fixed]” is marked as a ‘positive’ word, which makes me think there’s no context at all to this thing.
I have to say that, as one who excels at snark when I’ve a mind to, as a snark-meter, it’s sadly lacking.
In addition, I just ran a comment that, literally (as Joe Biden might say), caused World War 3 to erupt on Ricochet not too long ago. The analyzer says there are 0 anger words, 4 cheerful words, 1 negative word, 30 agreeable words, 17 conscientious words, 44 openness words, 3 analytical words, 1 confident word, and 2 tentative words, for a 4% emotion tone, an 89% social tone and a 5% writing tone.
The definitions of each of these traits can be found here.
Claire, I think you’re safe for the time-being. You’re a lot subtler than Watson.
Strangely close to the rating they gave Churchill’s Blood, Toil, Tears and Sweat speech. You scored a bit higher on openness, but that makes sense, given that he was waging war against a monstrous tyranny, never surpassed in the dark, lamentable catalogue of human crime, whereas you were sitting on the alleyway stoop, watching lilac bushes down the hill ripple as birds quarreled in the breeze.
What do I need this for. I’m already married.
Cool it Mike. Your tone is aggressive.
Not surprising, since words it recognizes as words of approbation or disapprobation get marked as “Emotion” words. For example:
I included a favorite phrase of Epstein’s, “a right good against the world”. What I’d like to know is why “a” and “the” are “Social” words.
Claire,
I have now input-ed my first comment on this post. The tone analyzer is like a Thesaurus on steroids with a text adjustment rate of a Gatling gun. If you simply get used to it as resource and not an expert it is quite interesting. Remember just think suggestion rather than red pencil correction.
I hope I didn’t offend it or frighten it by my belligerent tone. It appeared not to notice my implied threat at all. In fact it merrily went on offering suggestions of alternate words. Sort of like negotiating with John Kerry as I would imagine.
Regards,
Jim
I was struck a while back by the courtly, passive-aggressive style academics used to use to tell their learned colleagues off. I think that came from reading the book Beyond the Thunderdome: Hayek vs. Keynes but I may be a little too far-gone on WPLJs this hot, humid morning for that to be entirely accurate. That’s my Turing Test. Wake me when the machines can detect that.
I predict that Claire’s position is safe and corporate attorneys will have earlier tee-times.
Waste of time.
If you want to consider the tone of something you are writing, write it, and then wait until the next day (next week is better) to proof-read it.
It’ll make the process of online
censorshipmoderation that much easier.“The owners of this website reserve the right to delete any posts that Watson deems negative.”
I’m sure Herr Hitler found Mr. Churchill’s words to be “open” and “agreeable.”
This is what baffles me. It’s 2015, folks. Latent Dirichlet Allocation has been around for a while. Long enough that you don’t even need to implement it yourself; you can just download a good implementation and have at it. You certainly don’t need the kind of “oomph” Watson has if all you’re doing is glorified POS (Part of Speech—what did you think it stands for?) tagging. I conclude it’s not just doing glorified POS tagging—it’s just that humans tend to edit on a lexeme basis, so that’s how the feedback is provided.
If Cecily Richards can equip her medical directors with this, she won’t be having to apologize for their tone.
GGG,
Darn those Humans. This brings up an old story of mine. Long ago and far away in a small retail independent computer store (when such things existed) there was a salesman and a tech. As is often the case the 23 year old tech with the degree was running the whole show for the owner and was in command of about 50 people. The old salesman was a strange creature which the young tech liked because at least he wasn’t boring.
This was back in the days when everything was much less plug n play and you had to load drivers and actually know what you were doing and all. The young tech would assemble the clone machines himself but he wanted the salesmen to load drivers, interface peripherals (printers, modems,..etc) and load the software. We delivered a turn key to the customer and were actually justifying our retail status (until the bottom dropped out of everything). The tech by the name of Ray would multi-task doing like seven things at once. The old salesman would follow him around, similar to C3P0 following R2D2 and ask him nervously “Ray why are you doing that..what was that..how are you doing that..ah Ray explain that..” The old salesman could even feel his arms going out to his sides just like C3P0 trying to keep up with R2D2. It was mostly R2D2 doing the important work and C3P0 was just a 5th wheel. Then one day something was different. Ray R2D2 was becoming increasing frustrated. All of the customers wanted this new thing called Windows. R2D2 didn’t see the point at all. After all the damned thing took up tons of memory and processor power. If you used your function keys well on your 101 key board you could beat the Windows idiot and his drop down menus. R2D2 just couldn’t see the point.
Finally, C3P0 realized his function in life. He must explain to R2D2 about this Windows thing. He thought for a while and then it hit him. The perfect way to explain the Windows thing. He said “Windows is a Human driver. When you want to interface one of those large Human peripherals to the PC you’ve got to use the right Human driver.” R2D2 found this very funny and understood. At least the old guy wasn’t boring.
Regards,
Jim
I’ve read numerous times that automated lie detector tests are unreliable, yet they still have been accepted as evidence in criminal trials (or were in recent decades). Whether the system works or doesn’t work is less important to prosecutors than whether or not it is convincing enough to sway a jury.
We laugh at these automated analysis programs today, but we might not be laughing ten years from now.
Patrick gets an agreeableness score of 96%! Just like he was aiming for.
My cat is pretty good at this.