What is natural language processing? Working with human language
Natural language processing, or NLP, is the field of building computational methods for human language. It includes classifying text, extracting information, searching documents, translating and generating language. A language model is one tool within that field, rather than a synonym for all NLP.
SchoolWhool's current learning path covers Python, Data Tools, Maths, Statistics and classical Machine Learning. Open lessons' parts and key terms are free to read.
Common NLP tasks and what they return
How text becomes model input
Tokenisation divides text into units a model can process. These might be words, parts of words or other units, depending on the system. An embedding represents a token or a larger piece of text as a numerical vector. Neither step removes the need to define what the task should accomplish.
Consider a hypothetical collection of repair requests. The phrases 'screen will not turn on' and 'display stays black' may concern the same problem despite different wording. A useful exercise is to compare a word-based baseline with a representation that can capture more context, using the same evaluation cases.
NLP versus LLMs: do you need a large model?
Large language models are part of modern NLP, but smaller classifiers, rules and retrieval systems remain possible tools. If you only need to find an exact reference code in a predictable format, a carefully tested rule may be a better starting point than generation.
For the repair-request exercise, first define the categories and the messages that should be left for review. Compare a simple text classifier before asking a generative model to produce a category. Record invalid outputs and ambiguous cases as well as correct answers.
Build a useful first NLP project
Define labels and exclusions
Write a short guide to the categories. Decide what happens to messages with several issues, no clear issue or unsupported languages.
Inspect and split the messages
Remove inappropriate personal data from a practice dataset. Keep duplicate messages and related conversation threads in the same split.
Fit preprocessing on training data
Learn vocabulary and other fitted transformations from the training portion. A pipeline helps keep the same transformation applied during evaluation.
Review errors by language and wording
Inspect paraphrases, negation, spelling variation and rare categories. Document the kinds of message you did not have enough examples to evaluate.
What does a good NLP evaluation look like?
A routing classifier needs a different evaluation from a summary generator. For classification, inspect which categories are confused. For extraction, define how to score missing or partly correct fields. For summaries, check individual claims against the source instead of judging only readability.
Keep a small set of reviewed examples with explanations of the expected answer. When you change the representation or model, use those examples to investigate regressions as well as the aggregate metric. An attractive example should illustrate the result, not stand in for the evaluation.
What should you learn before NLP?
Practise Python strings, functions and file handling, then use tables to inspect a text dataset. Learn classification, train/test separation and statistics before treating an output score as evidence. These skills transfer from simple text classifiers to more advanced language systems.
SchoolWhool's Python, Data Tools, Statistics and Machine Learning tracks teach these foundations. NLP and LLMs are not released specialist tracks; this guide helps you identify the preparation you can begin now.
Where this guide takes you
- Use the examples to understand the task and its evaluation. They are suggested learning exercises, not measured results or production-ready systems.
- The linked SchoolWhool courses teach the foundations. Specialist deep learning, computer vision, NLP, generative AI and reinforcement learning tracks are not published.
- Open lessons' parts and key terms are free to read. A free account opens quizzes and saved progress, plus practice and other material in the free sample modules.
Questions
What does NLP stand for in AI?
Natural language processing: computational methods for working with human language, including analysis, retrieval and generation.
Is NLP the same as generative AI?
No. NLP includes tasks that return labels, extracted fields and retrieved documents. Generative AI creates content and can also work with images, audio and other media.
What is the difference between a token and a word?
A token is a processing unit chosen by a tokeniser. A word can be one token, several tokens or part of a larger representation, depending on the system.
What is an embedding used for?
An embedding is a numerical representation used as input to further computation, such as classification or similarity search. Similarity depends on how the representation was learned and used.
Can I start NLP without deep learning?
Yes. Begin with text inspection, word-based representations and a simple classifier. That makes a useful baseline before studying larger neural models.
Sources
Technical references and further reading from the authors, educators and maintainers below. Read dates record when these references were checked; the examples and suggested exercises in this guide are SchoolWhool's own.
- Hugging Face: NLP and large language modelsRead 2026-10-14
- Hugging Face: LLM course introductionRead 2026-10-14
Related reading
What is generative AI?Python for machine learningSupervised vs unsupervised learningData Tools trackBuild the foundations for this subject
Follow SchoolWhool's learning path through Python, Data Tools, Maths, Statistics and Machine Learning. Start with the skills you need, read open lessons without an account, then sign in free for quizzes and saved progress. Free sample modules also open their practice and study material.
