Natural Language Generation (NLG) includes techniques for translating information into text. NLG is a subset of Natural Language Processing (NLP). If natural language processing techniques can be separated into two, Natural Language Understanding (NLU) is the part that analyzes natural language, NLG is the part that synthesizes meaningful text for human consumption. NLP=NLU+NLG. For example, let's imagine that you are writing to a conversational AI solution. When you ask what the weather is like today, NLU techniques are used first to understand your intent. Weather data is then searched, the results are analyzed and an answer generated thanks to NLG techniques.

NLG techniques can be used to automatically summarize and explain data in a human-like manner. NLG can automate the transfer process of analyzed structured data to humans in seconds.

NLG saves time and costs as well as contributes to scalability and consistency via automating repetitive processes involving language.  

• Time and cost savings: NLG reduces time and costs as it speeds up the reporting processes.
• Reporting consistency: Reports can be precise and standardized as though they are created by one person. Accuracy of the reports can be improved by time as NLG solutions advance.
• Scalability: By automating parts of the reporting and content generation processes, it becomes possible to create large amounts of content.

Important factors in evaluating an NLG solution are: evaluating use cases, the existence of suitable data sources, security, language capabilities and vendor dependency.

• While evaluating NLG solutions, it is necessary to find vendors that provide solutions for a particular use case. For example, an NLG solution trained to prepare content in the field of e-commerce would not be suitable for reporting the efficiency of IoT devices.
• NLG solutions can be as good as the underlying data. The project team needs to ensure that they have the structured data which can be combined with the NLG solution to serve users.
• The audience of the reports or content is important. Publishing public content is always risky and would require a more robust quality assurance process.
• In situations where language skills are important, a robust assessment of the NLG tools' language capabilities is necessary. For example, correct use of language is far more important in media and journalism compared to a company's internal reporting.
• Large-scale enterprises can build their own NLG engines using open source solutions and minimize vendor dependency. Two aspects are important in this decision: 1) Is the use case highly specialized? If yes, it may be harder to find an NLG solution from a vendor for that case. 2)Is this a simple or difficult NLG problem? Simple problems can be easily resolved with open source packages.