24-7 service to your customers
Natural Language Processing (NLP)
Breakthroughs within NLP algorithms make up the core innovation behind modern chatbots. Long and extensive research within conversational interfaces, enabling machines to understand human written language is nothing short of a scientific breakthrough. World-class universities have worked closely with large tech companies that can supply the training data required to develop these NLP algorithms. With these breakthroughs, we now see the same companies making NLP algorithms easy to access and use for third-parties and cater to their specific needs. Many tools have been developed to help entrepreneurs, companies and ideamakers alike to build a state of the art chatbot within their field. Google has API.AI, Facebook has wit.ai, Amazon has Alexa and finally IBM with their infamous Watson. All of which offer low-cost, easy to use yet incredibly powerful tools to connect intents and entities to groundbreaking NLP algorithms. The question remains, who do you think out of these companies will build the strongest NLP algorithms for the future?
Intents and Entities
When you think about it, any conversation you have ever had can be broken down into two categories: What was the intent and what entities did my intent consist of. Every time you have written something or opened your mouth, you had an intent to utter. Within this intent, the words that made up your intent can be broken down into entities – concepts that the other person needed to understand in order to understand your intent. A simple example: ‘Where is my dog’. Here your intent is to find your dog, and the entities that the receiver had to know to understand in order to answer on the intent would be ‘Where’, ‘my’ and ‘dog’. The exact same structure applies to conversations between humans and machines. Whenever a human writes or says something, the chatbot developed needs to understand the intents and entities the sender has. This is why good training material is so important. Whenever Wiredelta builds a chatbot the first question is always what training data the client has before starting any work. Training data can both be created or restructured if already created by the client or the customer’s that the client has. Regardless, training material is the first step towards building a chatbot that imitates humans and becomes almost indistinguishable from another human supporter.
Conversations are all about context. As mentioned above, all conversations can be broken down into intents and entities, but the same intent requires different answers depending on the context. To take the example mentioned above for ‘Where is my dog’. If you stand at a dog training facility amongst many other dogs versus if you just entered the vet because your dog had an accident, both contexts qualify for the intent, but the answers can be very different because of the context. Humans act differently when they are in different situations and with different people, it’s a natural part of life, and we automatically assume that machines have to respond differently when in different situations. This means that chatbots built to service humans need to be able to understand the context to provide a good user experience that imitates their human counterparts. If not, we can all relate to the ‘dumb machine’ experiences when a robot was supposed to replace a human but failed miserably! Don’t be that company that fails miserably, let Wiredelta’s experts build a chatbot that understands contexts as we do.