Natural language processing explained: from text analytics to large language models
NLP is the field that taught computers to work with human language. Chat assistants are its most visible product, but most enterprise uses are quieter.
Markus Winkler / UnsplashNatural language processing, or NLP, is the branch of computer science and AI that lets computers understand and produce human language. IBM describes it as combining computational linguistics with statistical modeling, machine learning and deep learning.
Until a few years ago, NLP meant separate systems for separate jobs: one model to classify emails, another to pull names out of contracts. Large language models changed that. A single model can now do many language tasks from a plain instruction.
The core tasks
The building blocks of NLP still matter, because they explain what language tools can and can't do reliably.
| Task | What it does | Enterprise example |
|---|---|---|
| Named entity recognition | Finds names of people, companies, places, dates and amounts | Pulling party names and dates from contracts |
| Classification | Sorts text into categories | Routing support tickets to the right team |
| Sentiment analysis | Judges whether text is positive or negative | Tracking customer feedback trends |
| Coreference resolution | Works out when different words refer to the same thing | Knowing "the supplier" and "Acme" are one company |
| Word sense disambiguation | Picks the right meaning of an ambiguous word | Telling "Java" the language from "Java" the island |
| Summarization | Condenses long text | Meeting notes, incident reports |
| Translation | Converts text between languages | Localizing support content |
How large language models fit
According to IBM, NLP research is what made today's generative AI possible. Large language models such as GPT, Llama, Claude and Mistral are a deep learning approach to NLP: they are trained to predict the next word, and at large scale that training produces general language ability.
For IT buyers, the practical change is that many tasks that once needed a custom-trained model can now be done by prompting a general model. That lowers the cost of trying a use case. It also raises new questions about accuracy, data handling and cost per request that older NLP systems didn't have.
Where NLP shows up in IT
Service desks use it to classify and route tickets and to suggest answers from the knowledge base. Search tools use it to find documents by meaning rather than exact keywords. Security teams use it to flag phishing emails and summarize alerts. Legal and compliance teams use it to pull dates, obligations and parties out of contracts. Analytics teams use it to turn free-text survey answers and reviews into themes.
Choosing an approach
For narrow, high-volume tasks with clear categories, such as ticket routing, a smaller dedicated model is often cheaper and more predictable. For varied tasks that need reasoning or writing, a large language model is usually faster to deploy. Either way, test on your own data before you commit, and measure accuracy against a sample a person has checked.
What IT teams should do
- List the text-heavy processes where staff read and sort documents by hand.
- Pick one with a clear measure, such as ticket routing accuracy, for a pilot.
- Test on a sample of your own data checked by a person before going live.