Is the answer predicate part of the real-world subject matter?

Normally no. It is introduced for the query to carry the requested bindings.

Can it hold several values?

Yes. Multiple arguments can preserve the values of multiple query variables.

Does a disjunction of answer literals give separate definite answers?

Not necessarily. It may establish only that at least one alternative works.

AI Glossary / Graphs and symbolic AI / Advanced

Answer literal

Definition

An answer literal is a specially introduced logical literal that carries the values of query variables through a resolution proof. It helps a theorem prover return an answer, such as which object satisfies a condition, instead of merely establishing that an answer exists. It is a bookkeeping device within formal logical question answering.

What it means

Proving that some document is approved and identifying an approved document are related but different tasks. A yes-or-no proof may establish existence without returning the name the user needs.

An answer literal connects the requested variables to the proof process. As unification substitutes specific terms for those variables, the same substitutions appear in the answer literal. A suitable final answer clause can therefore expose the requested values.

Here literal has its technical logic meaning: an atomic formula or its negation. It does not mean a verbatim quotation, a straightforward natural-language answer, or a token generated by a language model. This entry concerns resolution-based symbolic reasoning.

How it works

For a simple query P(x), introduce a fresh answer predicate and attach Answer(x) to the negated query, forming the clause not P(x) or Answer(x). The special predicate is chosen so it does not conflict with an ordinary domain predicate.

Resolution combines this clause with knowledge-base clauses. When a matching positive fact P(a) is available, unification substitutes a for x and resolution removes the complementary P literals. The remaining clause is Answer(a).

The answer predicate is treated as an output marker rather than an ordinary proposition to resolve away. For a query with several requested variables, its arguments can carry a tuple of terms in a defined order.

A proof can also leave a disjunction of answer literals. That may establish that at least one alternative answers the question without identifying a particular one. A system must preserve this distinction rather than present every disjunct as a separately established answer.

Three practical examples

1. A symbolic catalog system answers which item satisfies a formally represented property, retaining the item identifier through the proof.

2. A rule-based assistant answers which person is assigned to which project by carrying two query variables in a single answer predicate.

3. A theorem-proving lesson contrasts proving that a solution exists with deriving a concrete answer term, showing why some existential conclusions are nonconstructive.

Think of it this way

Think of attaching a labeled return slip to a reasoning task. As the task discovers the requested identifier, it fills that identifier into the slip, which survives to the result.

The analogy has limits. The slip does not create evidence or make a guess correct. It records substitutions supported by the formal derivation. If the remaining information says one of two slips is correct, the system cannot truthfully select either as certain without more evidence.

Worked example

Consider an original knowledge base with the fact Signed(iris) and the rule that every signed document is Approved. In clause form, the rule is not Signed(z) or Approved(z).

The question asks for an x such that Approved(x). Add not Approved(x) or Answer(x), using a fresh Answer predicate.

First, resolve Signed(iris) with not Signed(z) or Approved(z). The substitution z = iris gives Approved(iris).

Next, resolve Approved(iris) with not Approved(x) or Answer(x). The substitution x = iris leaves Answer(iris). The returned identifier is iris, and the two resolution steps explain why it satisfies the query under the knowledge base.

Now consider a different knowledge base containing only Approved(iris) or Approved(cedar). It supports the conclusion that at least one is approved, but not either individual approval. Answer extraction can yield Answer(iris) or Answer(cedar). Treating this clause as two definite answers would overstate what was proved.

Illustrative case study

Fictional case: Willow Catalog adds a symbolic query layer to a collection of document records. Its earlier interface returned only yes when some record satisfied a policy. The team adds answer tracking so the interface can display a supported identifier and its derivation.

Reviewers also test a deliberately ambiguous record that states one of two items qualifies. The interface reports the unresolved alternatives rather than selecting a convenient name. This invented case illustrates faithful answer extraction, not a claim about an existing product or measured accuracy.

Common misconceptions and mistakes

An answer literal is not an answer set. Answer sets belong to a different semantic framework used in answer set programming; similar terminology does not make the mechanisms interchangeable.

Do not assume an existential proof always supplies a unique named witness. It may support alternatives or terms whose interpretation requires care.

The marker also does not repair an incorrect knowledge base. The returned terms are justified relative to the supplied axioms and inference rules. Missing facts, inconsistent assertions, or a mistranslated question can undermine the practical answer even when the proof machinery is operating as designed.

Business use

Answer literals help explain how symbolic question-answering systems can connect formal proofs to useful outputs. Returning an identifier, tuple, or structured term is often more valuable than returning a bare truth value.

An implementation should make the output contract explicit: which variables are requested, in what order they appear, and whether the result is a definite instance or an unresolved disjunction. Preserve the proof connection and avoid presenting a partially completed search as exhaustive. In general first-order reasoning, failure to find an answer within a resource limit is not the same as proving that no answer exists.

Check your understanding

A derivation ends with Answer(a) or Answer(b). May the system conclude that a definitely satisfies the query?

Answer: No. That clause alone supports the alternatives collectively, not either answer individually.

Related terms

Resolution
Unification
Clause
query variable
substitution
existential quantification
automated theorem proving
nonconstructive proof

Learn next

Next, trace a two-variable query through a small resolution proof. Then change the facts to a disjunction and compare exactly what the resulting answer clause licenses the interface to say.

Sources

Russell and Norvig, Inference in First-Order Logic, author-hosted chapter: https://aima.cs.berkeley.edu/4th-ed/pdfs/newchap09.pdf

Stanford Introduction to Logic, resolution and fill-in-the-blank questions: https://logical.stanford.edu/intrologic/sections/section_14.html?section=6

Primary educational passages consulted. The supplied term is used in its resolution-proof sense; all document examples are original.

Previous: Ansari-Bradley test Next: Answer set programming