NLP glossary
Key terms from this course, in plain English.
- Attention
- The transformer mechanism that lets each token weigh how relevant every other token is to its meaning.
- Bag of words
- Representing text by counting how often each word appears, ignoring order.
- Bigram / n-gram
- A sequence of two (or n) adjacent words used as a single feature.
- Corpus
- A collection of texts used for analysis or training.
- Cosine similarity
- A measure of how closely two vectors point in the same direction, from −1 to 1; the standard way to compare embeddings.
- Embedding
- A vector of numbers representing a word, sentence or document, learned so that similar meanings are close together.
- Few-shot
- Giving a model a handful of examples in the prompt instead of training it.
- Grounding check
- Verifying that extracted or generated values are supported by the source text.
- Hallucination
- A model producing plausible-sounding content that is not supported by its input or by fact.
- Hybrid search
- Combining keyword search (such as BM25) with embedding search for better retrieval.
- Information extraction
- Pulling structured facts — IDs, dates, amounts, names — out of free text.
- Lemmatisation / stemming
- Reducing words to a base form (running → run) so variants are treated alike.
- NER
- Named entity recognition: finding people, organisations, places, dates and other entities in text.
- NLP
- Natural language processing: techniques that let computers work with human language.
- NMF
- Non-negative matrix factorisation: a method that can discover topics as groups of co-occurring words.
- Normalisation
- Standardising text before processing, e.g. lowercasing or removing punctuation.
- PII
- Personally identifiable information, such as names, emails and phone numbers.
- Redaction
- Replacing personal or sensitive details in text with placeholders.
- Regular expression
- A pattern language for matching precisely defined text, such as order numbers or dates.
- Semantic search
- Search by meaning, using embeddings, rather than exact keyword matches.
- Sentiment analysis
- Classifying text by the opinion or emotion it expresses.
- Stop words
- Very common words often removed before analysis; standard lists may include negations like 'not'.
- TF-IDF
- Term frequency × inverse document frequency: weights words that are frequent in a document but rare across the corpus.
- Token
- A unit of text — a word, subword or character — that a model processes.
- Topic modelling
- Unsupervised discovery of themes as groups of words that tend to appear together.
- Transformer
- The neural network architecture, built on attention, behind modern LLMs.
- Vector database
- A store optimised for finding the nearest vectors to a query embedding.
- Zero-shot
- Asking a model to perform a task from instructions alone, with no examples.