I studied semantics twenty years before I had any use for it. The subject asks a question that sounds trivial and is not: what makes two expressions mean the same thing.
What do embeddings get wrong about meaning?
Embedding models answer the question of similarity numerically. They place text in a space where distance stands in for similarity. Almost everything written about them is about the mechanics: which model, which dimensions, which distance measure. Very little is about what similarity is being measured, which is the part that decides whether your system works.
Similar is not one thing
"The fare is non-refundable" and "no refunds apply to this fare" are similar in the way a retrieval system wants: same meaning, different words. Any decent model handles that.
"The fare is non-refundable" and "the fare is refundable" are also similar, in the way that ruins you: nearly identical words, opposite meaning. Negation is the classic failure, and it fails quietly, because the retrieved passage looks right until someone reads it carefully.
Linguistics has a word for the distance between what a sentence says and what it is being used to do. Retrieval systems mostly measure the first and are asked about the second.
Context is not metadata
The second thing semantics teaches is that meaning is not fully contained in the sentence. "That is not available for this partner" means nothing without knowing which partner, which product, which contract. Systems that treat context as a filter applied after retrieval have already lost. The search itself needed to know.
This is why the interesting design work in a retrieval system happens before any vector is computed. What counts as a document. Where a document begins and ends. What travels with it. Chunking is not a preprocessing detail, it is the point at which you decide what your system is capable of understanding.
Why this is worth saying
Not to claim that a linguistics degree is required. It plainly is not.
But the field spent a century on questions that retrieval systems now run into on week three, and the answers are written down. Ambiguity, reference, negation, presupposition, the difference between what is said and what is meant. None of that appears in the tutorials, and all of it appears in production.