This article misrepresents the theory of distributional semantics. The article implies that words that most often appear "in the vicinity of" each other, or words that are "colocated", are semantically similar.

For example, colocation as a predictor of similarity would imply the two words "bank" and "statement" are semantically related.

This is not how word embeddings are trained.

Distributional semantics states that words which appear in the same contexts have similar meanings.

For example, consider these two sentences:

The desert is hot

The desert is dry

"hot" and "dry" both appear in the same contexts, they both appear after "the desert is", this is what gives them semantic similarity.