🔤NLP & Language
Turning language into something models can use · 28 terms
NLP Basics
6What it means for a computer to work with human language. Also the text and models it starts from.
The part of AI that reads, writes and makes sense of human language.
The part of language AI that finds what text means. Then a program can act on it.
The part of language AI that writes. It turns data or ideas into text.
A big set of text that a language model learns from.
A model that guesses which word comes next. Repeat it and it writes.
A run of n words in a row, like 'green tea' for n = 2.
Text Processing
4The cleaning steps that tidy raw text before a model sees it.
Very common words like 'the' and 'of' that some tools take out.
Cutting word endings off by rule, so 'jumps' and 'jumped' match.
Each word goes back to its base form, so 'was' turns into 'be'.
Each word gets a tag, like noun or verb.
Text Representation
3Ways to turn words into numbers, from simple counts to points on a map of meaning.
Text turned into word counts. The word order is thrown away.
A word score. It is high when a word is common in one page but rare in the rest.
Each word as a list of numbers, so words with close meanings sit close.
Encoder and Decoder Models
4Model designs built for language work, from models that read a whole sentence at once to models that read one text and write another.
A language model from 2018 that reads text in both directions. It is built to understand, not to write.
A smaller, faster BERT that kept 97% of its skill.
A model trained by hiding words in text and guessing them back. BERT learned this way.
A model with two parts. One reads the input, and the other writes a new output from it.
Language Tasks
11The jobs language tools do for people. They read mood, find names, translate and sum up.
Text Analysis Tasks
6Jobs where a model reads text and pulls out a group, a goal, facts or themes.
Sorting text into groups that were picked first, like spam or not spam.
Finding what a user wants from what they say. Reset a password is one goal.
Reading text to tell if it sounds good, bad or in between.
Finding names of people, places, dates and firms in text.
Pulling facts like names and dates out of text. They go into a neat table.
Finding the main themes in many texts. Words that show up together form a theme.
Language Generation Tasks
5Jobs where a model writes new text, from free writing and answers to translations, summaries and database queries.
Making new text with a model. It adds one word at a time to a prompt.
A machine reads your text and writes it out in a new language.
Making a short version of a long text that keeps the main points.
Giving a direct answer to a question in plain words. Not a list of links.
Turning a question in plain words into SQL. Then a database can run it.