How do Amazon, Google, IBM, Microsoft and others edit?

I’ve just spent two fascinating days at Content Connections 2017, the annual forum for top US-based users of the Acrolinx artificial intelligence (AI) editing platform and other interested parties. Held in San Jose in the heart of Silicon Valley, I learnt how leading global organisations are using technology to improve the quality of their content. Some standout insights from the forum include:

  • Computers will play a much bigger role in writing and editing, and soon.
  • AI systems can now ‘read’ paragraphs; whole documents may be next.
  • Writers may prefer to get corrections from a machine than a fellow human.
AI is the future of writing and editing

The first insight dawned as I looked at the name tags on the registration desk. The big American users of Acrolinx are the world’s leading technology companies. Adobe, Amazon Web Services, Cisco, Google, IBM, Microsoft, PayPal and others are embracing AI editing because they see it as the only way to process large amounts of content quickly. Blending human and computing capabilities is helping companies enforce their language and style rules globally, while improving the writing skills of their people.

Tyson Nevil from Microsoft, for instance, showed how his Azure business unit uses Acrolinx to control the quality of technical articles written by more than 400 authors. Microsoft publishes these articles on the open-source GitHub platform to assist software developers (see this related story). Nevil explained that the company tried to manage this enormous stream of content using human editors but quickly became overwhelmed before seeking out a computing-based solution.

Some of the major technology companies are also seeking to blend Acrolinx AI with their own capabilities. Michael Iantosca from IBM, which uses Acrolinx to help manage more than 220 million pages of online copy and other content, said he was exploring ways to combine Acrolinx with IBM’s Watson AI system.

The mind boggles at what could happen when the smartest people on the planet combine a platform that has deep language capabilities (such as Acrolinx) with one that has an enormous capacity to digest and synthesise information (Watson). Perhaps we aren’t all that far from the days when AI systems not only help us create blogs and books, but start writing them as well. In fact, this is already starting.

It’s early days for such experiments but it’s clear we’re on the cusp of an extraordinary leap forward in the way we use and engage with language. Acrolinx’s founder and CEO Dr Andrew Bredenkamp has explored this idea, and it raises profound questions about the future of writing, editing and – given we think in language – how our minds work.

I should also note that there were plenty of non-tech attendees at Content Connections, including representatives from organisations as diverse as Chevron, Medtronic, Nestlé and Wells Fargo.

But what is AI?

With more than 15 years experience in the field, Dr Andrew Bredenkamp has been asked “What’s the AI in Acrolinx?” many times. The short answer is that when a system like Acrolinx – or Watson or Apple’s Siri – works with words, it draws on a natural language processing (NLP) engine. This is the cornerstone of how AI systems make sense of language to enable processes such as analysing whether a sentence is grammatically correct or deducing that you want to know the time.

Those NLP engines have been in development for well over a decade. They are essentially AI systems that draw heavily on the discipline of computational linguistics. Groups like Acrolinx build other processes around their core NLPs to deliver additional functionality, such as deciding whether a company is using its own terminology correctly. They also incorporate machine learning to spot patterns and enable their systems to deliver more value over time (which is what many people think of as AI), but the core intelligence rests in the magic that occurs inside an NLP engine. Unfortunately, that’s hard to explain quickly, so you might like to read this helpful Wikipedia description and this more general series on the implications of AI for business.

AI is moving from sentences to paragraphs, so what’s next?

You know you’re sitting in a room full of language geeks when the news that Acrolinx can now analyse whole paragraphs, in addition to words and sentences, draws a gasp. The shift from sentences to paragraphs is significant because it shows how quickly computer-powered editing systems are evolving. If you think about your experience of using spellcheck in Microsoft Word, you’ll recall it originally could only identify whether individual words were correctly spelt. More recently, it has been able to spot issues in the syntax (grammatical construction) of sentences. This was clunky for a long time but has really improved in the past few years.

Acrolinx can now look at how well written a whole paragraph is. Instead of comparing words to those held in a dictionary database or comparing sentences to its understanding of how language is usually used (or should be used), Acrolinx can look for things like an author starting two sentences with the same word or structure. For instance, it’s not a spelling or grammatical error to start two sequential sentences with the phase ‘On the other hand’, but it would be repetitive and probably illogical. The smart people at Acrolinx have worked out how to make their system look for those types of issues and flag them to authors as something to consider (which is how its ‘guidance’ typically works).

Interestingly, Acrolinx hasn’t (yet) been able to work out how to reliably advise on whether a paragraph is too long and should be broken into two or more new paragraphs. The software certainly advises on whether sentences could be broken up, but Acrolinx’s NLP expert Peter Tauter said his team was continuing to work on whole paragraphs.

Again, this has massive implications. If an AI system can both ‘check’ words and sentences and ‘read’ paragraphs, how long before computers can read whole documents and offer useful commentary – especially if they’re hooked into other AIs that give them subject matter knowledge like Watson, Siri or Google’s DeepMind (the AI that beat the world’s best human Go players)?

Furthermore, what role will remain for traditional proofreaders and editors? It seems that as computers can handle more of the grunt work involved in getting language right, editors will come to focus on two main things: applying their expertise to analyse the value of the author’s message, and deciding when to accept the spelling, grammar and structural advice offered by a computer. But just as oncologists now consult systems like Watson before offering cancer diagnoses, we are likely to see the relationship between writers, editors and AI quickly become deeper and more intertwined.

Writers prefer feedback from machines

If you chat to editors about these issues, many will say while it’s nifty that machines are getting better with words, authors will always prefer getting advice from a fellow human being than a computer. Unfortunately, Content Connections popped that bubble.

Tyson Nevil said many of the writers who contributed Microsoft’s GitHub articles preferred getting feedback from the neutral and consistent Acrolinx AI (think HAL in 2001: A Space Odyssey or Samantha, the AI in the more recent movie Her, though without the cool voices) rather than a traditional editor. “It was really awesome to get feedback that wasn’t from a human … Editors can be snarky!” he said, quoting his team.

As a professional editor, the owner of an editing business and, irrevocably, a human, this suggests two things to me. First, our firm should offer clients the most appropriate blend of computing and human capabilities to meet their needs – which is why we’re now helping our clients implement Acrolinx and other solutions. You can imagine that the people who write Microsoft’s software might be uniquely fond of interacting with computers, but my bet is they’re not alone. Second, human editors should strive to be very objective, consistent and downright nice to authors!

Other tidbits

Like any multi-day conference, there were lots of other interesting takeaways and factoids:

  • Stories are compelling. Kelvin Gee, Senior Director for Modern Marketing Business Transformation at Oracle, reminded the audience that it’s not enough just to write well. You also need to be saying something compelling and the most memorable way to say something was in a story. Those stories should have a setting, character/s, a point of conflict and then a resolution.
  • Post-sales content matters. Most businesses focus heavily on the quality and effectiveness of their marketing and communications material. However, attendees cited estimates that 55 per cent of copy read by customers is post-sales material such as product manuals and support documentation, so lifting the quality of this material can boost sales.
  • The oldies are still the goodies. According to the SiriusDecisions 2015 Buying Study, when it comes to generating business-to-business sales, the top six most effective types of content are: sales presentations, analyst reports, case studies, articles/publications, static brochures and white papers.
  • But content must also be fresh. Multiple speakers discussed the importance of ensuring that the content they produced was current and regularly refreshed. They said this was a greater – and still growing – requirement as more of their products morphed from being sold in one-off transactions to being delivered as ongoing and regularly revised services. This means that their related content has to remain as dynamic as the product itself.
  • Content marketing remains contentious. Phyllis Davidson from SiriusDecisions said there was still considerable debate in the US about the meaning of the term ‘content marketing’. Her firm’s view is that it’s too generic at a time when content is central to most sales activity. Instead, it is better to think of content marketing as the reports, blogs, tweets and so on designed to generate inbound traffic and leads, and everything else as ‘marketing with content’ (see the photo above). She also noted that ‘creative services’ teams were being replaced by ‘content factories’ that are part of wider content hubs.
  • Managers waste a lot of time editing. Deborah Bosley from The Plain Language Group estimated that the average manager spends up to five hours per week fixing the English written by their subordinates. Across a group with six managers and over a year, that could add up to a US$75,000 a year impost. Do your own cost calculations here.
  • Silos are trouble. The biggest obstacle to achieving high-quality and consistent content is silos. To overcome this, larger businesses are now creating centralised content governance groups and using Acrolinx to enforce language standards – and identify problems – across operating groups.
  • It pays to start small. Companies are finding it is best to start small with content governance programs and when implementing Acrolinx specifically. This means identifying the top 10 or 20 language usage issues the business is really concerned about, and which tie most directly to business goals, then seeking to improve performance around them.
  • There’s a lot of stuff online. According to Oracle’s Kelvin Gee, there are currently 305 trillion web pages. This will increase fivefold by 2020, so “attention is the new oil”.
  • Tone of voice matters. Steve Rotter, Chief Marketing Officer for the European software maker OutSystems, presented on the importance of tone of voice in writing. As an example, he said his company had increased the average time visitors spent on its website from four to nine minutes just by rewriting the copy to make it snappier and more appealing (which isn’t bad when there are 304.999 trillion other sites those visitors could be viewing!).
Learn more

If you’d like to find out more about Acrolinx specifically you can see:

PS…

This article was written with extensive guidance from multiple computing systems and human editors. As I wrote the first draft, Microsoft Word picked up some basic spelling errors and grammatical issues. Then I ran Acrolinx, which suggested I cut some long sentences in half and flip a number of sentences from passive to active English. Acrolinx also picked up that I had contravened a few Editor Group house style rules such as using ‘years of experience’ instead of ‘years experience’, which we put in place in deference to the Australian Government’s Style Manual. (The cool thing about Acrolinx – and perhaps its most important feature – is you can teach it your own rules).

Even now, Word and Acrolinx want me to change ‘a number of’ in the middle of the above paragraph to ‘numerous’ or ‘multiple’ for brevity, but I like ‘a number of’ for clarity and tone so I’m ignoring them. This is something you should stay brave enough to do – computers are brilliant and we should work with them, but never forget they were built by humans and they can’t read your mind (yet). Finally, this blog was proofread by Editor Group’s expert team, who looked at the language to make sure the whole piece made sense, ‘felt right’ and would help you, dear human reader. I hope it does.

Grant Butler is the founder and managing director of Editor Group. Contact him here. 

Read more

What is active content governance?

 

Pin It on Pinterest