New A.I. tech helps you write right
New A.I. tech helps you write right
The publicly available Watson Tone Analyzer example site
shows you how your words will come across on scales about emotion, social
propensities and writing style. Credit: Screenshot/Mike Elgan
The newest cloud tools (which are online now) can help
you fix your writing as you write.
By Mike Elgan Computerworld
| Jul 20, 2015 3:01 AM PT
This column is a little cheerful, slightly analytical,
both confident and tentative and just a tiny bit angry. But mostly, it's open,
agreeable and conscientious. At least that's what IBM's Watson thinks.
Last week, IBM revealed that its Jeopardy-winning
supercomputer has a new capability. It's called Watson Tone Analyzer. You can
use it like spell check, except instead of checking your spelling, it checks
the "tone" of your writing. (And, by the way, when I say you can use
Watson Tone Analyzer, you can literally use it right now on IBM's experimental
public page.)
Watson Tone Analyzer is part of a new generation of
writing tools that go far beyond the old spelling and grammar check. Instead,
they help you polish and perfect your writing to achieve very specific goals.
Of course, there is already software that does all the
writing for you. For example, companies like Automated Insights and Narrative
Science offer products that take published data (such as financial numbers or
sports scores, stats and other data) and turn it into prose, usually in the
form of news stories or financial reports.
Here are examples of current news stories written by
Automated Insights' product, which is called Wordsmith.
The fact that this technology can do this kind of writing
says a lot about this category of prose. It's all utilitarian information
delivery -- stories and reports that no writer wants to write and no reader
wants to read. But the information must be conveyed somehow, and artificial
intelligence is far cheaper than real intelligence.
Writing well is way too hard for computers. For example,
a good writer wants to delight, be evocative and hold the attention of readers,
and humor is a great way to do that. Thing is, computers just aren't funny.
New research from the University of Michigan, Columbia
University, Yahoo Labs, and The New Yorker has found that even the most
advanced artificial intelligence software can't tell the difference between a
funny New Yorker cartoon caption and an unfunny one, even after analyzing
massive data sets and hunting for correlations.
Prose-generating software robots represent interesting
technology, but they've got a long way to go before they can write stories
people want to read.
A far more interesting branch of technology helps people
write better, instead of trying to write for them. By write better, I mean to
use language in a way that helps readers reach specific goals.
Watson Tone Analyzer is the newest of these technologies.
It became available last week as part of the Watson Developer Cloud application
programming interfaces and software developer kits, which are the tools IBM
offers to developers so that they can use Watson's capabilities in their
software.
Watson Tone Analyzer rates your language in three
categories:
Emotion (negativity, anger, cheerfulness)
Social propensities (openness, agreeableness,
conscientiousness)
Writing style (analytical, tentative, confident)
You'll note that the "judgment" in these three
categories is positive or negative. One of Watson Tone Analyzer's potential
uses is to help you boost the positive and cut the negative.
Other uses might be to customize or adjust the tone of
marketing messages for specific target audiences, market research, PR and
automated contact center management, according to IBM.
One other potential application that is very interesting
is that the technology could give virtual assistants -- such as Apple's Siri,
Google's Google Now, Microsoft's Cortana or Amazon's Alexa -- the ability to
"understand" the "tone" of their users' requests and then
respond with an appropriate "tone." For example, by detecting elation
or sadness, one's virtual assistant could respond with excitement or empathy,
respectively.
When you process your words in the Watson Tone Analyzer,
it highlights and color-codes all the words that contribute to tone, and by
clicking on those words you can see Watson's suggested alternatives and
improvements.
IBM expects developers to create Watson Tone Analyzer
plug-ins for browsers, email applications, social-media front ends and other
applications.
While all this sounds great, the trouble is that
"tone" in language is, to date, impossible for software to deal with.
Back to my own example, Watson Tone Analyzer detected "anger" in my
prose, but only because I was describing the ability of Watson Tone Analyzer to
detect anger. The word "anger" repeated several times caused Watson
to say, in effect, "Hey, calm down. Why so angry?"
This highlights the vast distance that artificial
intelligence has to go before it can know the difference between talking about
anger and writing in an angry tone.
In that respect, Watson is useless for most people who
write well. But that's not the case with another major product in this
category, called Textio.
Textio
A cloud-based artificial intelligence service, Textio is
another example of how software is trying to help people write better -- or, at
least, more effectively in the achievement of specific goals.
Textio was founded by big-data experts from Microsoft and
Amazon. Their slogan is "words + data = magic."
Textio applies big data to link language with specific
outcomes. For instance, verbatim job ads can be fed into the Textio algorithm,
along with data on who applied, and the system can figure out which words and
phrases either succeeded or failed in support of the hiring company's
recruiting goals.
For example, if a company's goal is to not push away
qualified female candidates, it's useful to learn from Textio that phrases like
"under pressure" (as in, "we're looking for a candidate that
works well under pressure") tend to drive women job hunters away, while
phrases like "passion for learning" tend to attract women to jobs.
Some of the "gender biases" that Textio ferrets
out with its big-data crunching are puzzling, but they can be verified
statistically. For example, more women are more likely to apply for a job if
the word extraordinary is used in the listing instead of the word exceptional.
Gender bias in job postings is one problem Textio claims
to help with. The other is the need to simply hire better candidates,
regardless of gender.
Textio costs $59 per user per month. After you sign up,
you can just copy and paste a job posting into the Textio Web form, and the
system will spit out its analysis.
Textio will highlight words and phrases known to turn
away candidates, either because they're too jargony or too cliché, or because
they contain keywords that Textio's analysis has determined will result in
inferior candidates or unsuccessful hires.
Before Textio got into the business of detecting gender
bias, the company used its analytical kung fu to try to predict which
Kickstarter projects would be funded. And in the process, it discovered that
the way a Kickstarter page was built and worded had a bigger influence on
funding than what the product was.
Even now, according to Textio CEO and co-founder Kieran
Snyder, who has a Ph.D. in linguistics and worked at both Microsoft and Amazon,
people use Textio's service to evaluate all kinds of communications, even
though the system is optimized on a huge set of job listings.
Unlike Watson Tone Analyzer, Textio could be
significantly useful even to the most skilled writers, such as professional
novelists or, say, tech journalists.
Even great writers can (and often do) write in a way
that's unappealing to one gender or another, or in a way that will
unintentionally push away prospective employees or crowd-funding investors.
The reason Textio works is that it doesn't try to
understand human language -- something far beyond even the most advanced A.I.
Instead, it does what computers are good at -- it finds correlations in data
sets. It knows that the word exceptional in a job listing will attract fewer
women candidates than the word extraordinary, even though it has no idea what
exceptional means or why the correlation exists.
Some people may be tempted to fear that software,
supercomputers and algorithms are going to replace us all -- including those of
us who write for a living or for whom writing is a major part of how we make
our living. Some may be tempted to dismiss these fears and say that software
can never replace people in these discipline. Either way, this technology is
astounding.
But there is a winning combination in all of this. As the
technology gets better, it's becoming clear that, ultimately, the literary and
creative skills of human writers combined with artificial intelligence writing
tools can help us communicate better.
Comments
Post a Comment