Chatbot ABCs: An Introductory Glossary of Terms

Chatbots, Artificial Intelligence, conversational UI, Natural Language Processing — what does it all mean?

As an emerging technology set to have a huge and lasting impact across sectors, chatbots are getting a lot of press right now. Along with that coverage, we’re seeing a rising number of questions around the technology and just what it all means. In an effort to simplify the learning process and be sure everyone is on the same page when it comes to definitions of the more common terms and phrases in the chatbot world, we offer this brief introductory glossary. We’re limiting it to our top 9 definitions, as these are enough to get you started and open up deeper learning possibilities as you continue researching your chatbot options, without causing overload.

Chatbot

It’s always a good idea to start at the beginning, right? A chatbot, at it’s most basic level, is a computer program designed to converse with human conversation partners. With a history going back to the late 1960s, the underlying technology is by no means new. What is new, and responsible for the explosive growth in chatbots in recent years is the advent of a truer form of AI (see below) and the corollary developments that have flowed from it. Chatbots can carry out basic conversations, answer FAQs, show site visitors where to find the information they need, and collect user information for the purposes of filling in forms. They can also retrieve data from a variety of internal systems, compile reports, and even generate alerts based on an evolving set of definitions.

Artificial Intelligence (AI)

There are myriad definitions for AI, ranging from simplistic to so detailed and comprehensive as to require a computer science degree to parse. We’re going to go halfway and say that artificial intelligence is the branch of computer science concerned with creating computers that act like humans. Not necessarily in a physical way, rather that these computers and programs can “think” like a person. In the context of chatbots, AI and machine learning techniques are responsible for creating the algorithms that allow bots to “learn” how to generate unique responses, use different vernacular to match their conversation partner, and evolve responses to questions not present in their programming.

Natural Language Processing (NLP)

Combining computer science with linguistics and data science, NLP is the AI specialty concerned with teaching computers to talk more like humans. By feeding large amounts of data (primarily from chat logs) into analytics engines that parse the logs looking for ways to improve the algorithms, these data scientists are making human-bot conversations flow more naturally. The ultimate goal of NLP is to make human-computer conversations as close to human-human ones as possible. This includes things like teaching the computer to use the same syntax and slang as their human conversation partner.

Natural Language Understanding (NLU)

A sub-specialty within NLP, natural language understanding is the field that works toward helping computers parse the intent behind the words spoken or typed by humans in their interactions with computers like chatbots. For example, if you asked Siri, “what’s traffic like this morning?” NLU is what tells her your primary concern is for traffic between your home and office, not across town. It’s this understanding of context, syntax, and intent that is leading to the greatest chatbot breakthroughs.

Conversation as a Platform (CaaP)

Increasingly the public is relying on messaging apps as their primary interface with friends and family, as well as with the companies they do business with. This has opened up whole new ways to interact with customers as CaaP has come to dominate the customer experience. Whether it’s your chatbot on Facebook Messenger, or a live CS agent chatting with someone on the tech support page, the emergence of CaaP is opening new vistas for engaging with and delighting existing and potential customers alike.

Platform (or Channel)

There are many diverse definitions of the word “platform.” In discussions centering around chatbots, it most often refers to the website or app that a bot is deployed to. So Facebook Messenger is a platform, as is your customer portal or dedicated chat interface on your homepage. Platform is often used interchangeably with channel to describe the different ways people interact with one another online. Each social media site, networking app, and page on your website counts as a channel for our purposes as a chatbot can be deployed to each with equally stunning results for customer satisfaction.

Conversational UI/UX

User interface (UI) and user experience (UX) are not new concepts. What is new here is the creation of UIs based solely on humans having conversations with computers. A chat window on your portal or a user asking Siri for directions are examples of conversational UI. And it’s down to the development of more advanced NLP algorithms to define how good a UX these interfaces can provide. Chatbots are appearing at an unprecedented rate, with thousands of new deployments every week. What sets the great apart from the mediocre is how well designed their UI is and how much effort goes into the UX overall.

Selective Chatbot

There are two conversational models for chatbots, selective and generative. Selective bots were the first to be developed and rely on more basic programming to process a person’s question and select a scripted answer from their database. These databases are populated by chat logs and input from the human agents involved in the same conversations. Machine learning plays less of a role in this model, keeping deployment quick and relatively hassle-free. The other side of that coin is that the conversations tend to feel stilted and the answers don’t always match up with the intent behind the questions.

Generative Chatbot

The other model of chatbot is called generative because it uses more advanced forms of machine learning to generate unique answers to each question asked rather than picking a pre-scripted response. This requires more computing power as well as more involved algorithms that enable the bot to learn from each response it gives, growing its database organically and allowing future answers to be further tailored specifically to each question asked.