According to the New Scientist, it’s almost here:
Now a British company is about to launch a software program that can automatically gauge the tone of any electronic document. It can tell whether a newspaper article is reporting a political party’s policy in a positive or negative light, for instance, or whether an online review is praising a product or damning it. Welcome to the automation of PR.
Funny, I always thought it was. (ht: Charles Martin)
PJ Media appreciates your comments that abide by the following guidelines:
1. Avoid profanities or foul language unless it is contained in a necessary quote or is relevant to the comment.
2. Stay on topic.
3. Disagree, but avoid ad hominem attacks.
4. Threats are treated seriously and reported to law enforcement.
5. Spam and advertising are not permitted in the comments area.
These guidelines are very general and cannot cover every possible situation. Please don't assume that PJ Media management agrees with or otherwise endorses any particular comment. We reserve the right to filter or delete comments or to deny posting privileges entirely at our discretion. Please note that comments are reviewed by the editorial staff and may not be posted immediately. If you feel your comment was filtered inappropriately, please email us at story@pjmedia.com.
The problem of using software to understand human language, which is basically what this is about, is notoriously difficult. I wouldn’t object to the use of this technology for broad and general PR purposes…such as developing trend lines to see if the market loves or hates your new products. In that role, a 20% error rate will probably cancel out. But what really scares me is that people are talking about using similar technologies for the grading of student essays..in which case, the relevant sample size is one, and the errors do not cancel out. I think that’s irresponsible.
I make the irresponsibly bold prediction that this technology will work nearly as well as the grammar checker in Microsoft Word.
Jeff: “I make the irresponsibly bold prediction that this technology will work nearly as well as the grammar checker in Microsoft Word.”
That bad, huh? Yee-owch!
How is it going to be on irony? Even carbon based lifeforms have trouble with that.
Irony? For the scientist, it doesn’t matter what the article thinks it says, it’s what the program says it says… and (we’re almost there) what the writer says who knows what the program will say… I’ve got a friend working full time to model the stock market with a team of PhDs. So far any insight they get into how the market responds (has responded) to information and trends is quickly discounted in value. But, as he likes to say, we’re “almost” there. Such are our delusions.
Let me know what happens when you feed it a vintage Steyn or Hitchens. Or a Buckley or a Lileks for that matter.
So this is essentially the first robo-blogger. Read a couple of sentences and then detect “liberal bias” or “conservative bias” or “anti (blank) bias”. This is what most bloggers do already. We’re being replaced!
“We’re being replaced!”
No, we are not. That is strictly impossible. Language is intrinsically nebulous. It is not static and unchanging. The meaning of words will endlessly fluctuate. Human communication is not limited to mere verbal utterances. Our body movements and tone of voice are sometimes more important. Human beings often use irony. That is why there will never be real life HAL as depicted in the film 2001: A Space Odyssey.”
If this program works as well as that ‘where are you on the political axis test’ some British site runs, we’re liable to find out Operation Rescue endorsed John Kerry. These people are trying to create a program that not only reads humans’ imprecise or manipulative use of language, but will do so across cultural lines. These Brit scientists have as much chance of that as I have of understanding the Workers’ Revolutionary Party…
I dunno, guys. There’s a lot of “impossibles” being thrown around that I don’t think are all that impossible. First of all, I don’t think that detecting the slant of an article is all that cognitively difficult, it’s not a full-fledged understanding problem. Consider, for example, the difference in connotation between “Senator Kennedy today said the Bush Administration didn’t have a plan to save Social Security,” and “President Bush today claimed to have a plan to save Social Security.” You don’t need to know what “Social Security” or “Senator Kennedy” or “President Bush” denotes to identify that the person who put together the sentances gives more credence to Kennedy than Bush.
But then the “Latent Semantic Analysis” work on essay grading is rather more complicated and nuanced than you might imagine too. I happen to have been following it for a while, and a lot of that stuff was done here at Colorado. Here’s a web site: Latent Semantic Analysis.
Whatever you may think of it philosophically, it appears the LSA essay grading is about as good as the grad students they normally use to do the grading.
(Which is to say: Molly Bloom’s Soliloquy may not do well, but then would the usual grad student give it a good score if they weren’t told ahead of time it was Art?)