Natural language generation (NLG) is the field of study concerned with computer systems that produce original, coherent text. You may notice that this definition doesn’t specify an input, rare for a technology that relies on machine learning. That’s because inputs for various NLG models can vary widely, and may include:
- Pre-existing texts, in which case the output may be a summary or paraphrase, e.g., Quillbot
- Images or video, as in software that applies text descriptions to visual data, e.g., Microsoft’s Seeing AI
- Structured datasets containing text, numbers, or both, e.g., Arria NLG, Narrative Science’s Lexio, and Automated Insight’s Wordsmith
We’ll focus on the last category, as most of today’s business applications use natural language generation from structured data. So how is NLG being applied to business challenges? Below are just a few natural language generation examples, organized by the capabilities they represent.
Pair natural language generation software with text to speech (TTS) from ReadSpeaker to create powerful voice experiences for your customers. Interested? Contact us to start the conversation.
5 Natural Language Generation Examples
1. Deep, Automated Personalization
Natural language generation software can personalize communications at a scale limited only by your customer data—and that goes a lot deeper than addressing recipients by name. The Orlando Magic organization used NLG to email user-specific advice on how to get the most out of its loyalty program based on the customer’s past usage. These emails improved engagement, with a positive reception rate of more than 80%. With NLG, marketers can apply similar personalization to chatbots, conversational IVR systems, SMS messages, and more.
2. AI-Generated Narrative Reporting
In 2018, the Associated Press used NLG to “write” more than 5,000 previews of NCAA Division I men’s basketball games—articles no sportswriter would have the time to file. And if a natural language generation algorithm can write basketball previews, imagine what it could do for financial reports. Or monthly marketing reports. Or any reports for which you have organized datasets, for that matter. The use of NLG to help convey business intelligence (BI) has spawned an emerging discipline—data storytelling—which purports to make big data analytics accessible to everyone.
3. Advanced Monitoring for the Industrial Internet of Things (IIoT)
Add an NLG module to IIoT infrastructure to receive automated status reports, maintenance updates, and other system analytics written in plain language. This technology is already being used across a range of industrial sectors, from utilities to shipping to manufacturing. One such NLG-powered system demonstrated the potential to almost double the speed of alarm-processing in a manufacturing IIoT system, suggesting the possibility of savings worth millions of dollars a year.
4. Content Creation at Scale
As it stands today, even advanced NLG isn’t a replacement for human bloggers—but it’s already being used to augment our composition. One familiar example of this is Google’s Smart Compose, which suggests words as you write sentences. However, today’s NLG systems are already being used to scale up production of shorter, more data-heavy content. In 2020, online children’s clothing dealer Babyshop used an NLG service to generate its product descriptions. Testing has shown that its NLG content converts traffic at similar or even higher rates than copy written by humans.
5. Conversational AI
Natural language generation is a crucial part of conversational AI systems like chatbots, voice user interfaces, and smart assistants—and these voice technologies are an essential tool for business. In 2021, nearly 92% of marketers considered voice assistants an “important” marketing channel, and almost 30% characterized the technology as “extremely” important.
Interestingly, composing responses is just part of NLG’s potential role in conversational AI. An NLG module could also generate lines of code—using Speech Synthesis Markup Language (SSML), or something like it—that instructs a text-to-speech (TTS) engine to express words more like a human speaker. For example, if the system’s dialog manager determines that a response needs to contradict a user, the NLG module could instruct the TTS engine to emphasize a certain word: “Your account is active!”
Even more exciting is the potential for emotionally responsive conversational AI. A TTS markup language with tags for emotionality—levels of excitement, regret, hilarity— would create the potential for extremely human interactions with our machines. With NLG sending instructions, TTS engines may soon express excitement (“Your account balance has gone up!”) and sympathy (“I’m afraid it’s too late for a withdrawal today”) within the same conversation, just like a human agent.
In previous blogs, we’ve discussed the dawning possibility of emotionally responsive voicebots. This is how they will work. Natural language generation and lifelike TTS from ReadSpeaker are ushering in a generation of voicebots that will scarcely feel robotic at all. Listen to sample ReadSpeaker TTS voices here.