Text automation – the future of writing?


At text2net, we offer automated text creation alongside other content management services. Interestingly, in business and private conversations on this topic, we repeatedly get the same reaction particularly from people who write professionally themselves (e.g. copywriters, journalists), ranging from being shocked to being outright dismissive.

“Machines can really write texts?” “These texts can’t be any good, can they?” “A danger to the writing profession and an assault on creativity!” As involved party, we know text automation from professional practice, and perhaps we can mediate a little.

Texts off-the-rack?

What is text automation all about? It boils down to a complex software being trained to generate semantically meaningful and syntactically correct sentences for a certain set of given data. This is made possible by Natural Language Generation (NLG). Natural language is processed computer-based with the help of rules and algorithms. NLG takes into account methods and results from linguistics and enhances them with modern computer science and artificial intelligence (AI).

Dealing with relatively short, similar, less complex texts in large numbers, text automation can result in significant time and thus cost savings. Prominent examples are product texts, stock exchange news, weather reports, but also sports reports. For this purpose, text frameworks with different variables are created, making possible the generation of unique (SEO!) texts in large numbers within a very short space of time. Once created, these text templates can easily be adapted as needed and used again and again to generate new texts.

What was still a dream of the future not so long ago is now so good that properly edited, automated texts can no longer be distinguished from manually written ones – and the learning machines get better with every task. Even automated translations into pretty much any language are already feasible.

Writing without a future?

Is this now a threat to novelists, essayists, journalists, copywriters and bloggers? First of all, no. Rather, this is often the subjective fear of a profession that doesn’t feel appreciated and that keeps on hearing that it is dispensable (“You can really earn money with that?” – “Everyone one can write!”). This is damaging to self-confidence – no wonder digitisation quickly appears to be a superior enemy. Like the craftspeople in industrialisation, members of the writing guild already see themselves sitting on the street, replaced by more efficient but bland machines. But be assured, that won’t happen quickly.

Every stand-alone, longer, and deliberate text has a value that no machine has yet been able to generate. The engine simply cannot part ways with the data framework. Or to paraphrase Thomas Ramge, Weizenbaum Institute Berlin (Research for the Networked Society): “In the data-free space, AI systems are disoriented. Authors who have the ambition to conceive and write down new ideas have nothing to fear for the time being.” (Source)

An engine also does not (yet) have a feeling for language. This can be well illustrated by a simple adjective like “great”: If we tell the engine to design variants for “a great pair of trousers”, it looks for synonyms for “great” – and there are quite a few, some of which fit “trousers” well, some not. For example, “significant” is probably a bit overstated for a pair of pants, and not every customer wants to find characteristics like “swell” or “smashing” in their product description, even though these are undoubtedly synonyms for “great”

Symbiotic coexistence

As of now, none of the creative writing jobs mentioned above is in serious danger. Rather, it is the workers who offer their services on platforms to make ends meet by writing hundreds to thousands of short texts in several variations who are affected. A Sisyphean task that is not really satisfying.

This is where the hybrid content editorial model can make sense, as Saim Alkan from our partner AX Semantics explains. This can also lead to a win-win situation economically, because the integration of NLG allows the editorial range of texts to be scaled significantly. These additional revenues can in turn be invested in manually created articles. So doing, high-quality journalism can even be promoted.

The NLG provider Retresco, with whom we also cooperate, emphasises the importance of text engines as well, for example in the coverage of election and sporting events, but also – very up-to-date – for news formats such as daily corona and vaccination updates.

For the foreseeable future, the use of text robots will not make humans superfluous. On the contrary, only with linguistic competence and technical understanding can humans set up and train NLG software in such a way that the quality of the automatically generated content is sufficiently high and worth reading.

That’s why text automation offers completely new opportunities and possible uses! And these tasks are far more challenging and appealing than writing a thousand product texts about pants.