Combining computing technologies with human language has become a driving force for modern-day technology.
The experience of using a smartphone, for example, wouldn’t be quite the same without the ability to pull up a map with a computerized voice navigating your next turn. Tools like Google Lens, which can translate words captured by a camera on the fly, would not be quite as impressive.
These tools represent just some of the power of natural language processing (NLP), a form of artificial intelligence that promises to have use cases far beyond smartphones.
For businesses, the ability to process speech and written words in real time could prove essential as organizations hope to better understand consumer and employee sentiment, analyze data and automate tasks that once required careful manual analysis.
Still, we may be only scratching the surface of NLP.
What Is Natural Language Processing?
At a high level, natural language processing describes a computer’s ability to process and comprehend language, whether in written, spoken or digital form.
It’s often thought of as a very recent capability of computers. In fact, however, NLP dates to the earliest days of computers. For example, early optical character recognition systems relied on specialized fonts that computers could detect.
Today, natural language processing is seen as mainstream and practical, with AI-powered smart assistants such as Google Assistant, Apple’s Siri, Amazon Alexa and Microsoft’s Cortana well established as mainstream use cases.
AI has become crucial in business as well, and NLP is seen as a major area of growth for many companies’ AI strategies. The Global AI Adoption Index 2021, an IBM Watson project, found that nearly half of businesses are using some form of NLP technology, with another quarter of businesses expected to use it within the next 12 months.
“The top use cases for NLP today — improving the customer experience and helping employees reach new levels of productivity — are critical priorities for nearly every business,” says Dakshi Agrawal, an IBM fellow and CTO for AI at IBM.
What Are the Steps in NLP?
The steps involved in natural language processing start with having access to data in its original form (a written message in a database, for example) and a language base to compare it with.
After the data is collected, the information is broken down using several data preprocessing techniques. Among them:
REGISTER: Learn more about leveraging AI to drive business initiatives in the weekly CDW Tech Talk series. Click the banner below to register.
After preprocessing, the data is analyzed using a variety of AI techniques, such as machine learning, to deduce meaning in a given use case — such as what a customer is asking for when calling an automated phone system. That information can then be acted on in that use case.
Different technology firms have different approaches to AI and NLP. IBM approaches AI through a four-step system it calls the AI Ladder, which involves collecting, organizing and analyzing data, then spreading the lessons of that data throughout the organization.
While the technology tools matter, Agrawal emphasizes that humans should also play a role in determining the result of an NLP use case.
“Every job function that touches the AI application — from the end users in the line of business to the application developer and the production management — should understand and have a clear view of how they can assist and be assisted by AI and NLP,” he says.
What Are Some Top NLP Techniques?
Agrawal notes that there are three different types of techniques used in natural language processing disciplines:
- Deep neural networks, which can be used to model information and determine a preferred result in a given use case
- Machine learning and other traditional approaches to AI, which rely on the use of “training data” to make decisions based on statistical methods
- Rule-based techniques, which make decisions based on a given set of parameters
“Among these, deep neural network-based techniques tend to get the most attention,” he says. “However, a business building out its NLP discipline should take a balanced perspective with use case-specific requirements and long-term maintenance of NLP pipelines. Ultimately, a business should select the right technology that fits the use case.”
What Are Examples of Natural Language Processing?
For IT teams, one good use case for natural language processing is document classification. Such classification might be good for the basic sorting of information, but it can also have uses in security. As Microsoft explains in the documentation for its Azure cloud platform, NLP can help identify whether an email is spam or is in some way sensitive, offering a potential way to filter out dangerous or problematic information before it reaches end users.
“The output of NLP can be used for subsequent processing or search,” the company explains.
Another area where NLP can come in handy is business analytics, allowing users to look for information using common phrases rather than having to adjust their wording to what the search engine or business intelligence tool will understand.
Personalization is also an important use case for many companies, with its use seen as a major element of understanding customer sentiment and offering services tailored to their needs.
NLP also can help analyze large databases to gather a deeper level of intelligence for making big decisions, a use case that carries lots of potential for scaling up. IBM Watson currently is being used to help manage an AI-driven stock index that evaluates potential investments based on in-depth analysis of data gathered on the largest publicly traded corporations. Right now, it is outdoing the S&P 500 by nearly 5 percent.
NLP vs. NLU and NLG: What’s the Difference?
Natural language processing can take on a variety of forms, but all are generally driven by two subsets of NLP that have similar names, sometimes used interchangeably. However, the use cases are significantly different.
Natural language understanding (NLU) refers to the comprehension and parsing of text or speech in a way that can determine the meaning of information in real time — and even grasp the emotions of the person writing, a boon for companies seeking to improve customer service.
“By reviewing comments with negative sentiment, companies are able to identify and address potential problem areas within their products or services more quickly,” Agrawal says. “NLU can also be used to spot trends in customer feedback to address concerns more quickly, reduce churn, improve the customer experience and drive revenue.”
In many ways, the difference between NLU and natural language generation (NLG) is the difference between the production of language and comprehension. Agrawal says that NLG “essentially enables computers to write.” Often used in chatbot applications, NLG is useful as a way to effectively automate responses in a natural way, and it can even be combined with text-to-speech technology to allow for voice conversations.
“Key benefits of NLG include automating redundant or mundane tasks and enabling higher levels of personalization at a greater scale,” Agrawal adds.