AI Vocabulary

Semantic layer: what does it do that your DMS does not?

A legal semantic layer is an intermediate level between a firm's documents and the systems that process them. It holds what documents themselves do not carry: the decisions taken, their reasons, the links between the pieces, and the terminology particular to each matter.

Where the term comes from

Not from law. In computing, a semantic layer has long meant a level that translates technical data into business notions: where a database holds tables and columns, the semantic layer says revenue, client, financial year.

Applied to legal work, the principle is the same and the object changes. What is translated is not a database but documents; and the notions are not accounting ones but legal: a position, a trade-off, a licensed scope, a clause that alters another.

What it covers in practice

Take a framework agreement, the amendment signed six months later, the internal note interpreting it, and the letter sent to the client a year on. Those four documents exist, they are archived, they can be found.

What exists nowhere is the link between them: that the amendment alters a definition in the agreement, that the internal note uses that definition in a wider sense, and that the letter relies on the note rather than on the agreement. That link is what a semantic layer preserves.

The difference shows in one question. Ask a document management system what the assignment clause says: it produces the agreement in three seconds. Ask whether that clause has since been amended, and by what: it cannot answer, not through any deficiency but because the link belongs to no document.

What it is not

Not document management. Document management files and retrieves; it does not know that an amendment alters a definition used 17 times elsewhere.

Not an AI model. A model reasons on what it is given; the layer decides what it is given and keeps what comes out. The two are distinct and work together.

Not one more tool in the stack. A layer sits beneath the existing tools rather than beside them: it does not add another environment, it connects those already there.

And not a knowledge base. A knowledge base holds general knowledge — precedents, notes, model clauses. A semantic layer holds the particular: this matter, this client, this decision, on this date.

Document management knows where the document is. A semantic layer knows what the document changed.

What it does not solve

It holds only what it is given. A trade-off taken orally and recorded nowhere will not appear because a layer exists to receive it: the layer can make recording easy, it cannot invent it.

It replaces neither an exhaustive case-law database, nor a citator, nor subscription commentary. Those are distinct functions answering needs the layer does not cover.

And it dispenses with no verification. Traceable work is still work to be reviewed; traceability shows where to look, it does not substitute for the judgement of whoever signs.

Why it matters to a lawyer

Because the question one asks on picking up a matter is almost never documentary. Not where is the contract, which search has handled for years, but why did we accept that clause, and what has changed since.

That question today has an answer only in people's memories. Displacing it is precisely what a semantic layer is for.