Natural language processing is the hottest area of artificial intelligence (AI) as “huge models, large companies and massive training costs” dominate the arena. And a new generation of transformer language models are unlocking new NLP use-cases.
AI investors Nathan Benaich and Ian Hogarth shared those findings in their State of AI report. NLP applications, which enable machines to analyze, understand and manipulate language, continue to expand their footprint and are implemented in Google Search and Microsoft Bing, according to the report.
So what does the surge in NLP use cases and technologies mean for AI in marketing? It’s huge if you ask Paul Roetzer, CEO and founder of the PR 20/20 and the Marketing Artificial Intelligence Institute.
“Major advancements in natural language processing equals innovation within the marketing space, because it makes the understanding and creation of language possible,” Roetzer told CMSWire. This leads, Roetzer added, to outcomes like AI being able to write a first draft, something not possible without these advancements in NLP. “Advancements in NLP,” Roetzer said, “mean trailing advancements in the marketing space and the use of AI.”
Why Language, Vision and Prediction Matter for Marketers
Roetzer’s research for the marketing use cases for AI finds three main current buckets: language, vision and prediction.
Specifically, through analysis of the mega vendors in the arena, he reports uses cases available for marketers this way:
- Natural Language Processing
- Natural Language Generation
- Sentiment Analysis
- Text Analysis
- Text Extraction
- Text Generation
- Voice Generation
- Voice Recognition
- Emotion Detection
- Image Analysis
- Image Recognition
- Facial Recognition
- Movement Detection
- Video Recognition
- Pattern Recognition
“The majority of the tools that we see with immediate value for brands are what I would consider language-based tools,” Roetzer told CMSWire. “They're tools that are using variations of natural language processing, which is just understanding language, and then natural language generation. And then on top of that, you can do things like sentiment analysis of language. This can be the spoken word or it can be the written word. You can do conversion of speech to text, like when you record a Zoom meeting, and then that transcribes with a tool like Otter.ai. They're using forms of natural language processing. Grammarly is built on natural language processing. All these companies, the base of what they're doing is natural language processing. GBT-3, which we hear a lot about, being able to create language, it's reliant on advancements in this area.”
Related Article: Cutting Through AI Marketing Hype: It's About Machine Learning
A Matter of Scale
When it comes to the NLP use case in marketing and arenas like customer experience, it’s very much a matter of scale, according to Jim Sterne, author and keynote speaker on AI in marketing and founder of Rising Media.
“If you have a huge call center with a massive number of phone calls and emails coming through every day, then NLP is crucial for keeping on top of changes in Voice of Customer opinions and attitudes,” Sterne said. “It's also useful if you're tracking a large number of social media posts and comments.”
As well, it can be helpful when considering messaging, Sterne added. “Keeping on top of the vernacular of your target audience is a tough challenge,” Sterne said. ‘But with NLP, you can get a feel for how language changes over time.”
NLP and the Connection to Marketing Analytics, Search
One of the most important things to have happened in marketing over the last 20 years is the advent of digital tracking and analytics, and the next frontier of tracking is understanding intent and sentiment of users, according to Justin Schmidt, VP of marketing at Capacity, an AI-powered helpdesk tool.
“NLP,” Schmidt said, “can parse intent and sentiment in visitors as they chat with a bot or use your site search. This will usher in a new era of engagement and understanding. So yes, I strongly believe marketers should embrace products that have good NLP and leverage those products to create new levels of engagement with the consumer.”
The most powerful NLP is also one of the most important sources of traffic for your site, Schmidt said. “Google," he said, "is at the absolute tip of the spear with NLP and understanding human language, and leveraging that understanding to deliver value to the user. This has implications for how you SEO your content. If you have a nuanced point of view with your product or service, then you should feel more free to describe and write about that offering as you would for another human. Keep your keywords relevant in your content, but don’t write for the machines. Write for your audience.”
Related Article: 8 Considerations When Selecting an AI Marketing Vendor
It's Still a 'Few Years Away' for Marketers
Is NLP right now the most common/practical use case for marketers? Maybe not.
The current AI renaissance can be traced back to 2012 when a team of researchers from the University of Toronto built a convolutional neural network that won over the outperformed humans at identifying objects in pictures for the first time, according to Brandon Purcell, Forrester analyst who serves customer insights professionals.
2019 saw a similar moment in the field of NLP, he said, when a machine bested the General Language Understanding Evaluation benchmark for the first time. This past year, advancements in NLP have continued with the evolutions of BERT, ELMO and GPT-3.
“For marketers, this may sound like great news,” Purcell said, “but the truth is that NLP adoption by marketers today is low and use cases are sparse. NLP works its magic on large sets of unstructured text data. In most companies, the customer service or customer experience team owns the largest set of customer-generated text data, and they apply NLP to this data to identify and remedy customer pain points. Marketers typically don’t have access to this data, and most of them are struggling to wrangle their structured customer data into the elusive 360 degree customer view.”
Purcell finds that eventually, as companies break down internal silos and marketers get access to more unstructured text data from customers, they’ll be able to tie this data back to behavioral and transactional data on these customers. This, he said, will allow them to derive richer insight and improve predictive models.
“But for most organizations, this is a few years away,” Purcell added. “Today, marketers who wish to experiment with NLP can apply it to their marketing content to tag it consistently and analyze which types of content perform best with different segments of customers. Or, they may even turn to companies like Persado that use natural language generation to create optimal text for different types of customers.”
What’s in the Public Domain?
Neil Yager, chief scientist at Phrasee, an AI copywriting tool for digital marketing campaigns, noted that there is a significant cost to training massive language models, so only large companies with significant R&D budgets can build them in-house.
“However, many, if not most, of these models are released to the public domain,” Yager said. “GPT-3 is the exception. However, even for GPT-3, there will eventually be paid access to the model via an API. Therefore, even small companies with limited budgets and resources can access these massive and powerful models. It is a very exciting time for NLP.”
NLP, Yager finds, has not traditionally been the main application of AI for marketing. So far the focus has been on personalization, audience segmentation and targeting, recommendation engines, etc. “Some areas where NLP has been used are sentiment analysis and information extraction,” he said. “These techniques can be interesting but are not necessarily impactful or revenue-generating.”