Natural Language Processing & Natural Language Understanding: In-Depth Guide in 2021

Computers excel in responding to programming instructions and predetermined plain-language commands, but we are just in the early phases of them understanding natural language.

A simple command like “Hang up the phone,” for example, has historical and colloquial contexts that shape its meaning. The human mind understands this phrase quickly, but computers might not.

Fortunately, advances in natural language processing (NLP) give computers a leg up in their comprehension of the ways humans naturally communicate through language.

Success in this area creates countless new business opportunities in customer service, knowledge management, and data capture, among others. Indeed, natural language understanding is at the center of what Botpress seeks to achieve as a company—helping machines to better understand humans is the goal that inspires our development of conversational AI.

Although implementing natural language capabilities has become more accessible, their algorithms remain a “black box” to many developers, preventing those teams from achieving optimal use of these functions. Grasping the basics of how it works is essential to determine what kind of training data, they will use to train these intelligent machines. Selecting and applying the right training data is critical to success.

In this article, we review the basics of natural language and their capabilities. We also examine several key use cases and provide recommendations on how to get started with your own natural language solutions.

What Is Natural Language Processing?

Natural Language Processing is a subfield of artificial intelligence studying the interactions between a computer and human language. It's a field of study that combines linguistic and computer science. The purpose of NLP is to transform a natural language input into structured data. It uses a multitude of tasks to do that, such as; part-of-speech tagging, named entity recognition, syntactic parsing, and more.

What is Natural Language Understanding (NLU)?

Natural Language Understanding is about the comprehension of the language. Similar to us, the technology can hear or read something without understanding it. The NLU is the technology that powers conversational interfaces. Without the understanding part, the conversation is nearly impossible or at best awkward.

How Does NLU Work?

Like other AI solutions, this technology requires training. Intent detection depends on the training data provided by the chatbot developer and by the platform engineers’ choice of technologies. These specialists must supply training data to ensure the tool understands users within the context of its function—whether that function is servicing external customers or assisting internal users with knowledge management. Even with training, NLU will get lost as conversations steer away from its core functions and become more general.

Fortunately, these technologies can be highly effective in specific use cases. Optimizing and executing training is not out of reach for most developers and even non-technical users. Recent breakthroughs in AI, emerging in part because of exponential growth in the availability of computing power, make applying these solutions easier, more approachable, and more affordable than ever.

“To gain that understanding, machines need to be able to understand and generate parts of speech, extract and understand entities, determine meanings of words, and use much more complicated processing activities to connect together concepts, phrases, concepts, and grammar into the larger picture of intent and meaning.” Forbes, “Machines That Can Understand Human Speech: The Conversational Pattern Of AI,” June 2020

Language is complex—more so than we may realize—so creating software that accounts for all of its nuances and successfully determines the human intent behind that language is also complex. But as with human intelligence, sufficient training of AI enables a machine to overcome these complexities (if the training data is well-shaped enough).

Training AI has specific requirements unique to each AI’s use and context. For example, let’s assume we intend to train a chatbot that employs NLU to work in a customer service function for air travel. The chatbot will process the natural language of customers to help them book flights and adjust their itineraries.

In this case, a chatbot developer must provide the machine’s natural language algorithm with intent data. This data consists of common phrases travel customers may use to create or change their bookings. The natural language algorithm—a machine learning function—trains itself on the data so that the conversational assistant can recognize phrases with similar meanings but different words.

Ideally, this training will equip the conversational assistant to handle most customer scenarios, freeing human agents from tedious calls where deeper human capacities are not required. Meanwhile, the conversational assistant can defer more complex scenarios to human agents (e.g., conversations that require human empathy). Even with these capabilities in place, developers must continue to supply the algorithm with diverse data so that it can calibrate its internal model to keep pace with changes in customer behaviors and business needs.

To this end, a method called word vectorization maps words or phrases to corresponding “vectors”—real numbers that the machines can use to predict outcomes, identify word similarities, and better understand semantics. Word vectorization greatly expands a machine’s capacity to understand natural language, which exemplifies the progressive nature and future potential of these technologies.

Tips to build your dataset

• Stick to one concept per intent (an intent contains multiple utterances)
  
• Try mixing synonyms in the utterances
• Write your utterances with language your persona would use
• Use entities
• Avoid spelling and grammar mistakes

Here is our complete guide to build a training dataset for your chatbot.