How does natural language generation work?

Natural Language Generation is a subset of Natural Language Processing and is the art of generating text from structured data using AI. It is defined as the process of taking structured data and turning it into text. It is a marketplace that is set to grow exponentially: recent research posits that the value of the NLG market is expected to increase USD 322.1m to USD 825.3m over the next five years.

NLG projects are formed using templates and conditions. Templates are essentially sentences with gaps that are filled using data and lexicalisation algorithms. Conditions are the circumstances that need to be met in order for the template to be utilised. The process happens through a software engine that arranges the templates into a determined order known as a ‘narrative’. The predetermined order that templates are arranged into is known as a ‘storyplot’.

Natural Language Generation does more. It is also able to use big data to spot newsworthy developments such as a sharp and unexpected increase in one value. This would trigger the production of a special story highlighting this shift. While a very basic use fills in gaps with data, more in-depth projects analyse the data, make sense of it, and draws conclusions. An example of this is financial reporting. Such a project involves looking at raw data and interpreting not only what the various financial events are, but ranking them in order of importance.

The system is also taught to combine different information (‘messages’), depending on the underlying data. Variation is made possible by adding alternative words, phrases, adverbs, and synonyms that the system can choose from.

How is NLG used?

In protection and Health Insurance Natural Language Generation can reduce underwriting time, being able to analyse data and produce reports on individual customers, giving a comprehensive view of the risks and the premiums that should be charged. It is a way to speed underwriting without compromising on quality or customer care.

Basic stock market reporting that uses NLG already exists, being able to analyse data and inform as to the best-performing and worse-performing stocks, their performances, and the respective indices on which those stocks are listed.

Figures such as EBIT, EBITDA, Revenue, Operating Expenses, and Net Income can be found, turned into text, and given proper historical context. Natural Language Generation systems in the field of Financial Reporting are also adaptable, meaning that they can produce different styles and types of reports, each reflecting the diverse needs of specific target groups.

One of the provisions under Mifid II is the beratungsdokumentation, a mandatory form for advisers that outlines why the advice they are giving to customers is right for them. This type of document could be generated automatically using NLG.

Chatbots are another feature of automation that is rapidly gaining traction. They can be created using NLG technology, which leads to nuanced and sophisticated dialogues between users and bots. Chatbots can be created by an external provider or through a piece of software, known as an enterprise platform, which is developed from outside and then implemented in-house.

Ecommerce providers use NLG technology that allows them to automate listings and product descriptions. The advantage that NLG has in this sphere is that it is possible to adjust texts wholesale within the blink of an eye by simply swapping or adding datasets.

Benefits of natural language generation for companies

NLG has many benefits for companies. On average, each company in the UK spends 120 days each year on administrative tasks such as filing taxes, writing reports, and sending out invoices. NLG solutions can automate many of these tasks, including:

Speed: Natural Language Generation reduces the time taken to produce text from minutes to milliseconds. Answers and processes reliant on data can be gleaned, gained, and utilised almost immediately.

100 per cent accuracy: Unlike human writers, NLG systems reduce errors. As good as the data they draw from, NLG systems do not make spelling, grammatical, or syntactical errors. This significantly reduces costs in the long term by reducing the need to audit every text produced.

Real-Time Monitoring: NLG systems are able to spot changes in real-time and recorded data, and react accordingly, producing narratives that highlight those changes and inform readers about them.

Empower Employees: An NLG system frees human workers from monotonous, rote work and allows them to do more creative, contextual analysis that is beyond the capabilities of an NLG system.

Optimise and Streamline Operations: Small businesses can use off-the-shelf automation products to curb expenditure. Instead of spending money using humans to perform some administrative tasks, these could be farmed out to computer programs.

Improving Customer Service: While there are some fears that NLG will replace workers, those with knowledge of the landscape believe that AI solutions like NLG will instead empower workforces. As Dutch banking giant ING were recently quoted by Financial Times, “We would like to use AI to bring smarter solutions to our customers, and be more effective in our decision making processes. As such, rather than ‘AI replacing workforce’, we believe in the power of ‘AI-empowered workforce’.”

Retresco has two NLG products. The first is self-service NLG, made possible through a platform, usually a web interface, designed to be accessible to non-developers. The second is fully-managed service, in which NLG projects are realised by a team of experienced developers, data experts, and language architects.

References & PDF:

http://www.inf.ed.ac.uk/teaching/courses/nlg/lectures/2011/NLG2011Lect1.pdf