© Glenn Carstens-Peters - Unsplash
Artificial intelligence is ubiquitous. Whether in smart factories, autonomous vehicle technology or streaming services, AI fundamentally changes the way we work and live. However, numerous users have often never heard of a specific area - even though they come into contact with it almost every day through apps, chatbots, smart assistants or often in numerous online shops: We are talking about Natural Language Generation (NLG) - a branch of artificial intelligence that deals with the automatic generation of natural language by software.
Where can NLG be used?
Natural Language Generation can be used wherever large amounts of structured and semantic data are generated - e.g. in e-commerce, on the stock exchange or in reporting for sports, weather or election events. More and more websites are using automatic text generation to provide targeted, personalized content. Some news portals already rely on so-called robot journalism. Here, too, NLG software is used in the backend to produce sports news, weather reports and stock market updates. The same technology is used by numerous real estate portals and online merchants to create offers and product descriptions for e-commerce systems - and the trend is rising. NLG software is also becoming increasingly relevant for marketing managers: In particular, marketing professionals who specialize in SEO and, for example, have to produce Google Analytics reports on a regular basis, know how much explanation is needed for some analytics KPIs - especially when the counterpart does not work with terms such as bounce rate, conversion or landing page on a daily basis. Natural Language Generation uses the data from Google Analytics to automatically create one-off or recurring reports and thus helps to better understand and classify information for interested parties and decision-makers at all levels.
80% Data, 20% Magic - How does NLG work?
Good examples for the meaningful use of NLG systems are those areas in which - as already mentioned above - a lot of structured, semantic data is generated. NLG takes these data and converts them into texts in natural language. The underlying processes include a number of technological and computational linguistic refinements: Large amounts of primary data - such as information about names, frequencies, product properties, measured values, etc. - are first added to the software's database. In order to create texts in natural language, however, so-called templates and conditions must first be defined. Templates are essentially gap texts, i.e. pre-formulated sentences that are enriched with a large number of variants, synonyms and adverbs using data and lexicalization algorithms. Conditions, on the other hand, are prerequisites that must be met in order for a particular template to be used. To automatically generate text, the system determines a relevant condition, combines information from the templates with event- or product-specific data, and creates the end product using intelligent linguistic analysis. By the way: NLG software usually also knows how to arrange the templates in an order - also known as "story plot" or "narrative" - so that they make sense for the human recipient and are easy to read. In the initial phase, human editors only have to create templates and define conditions - but once this information is available, the NLG system can work completely autonomously.
In the field of football reporting, computer-generated texts are produced in this way, which cannot be distinguished in the reader's perception from reports written by a (human) editor.
The potentials and challenges of NLG
NLG opens completely new possibilities for all those - including online shop operators, publishers or content marketers - who want to create large amounts of content faster and with less time and human resources. Automatic text generation can create personalized, variant-rich, unique and SEO-optimized content in real time and at the push of a button - scaling is only limited by the availability of the data.
NLG software can even be used to create texts in many different languages - either by translating an original source text (also automatically)