Last Friday, at the Apple Store near the Paris Opera House, I paid my annual Microsoft tax: €140 ($194) for the 2011 edition of Microsoft Office. My hopes: more speed, less bugs, and smarter features. All in the service of producing all manners of text and presentations required by my multiple jobs. So far, no mind-blowing features, nothing more than a superficial makeover.
To look at this new iteration of Word, I use the framework built on my experience of Microsoft’s R&D effort. A few months ago, I spent three days at the Microsoft Tech Fest in Redmond. At first, I felt like a kid in a candy store, chatting with some of Microsoft Research 900 plus PhDs who work on exotic fields such as Machine Learning or Epidemiology. But the amazement subsided and was replaced by doubt: How did this tremendous intellectual firepower actually make a difference in the Microsoft products I’ve been using for 15 years. In fact, Microsoft R&D has very little impact of everyday products. This is but one of Microsoft’s many problems: see the long piece I wrote in Le Monde Magazine.
Let’s go back to the subject of this column. Knowing what I know about Microsoft’s vision of computer science, I had envisioned of a quantum leap for applications I use the most, such as the very word processor on which I’m using “as we speak”. No joy. Let’s ignore the letdown and, instead, speculate a little bit about the next generation of text creation tools branded Microsoft Word, or Apple Pages (which comes with fewer bells and whistles, but is tidier).
First, text creation. One of the biggest challenges, and a growing one, is spelling, syntax, and grammar. In a country such as France, whose language is loaded with utmost (and sometimes absurd) complexity, the quality of writing is in steep decline. For the youngest part of the population, it is accelerated by the demise of a school system where teachers in effect gave up on written language. As for the 30-40 age bracket, the bombardment of daily interactions (email at work, SMS, chat on social networks) has made proper spelling and syntax secondary. Quite often, coming from a manager or even an attorney, you’ll receive a business document riddled with spelling errors well beyond the typos acceptable in a hastily written piece.
Unfortunately, today’s word processors do a very poor job when dealing with mangled spelling and grammar. All of us have in mind examples where the Word application becomes absurdly creative when dealing with the unknown: regardless of context, and with no learning capabilities whatsoever, Word will stubbornly keep suggesting an alternate spelling instead of simply skipping an unrecognized term.
Let’s dream for a moment; let’s picture what a text processing software could look like in the light of existing technologies.
When I install my 2013 version of MS Word or Apple Pages, it asks me to load a “reference corpus” of texts it will learn from. Since I write both in French and in English, I will feed the app with the final versions (edited, and proof-read) of articles I published and I’m comfortable with. Grammar and syntax will be helpful for English and thesauruses will be used for both. Since I currently write about media and technology, the application dictionary will soon be filled with the names of people, places, companies I mention, as well as with the technical jargon I allow myself to use. Alternately, if I don’t want to feed the word processor with my own writings, I can direct it to URLs of texts I find trustworthy: great newspapers, magazines, or academic papers…
Similarly, a lawyer or a doctor will feed the word processor with texts (from his own production, or found online) to be used as reference for professional vocabulary and turns of phrase. In my dream, third-party software vendors have seen a business opportunity: they sell industry- or occupation-specific plugins loaded with high-quality reference corpuses. This results in reliable auto-correct for Word and Pages. Some vendors even provide their corpuses as on-line subscriptions, constantly updated with state-of-the art content.
Then, as I write, the application watches my typing and matches it against the relevant corpus. Instead of relying on rigid hit-or-miss grammatical rules, it uses a statistical algorithm to analyze a word, or a group of words within their context of intended or inferred meaning. Take this gross mistake: “GM increased its sails by 10 percent”. The word is spelled correctly but, in this context, wrong. Because it lacks a context in which to detect the misspelling, the 1998-vintage word processor won’t change “sails” into “sales”. Conversely, the 2013 statistical-based language model flags the mistake by using the proper body of reference to see that “sails” is unlikely in an auto industry context.
Just a year ago, Google introduced Wave, an ambitious reinvention of email, seemingly ahead of its time. Among other advances, Wave featured a spectacular implementation of Google’s huge statistical model of language. In this video (go to the 45th minute) you’ll see Google Wave’s product manager Lars Rasmussen type the following sentences: “Can I have some been soup? It has bean a long time. Icland is an icland”, etc. Each time, the software automagically corrects the mistakes as they are typed, confident in the power of its algorithm and of its immense body of reference. This statistical approach works with gross, obvious mistakes, but also with more subtle ones.
Of course, I am aware of the difficulties in applying statistical language models to personal software: such algorithms are bandwidth and CPU intensive. This could explain why Google did not deploy the Wave spelling demonstrator on Gmail, or on Google docs. But the underlying algorithms do exist. A less sophisticated version, limited to professional dictionaries and thesauruses at first, could be fantastically helpful in properly spelling Zhengzhou, if you happen to write about Asia, or Neuroborreliosis, if you are a medical student.
Second, the use of texts. A significant proportion of writings goes to blogs and other social environments. As a serious user of the WordPress platform [today’s Word can’t even change WordPress into the correct WordPress, I had to check on Google…], I would gladly pay for a Word or Pages plug-in allowing me to compose a clean post with text, images, tables, links, typographical enrichments and, when done, letting me click “publish on my blog” or “send it to the mailing list”. No more cut & paste surprises or image resizing headaches. The word processor plug-in could be provided by the same developer who designed the style sheet (CSS) for my WordPress (or Blogspot, or TypePad) site. Or I could go for the auto-settings by inserting the CSS code in the plug-in that will, in turn adjust the word processor’s dials, from fonts and sizes, to background colors, etc.
You get my point: self-correcting spelling systems that guarantee (or at least vastly improve) decent grammar, syntax and the proper spelling of nouns and names can be a huge improvement for all professional writers – especially in a globalized economy where a greater number of us produce documents in a foreign language. Such auto-correct systems can even offer educational value in helping bloggers improve their basic writing skills.
I’m writing this on Word version 14 (yes, fourteen). How long will I have to wait for this quantum leap, Mr. Ballmer? Or Mr. Jobs?