What Is Natural Language Generation?

Natural language generation is a software process in which an artificial intelligence application produces an output with language that is understandable in English or another spoken or written language. Many of the professionals within the technology world have spent the last 20 years instructing human programmers on how to speak machine language; however, they are now doing the opposite—teaching machines how to speak human language, which is the NLG process.

NLG has many real-world applications. When you interact with automated chatbots, such as Amazon’s Alexa or Apple’s Siri, they generate natural language to respond to spoken inputs. If you employ ChatGPT to write your cover letter, it uses NLG, among other machine learning and AI concepts, to generate those outputs in written language for your document. In this article, you’ll discover more about natural language generation, which is a part of natural language processing (NLP), and how NLG technology generates human language.

What is natural language generation?

Natural language generation is a subset of natural language processing. Natural language processing is a broader concept that describes the application in technology of natural language generation and natural language understanding (NLU). The early versions of NLP technology produced manually-coded systems that could perform some processing tasks, but could not scale to adapt to an ever-increasing volume of text or audio recordings.

Statistical NLP, which incorporated machine learning and deep learning into computer algorithms, allowed models to effectively classify language and establish an “understanding” of words, phrases, and grammar structure based on the statistical likelihood of the meaning of each element. Once the model “understands” these key components of language, it can then use natural language generation to generate human language and communicate with you.

How does NLG work?

As part of NLP, natural language generation commonly takes two different approaches: extractive and abstractive text summarization.

An extractive approach crafts an output by pulling sentences directly from the information in the input text. After the extraction completes, the NLG process uses the words and phrases from these sentences to generate an output that summarizes the text. The extractive approach can pull key ideas, phrases, or main ideas from the original document.

An abstractive approach generates new language by identifying key concepts in the original document. Unlike the extraction method, the abstractive approach must first understand the key concepts in the text before the NLG algorithm creates a summary of the original document. Operating in this two-step process makes the abstractive approach more aligned with how a human would study a text and then write the summary.

Some of the most effective text summarization applications also use semantic reasoning to include important context and draw insightful conclusions. Semantic reasoning is the process by which technology deduces the meaning of a single word by recognizing its use in various contexts.

Stages of natural language generation

NLG utilizes three stages—content selection, sentence planning, and sentence realization—to create a new text.  To accomplish this, AI or machine learning models generate language according to grammar, syntax, or lexicon rules. The NLG model starts at a conceptual level (content selection), then scales down to perfect language conventions such as correct spelling, grammar, and word choice (sentence planning), and finally, produces a chain of words in the form of a sentence (sentence realization). The following provides a closer look at the three stages of NLG.

Content selection

In this stage, the language processing model describes the content and selects information to organize into a standardized form, such as grouping related phrases together or creating a table with labeled columns.

Sentence planning

The model separates content into paragraphs and sentences, incorporating grammar and punctuation. Pronouns or conjunctions replace other parts of speech where applicable.

Sentence realization

The final stage of the language generation process involves checking for grammatical accuracy and refining sequences of words. The model properly conjugates words—for example, “catched” would change to “caught.”

What is NLG used for?

One of the most common uses for NLG is text summarization. This process allows you to understand large volumes of information much more quickly than manually reading and processing information from media such as books, videos, or articles.

The advent of ChatGPT has put NLP, and its component NLG, on the map as part of the hottest new technology, which can contribute to your business operations, assist and even automate tasks, and help your employees generate content. You can also use NLG  for many more applications in your daily life, such as chatbots, voice assistants, and writing blogs.

Real-world examples of NLG

Conversational AI platforms: Technology you may use daily involves language generation. Conversational AI platforms such as Siri, Amazon Alexa, or Google Home use natural language understanding (NLU) to analyze spoken input and then use NLG to generate a helpful response.

Image-to-text generation: NLG can neatly draw conclusions from charts or graphs that you, as a human, might miss. These models analyze large amounts of data, deciding what to highlight and what to discard, to convert that data into language that is easy to read and understand.

Translation: Accurate translation involves much more than converting each individual word from one language to another. NLG applications can consider tone and figures of speech when translating language to make it as natural as possible for you, the reader.

Accessibility: NLG applications can aid the visually impaired in understanding information presented visually in its original format. One technology known as the SIGHT system converts images to text descriptions, including audio summaries, to aid people who are visually impaired.

Who uses NLG?

Individuals, companies, and organizations can use NLG for a variety of purposes, such as virtual assistants, chatbots, business intelligence, and reporting generation. If a career involving NLG interests you, it is encouraging that NLP, as a whole, is a growing interest in business contexts. A 2023 report found that 77 percent of businesses surveyed planned to increase their spending regarding NLP in the next 12 to 18 months [1].

If you have an advanced skill level in digital technologies, you are seeking work at an opportune time because companies are hiring people with your skill set at a higher rate than in past decades. The share of workers in occupations requiring skills in digital technologies rose from 4.8 percent in 2002 to 23 percent in 2016, according to a report by The Brookings Institution [2]. The report claimed that technology-producing firms must hire professionals capable of training AI systems such as natural language processors.

Top companies for NLG Engineer positions include Apple, Bloomberg, Google, and Bosch.

Job titles related to NLG, along with their average annual salaries, include:

• NLP engineer: $122,889 [3]

• NLP research engineer: $138,373 [4]

• Machine learning engineer: $125,748 [5]

Pros and cons of natural language generation

Pros

Document summarization

Natural language processing, with NLG being a step in that process, can summarize a document instantly. It can take humans a great deal of time to read a document, comprehend its information, and grasp key concepts or main ideas. NLP saves time and can make documents packed with information easily understood by humans.

Convenience

Virtual assistants, such as Apple’s Siri or Amazon’s Alexa, use NLP applications to answer your questions and perform tasks simply by listening to your voice. Within the NLP process, the NLG step helps to generate these answers. These technologies can tell you how the weather looks or even order a product simply by listening to voice commands.

Return on investment

As a business owner, you might be particularly interested in NLG because it, as a part of NLP, can replace human tasks, making your business more cost-effective regarding your budget. Business activity with NLP models is increasing because of the potential for AI to maximize every dollar spent on work pertaining to content creation, record keeping, data summarization, and more.

Cons

Lack of human touch

Though NLP technologies can produce outputs that humans easily understand, they can often seem shallow or lack a human touch. NLP may not always generate language in a way that feels similar to how you, as a human, would write or talk. You cannot rely on NLG to complete complex tasks, or to outright replace human language tasks.

Ethical concerns

The high-quality outputs of ChatGPT—an NLP-based technology—high-quality outputs make it difficult to ascertain whether a human or technology developed it, which leads to an ethical issue in the realm of academics or the development of content in any context. The elimination of jobs is another concern surrounding the rise of NLP applications; for example, this technology, with its ability to generate content, could replace the work completed by computer programmers.

Security concerns

Plagiarism and unauthorized access to data, especially proprietary business information, are concerns for NLP models. An NLP model can access your business data when people use it to generate content or summarize documents.

How can you start a career in NLG?

To pursue a career in NLG, you will most likely need a bachelor’s or associate degree in a related field, such as data science, computer science, or engineering. Some employers may prefer advanced degrees, such as a master’s degree or PhD, for higher-level positions.

If you’re looking to build a foundation of knowledge for a career in NLP, which includes NLG, it’s a good idea to develop your skills in areas such as data structures, artificial intelligence, and computer logic.

Finally, as you develop your abilities, Python, a popular coding language for AI practitioners, is essential for pursuing your career in NLP. Python works well with NLP projects because of its simple syntax and semantics. It contains many libraries that optimize NLP projects. Java and R are also important coding languages for NLP use.

Getting started with Coursera

You can take the next step toward a career in natural language processing with DeepLearning.AI. The following course on Coursera includes lessons on several aspects of AI technology, such as sentiment analysis, text generation, and named entity recognition. Learn more about natural language processing and generation with DeepLearning.AI’s Natural Language Processing with Sequence Models, which is the first part of the Natural Language Processing Specialization.