14.3 Embeddings

An embedding turns a token into a vector — a list of numbers — that captures its meaning. Words used in similar contexts get similar vectors, so 'king' and 'queen', or 'coffee' and 'tea', land close together in this numeric space. This lets models reason about meaning, not just spelling.

Words with related meanings cluster together in embedding space.

Picture an Ethiopian e-commerce search: a shopper types 'ስልክ' (phone) but a product is listed as 'ሞባይል' (mobile). With embeddings, the system knows these two words are close in meaning and still shows the right products, even though the spellings differ.

Scenario

You want a search engine that returns results for 'ሐኪም' (doctor) even when a document only says 'ዶክተር' (doctor). What technique enables this?

Check your understanding

1/4 · 40 XP

Embeddings represent words as numeric vectors where similar meanings are placed close together.