If machines could understand the language we humans use, it will be a step in the direction of accomplishing the tasks. This is what we mean by NLP which uses a lot of machine learning. Here a pre-trained language model (PTLM) is used. A machine is made to read a lot of text, possibly in billions, to understand the syntax and come out with better responses to a question. The size of the language model is measured by the size of training data. In 2018, Google’s BERT was trained in 3 billion words. GPT-2 by OpenAI, San Francisco was trained in 40 billion words. GPT-3, its advanced version is based on 500 billion words. All areas of NLP benefit from these models. Information extraction, question-answer and summarisation all use PLTMs.

Chatbot is another example of NPL. It is able to read a conversation, understand what the user said, and construct new sentences to keep the conversation going.

Information extraction is another example of NLP. Here input is converted into information.

NLP can do sentiment analysis — understand the mood of the user. It can do topic modelling — discovering abstract topics. It can do text categorisation — arrange text in organised groups.

These days there is research in progress on multi-lingual NLP where systems converse or understand or interact in multiple languages at the same time.