What is Natural Language Generation?

‍

Natural language generation (NLG) is a subsection within Natural Language Processing (NLP), the border domain that encompasses all software in charge of interpreting and generating human language.

NLG is the domain responsible for converting structured data into meaningful phrases in the form of natural language. It can communicate narratives in a human-like manner at an extraordinary pace, analyzing, interpreting, and summarizing thousands of pages per second.

Besides being able to generate unoriginal texts, NLG can also be used to write personalized content such as automated custom reports and custom web or mobile content.

However, although NLG can write, it cannot read. Natural Language Understanding (NLU) is the part of natural language processing that turns the unstructured data into understandable structured data.

*2019, “A Comprehensive Guide to Natural Language Generation.” Medium, Sciforce.*

The Evolution of NLG

NLG uses different writing structures depending on the context, audience, and purpose of the text it is given. NLG uses different methods to function in such an adaptable manner; however, each system continues to follow the same distinguishable three processes:

Document planning: conjuring an outline of what is to be said and structuring the information that will be displayed.

Microplanning: also known as sentence planning, microplanning is tasked with tagging words, expressions, and other nuances alongside fleshing out other document specifications.

Realization: Realization takes the specifications that were bundled up within the "micro-planning" phase and uses them to create natural language texts through using its knowledge on syntax, morphology, etc.

Although this outline demonstrates the pillars of natural language generation, specific approaches vary drastically with each new development of the technology.

Currently, NLG systems are incredibly advanced and capable of generating text in natural language. However, NLG came a long way to get to this point, from simple, straightforward templates to today's state-of-the-art system:

Filling in the gaps
One of the earliest stages of NLG, 'filling in the gaps' is a concept built on a template-based system capable of filling in missing words.

This somewhat simplistic template works for texts that already have an existing structure and only require small amounts of data to be entered.

Rule-based texts
The previously mentioned gap-filling approach was expanded upon through the addition of rule-based general-purpose programming structures. This approach integrates templates within general-purpose programming languages (such as JavaScript and Python) to enable the generation of statements such as complex conditionals. Though this form of NLG is a level up from simply filling in gaps, it still lacks the needed linguistic capabilities to generate complex texts.

Basic grammatical functions

A step forward from embedding templates within a general-purpose programming language was to add word-level grammatical functions to create more grammatically correct texts and better mimic natural language.

Dynamic sentence creation
Finally, a dynamic NLG, capable of generating complex natural language sentences, is created. This advanced form of NLG can carry out rational tasks without needing a programmer to explicitly write code for each individual case. The NLG is also able to optimize sentences by automatically including references, aggregation, connectives, etc.

Main NLG Applications

NLG creates fast and easy data-driven reports, product descriptions, memos, and so on, minimizing the burden analysts face when needing to summarize data and write such tailored reports. Therefore, it is no surprise that the current practical uses of NLG are centered around the writing and communication of necessary information. In the financial sector, for example, NLG is used in the following ways: