Definition
You searched, typed, translated and asked a question today. How many times was a language model involved?
Natural language processing is used in any product that reads or writes human language: search engines, translation, spam filters, typing suggestions, grammar checkers, summarisers, sentiment analysis, voice assistants and chatbots.
In simple words
Language is how people store and share almost everything they know. A computer that can work with language can help with almost anything people do with words: find, sort, shorten, translate, answer and write.
Why it matters
NLP tools shape what information reaches you. A search engine decides which pages you see, a spam filter decides which emails you never see, and a summary decides which details are dropped. It helps to know a model is making those choices.
Key terms
- Search engine
- A system that finds and ranks pages relevant to the words you type.
- Ranking
- Putting results in order, with the ones judged most useful at the top.
- Machine translation
- Automatic translation between languages.
- Summarisation
- Producing a short version of a long text that keeps its main points.
- Named entity recognition
- Finding the names of people, places, organisations, dates and amounts in text.
- Autocomplete
- Suggesting how to finish what you have started typing.
Explained in detail
Search: the application everyone uses
A search engine does three things. It visits billions of web pages and keeps a copy of their text. It builds an index, like the index at the back of a book, listing which pages contain which words. And when you type a query, it ranks the matching pages and shows the best ones first. Language AI matters most in the last step. Early search engines matched keywords exactly, so a search for "cheap flights" missed a page about "low-cost air tickets". Modern ones use language models to match meaning, and to work out what you probably want: someone typing "apple" beside "recipe" wants the fruit, not the company.
Tools that help you write
- Autocomplete and next-word suggestions predict what you are about to type.
- Spelling and grammar checkers compare your sentence with patterns of correct writing and suggest changes.
- Transliteration keyboards let you type Hindi, Tamil or Bengali using English letters.
- Voice typing turns dictation into text.
Tools that help you read
Summarisers come in two kinds. An extractive summariser picks the most important sentences from the original and shows them unchanged; it cannot invent anything, but the result can feel disjointed. An abstractive summariser writes new sentences that express the main ideas, as a person would; it reads more smoothly, and it can introduce a statement that the original never made. Translation tools let you read a page in a language you do not know. Question-answering tools find the passage that answers your question and state the answer directly.
Tools that work behind the scenes
| Task | What it does | Where it is used |
|---|---|---|
| Text classification | Sorts text into categories | Spam filters; routing complaints to the right department |
| Named entity recognition | Picks out names, places, dates and amounts | Filling forms from documents; finding people mentioned in news |
| Sentiment analysis | Judges whether text is positive or negative | Reading thousands of customer reviews |
| Content moderation | Flags abusive or dangerous posts for review | Social media and comment sections |
| Information extraction | Turns free text into rows in a table | Reading invoices, medical notes and legal documents |
Reading the results wisely
Each of these tools makes choices on your behalf. A search ranking is influenced by many signals, including how popular a page is and sometimes who paid for a place, so the top result is not automatically the most accurate. A spam filter occasionally hides a real message. A summary drops details that may be the ones you needed. A translation can change a meaning. Used with that in mind, language tools save enormous time. Used without it, they quietly decide what you know.
How it works
| Application | Task | What goes in | What comes out |
|---|---|---|---|
| Search | Find the most useful pages | A few words | A ranked list of pages |
| Translation | Carry meaning across languages | A sentence in one language | The sentence in another |
| Spam filter | Sort messages | An email | Spam or not spam |
| Typing suggestions | Predict the next word | The words typed so far | Three likely next words |
| Summariser | Shorten a long text | A long article | A few key sentences |
| Sentiment analysis | Find the feeling in text | A review | Positive, negative or neutral |
| Question answering | Reply to a question | A question | An answer in sentences |
Two families of language task
- Understanding tasks read text and give a judgement: sorting, searching, finding the feeling, picking out names and dates.
- Generation tasks write new text: translating, summarising, answering, drafting.
Try it yourself
One email meets four language tools
A long email in English arrives for a shopkeeper who prefers Hindi.
Match each task to an application
Which product would you use for each need?
- Read a notice written in another language
- Know whether customers liked a new product
- Get the main points of a twenty-page report
- Keep fake prize messages out of the inbox
- Find pages about growing tomatoes
Common mistakes
- Thinking NLP means only chatbots. Search, spam filtering and typing suggestions are NLP too.
- Treating a summary as the full text.
- Thinking a search result at the top must be the most accurate. It is ranked by many signals, not by truth alone.
Check yourself
Summary
- NLP powers search, translation, spam filters, typing suggestions, summaries and chatbots.
- Some tasks understand text; others generate it; many products do both.
- These tools choose what you see, so check the original when it matters.
Finished reading?
Sources: Elements of AI (University of Helsinki) · AI4K12: Five Big Ideas in AI