Wortmathematik: King minus Man plus Woman

Because words are now vectors, you can add and subtract them like arrows. This unlocks a playful idea called word math.

Take king, subtract man, add woman, and the nearest vector is queen. This analogy stunned researchers when it first appeared. 👑

The step from man to woman is a consistent direction in space. Adding that same direction to king lands you near queen.

Many relations become repeatable vector offsets. The gap encoding gender or tense behaves like a fixed arrow across word pairs.

The trick generalizes nicely. Paris minus France plus Italy lands near Rome, capturing a capital-city relationship.

You compute the target vector, then search for its nearest neighbor among all word vectors using cosine similarity.

gensim wraps the whole analogy in a single call. You list positive and negative words and read the top result.

Even without subtraction, you can ask which words sit closest to one term. These neighbors reveal what the model thinks is similar.

Analogies often work, but not always. Bias in the training text can produce unfair or wrong results, so stay critical.

Embeddings absorb the stereotypes present in their data. Always remember that bias in text becomes bias in your model.

Simple arithmetic on learned vectors reflects real human relationships. That is the clearest sign embeddings truly capture meaning. ✨

What does king minus man plus woman tend to produce?

Word math adds and subtracts vectors to solve analogies like king to queen. It shows embeddings encode real relationships, bias and all. ✅