No one reads documentation
It’s the year 2017 and people use apps for everything; no one reads documentation. The job of a technical writer has been replaced by an icon on a mobile device. Yet for many companies, documentation is still necessary to fulfil specific legal and safety requirements. Technical writers still strive to create consistent and usable documentation out of professional pride – documentation which no one reads.
Reading habits have changed for numerous reasons: some people are more discerning and only read what has been highly endorsed by others; others are less discerning and want “snackable content” that they can consume quickly and while on the go. It has become harder to capture readers’ attention because social conventions have an enormous impact on how people consume information. Technology coupled with modern lifestyles has led to fundamental changes in the relationship between writers and readers as well as within writing professions.
People read novels (sometimes)
Technology has also led to a decline in the purchase of printed media but is not to blame completely for the decline in overall reading; E-book sales have also flattened. The publishing industry in general is struggling. Documentation is a necessary by-product; it might be packaged as part of a product, but product users would rarely expect or be expected to pay for the documentation. By contrast, a novel is a product by and large written for commercial reasons and promoted to make money. For this reason, in a world where anyone can instantly publish their own book, publishers are eager to assert their relevance by discovering the next big bestsellers.
A bestseller formula
Studies show that algorithms can predict whether a manuscript will become a bestseller with astonishing accuracy. Plot, theme, and topicality influence the popularity of a book; however, the data analysed by the algorithms also has a strong focus on linguistic aspects such as style, rhythm, structure, and choice of words. When asked by The Guardian if plot is more important than style, Matthew L Jockers, co-author of The Bestseller Code, responded with an unequivocal “No” adding that “If your style is no good, no one will read it.”.
It has taken a machine to reveal through quantitative analysis of text that stylistic strategy and linguistic style play a significant role in the success of a novel. Yet even though these findings resonate with technical writers, the bestselling technical writer is and will remain an oxymoron.
The tide has turned
The quality of the text (the language) is the metric by which readers implicitly or explicitly measure the quality of the documentation and, in some cases, also the quality of the product it describes. Text is still the mostly widely used medium of the Internet and forms the basis of other media. Text can be compared, chunked and recompiled in different combinations, revised, structured, summarized, labelled. And text can also be processed by algorithms and used as data to train a machine to do the job of a technical writer.
Technical Writing vs. Machine Learning
If a machine can accurately pick out the bestsellers from 20,000 novels written over a period of 30 years, then the natural conclusion is that algorithms and, one day, an AI will write documentation. The question is only when. New Scientist predicts that a machine will be able to write a bestselling book by 2049; however, many experts are quick to point out that while algorithms help to gain insights into data, an algorithm or a program can only learn based on data fed to it by a human, unless it is a self-learning AI.
The old adage applies: GIGO
Garbage in, garbage out (GIGO) should possibly be the guiding principle of all documentation projects and especially those involving multiple language versions and computer-based translation. Machine translation not only requires a vast amount of data, but also defined rules and suitable text. The better the quality of the source, the less need for manual postediting. For people to absorb information effectively, information needs to be cohesive; the same principle applies to machines.
Technical writers can make a valuable contribution to machine learning, but they need to expand their traditional skills and see text for what it is, namely as data for machines, indexed by huge databases.
While machine translation is not strictly based on machine learning algorithms, parallels with supervised machine learning exist, including the need for experts with in-depth knowledge of language and text; knowledge that both technical writers and translators possess. If text is so crucial to machine learning, who better to train machine learning algorithms to create documentation than technical writers.
The Future of Technical Writing
Within five to eight years, our jobs will be taken by algorithms backed by huge databases with a humanized UI that speaks to you, that is, a digital assistant; this will open a new field for technical writers: natural language processing, a subset of computational linguistics. The task of technical writers will be to support natural language generation.