Is a vocabulary the same as an ontology?

Usage varies. A vocabulary names representational elements, while an ontology commonly also organizes their meanings and relationships.

Can a finite vocabulary describe many facts?

Yes. Symbols can be reused in many expressions, and function symbols can support further combinations.

Does adding time always improve a model?

No. It adds complexity and is useful when temporal distinctions matter to the task.

AI Glossary / Graphs and symbolic AI / Intermediate

Vocabulary

Definition

A vocabulary is the collection of symbols a representation uses to express its subject matter. In a first-order knowledge base, it commonly includes constants naming objects, predicates describing properties or relations, and function symbols denoting object-valued mappings. Choosing the vocabulary determines which distinctions the knowledge base can express directly.

What it means

A vocabulary supplies the building blocks; facts and rules state what is true using those blocks. Declaring a predicate called Approved does not establish that anything is approved, define who may approve it, or explain how approval changes over time.
The term has other meanings in AI. In language processing, a vocabulary often means the set of tokens recognized by a model or indexing system. Here the emphasis is symbolic knowledge representation, matching the knowledge-engineering context. Both meanings involve a chosen set of representational units, but a logical predicate vocabulary and a tokenizer's vocabulary serve different purposes.

How it works

Begin with the questions the system must answer. Identify the relevant objects, relations, and distinctions, then assign symbols with clear meanings and argument conventions. A predicate's arity specifies how many arguments it takes; a two-argument relation must not silently alternate between different argument orders.
Next, supply facts, axioms, and constraints that give the symbols their intended use. Review whether a function is appropriate for a uniquely determined value or whether a relation better represents multiple possibilities. Add time, provenance, or other arguments when the task requires them. Vocabulary design is therefore a modeling decision, not merely a naming exercise.

Three practical examples

A knowledge-base project can define symbols for documents, reviewers, and approval relationships.
A planning model can distinguish locations, objects, and the relations that determine available actions.
A data-integration exercise can align two systems' different names for the same relation while preserving argument order and meaning.

Think of it this way

Think of designing the headings and labels on a structured form before filling it in. The chosen fields determine what can be recorded clearly. The analogy is limited because a logical vocabulary supports nested expressions and inference, while a form's labels alone do not provide formal semantics or reasoning rules.

Worked example

Consider an original document-review knowledge base. Its initial vocabulary has constants docA, docB, and reviewerR. It includes predicates Document(x), Reviewer(x), and ApprovedBy(d,r), where the first argument is a document and the second is a reviewer.
Add facts Document(docA), Document(docB), Reviewer(reviewerR), and ApprovedBy(docA,reviewerR). A query asking which named document is approved by reviewerR can now obtain docA from the stated approval fact.
However, the absence of ApprovedBy(docB,reviewerR) does not, in ordinary first-order logic, by itself prove its negation. The system's treatment of missing information requires a separate assumption or rule. That behavior does not follow from the vocabulary's names.
Suppose another dataset uses ApprovedBy(r,d), reversing the argument order. Copying its expressions directly would create the wrong relationship even though the predicate spelling matches. Integration requires an explicit mapping from its first argument to this knowledge base's second argument and vice versa.
Now the task changes: the system must distinguish current approvals from approvals that existed yesterday. The original two-argument predicate cannot express that distinction directly. One option is to introduce ApprovedByAt(d,r,t), a three-argument relation with time values. The old and new predicates are not interchangeable simply because their names are similar.
Finally, consider representing a document's reviewers with a function ReviewerOf(d). A standard function supplies one value for each input in its domain. If a document can have several reviewers, a relation such as Reviews(r,d) may represent the intended multiplicity more naturally. The function could still be used for a specifically defined lead reviewer, but that would be a different meaning that needs documentation.
The example shows three independent vocabulary choices: which concepts are named, how arguments are organized, and which distinctions are explicitly representable. Correct inference depends on those choices plus the facts and axioms; attractive symbol names alone do not provide the needed knowledge.

Illustrative case study

In a fictional publishing-platform project, editors and developers review a proposed approval vocabulary together. Editors explain that an approval applies to a particular document revision, not to every future version sharing the same title.
The team adds an explicit revision concept before importing records. It then checks sample queries with known answers and documents argument meanings. This fictional case illustrates how domain review can prevent ambiguous knowledge representation, without claiming a deployed integration result.

Common misconceptions and mistakes

A vocabulary is not the same as all knowledge expressed with it. Symbols need interpretations, facts, and rules. Two systems can use identical names with different meanings, or different names with equivalent meanings.
Do not assume that a predicate's English name supplies constraints such as uniqueness, symmetry, or transitivity. Do not treat all functions as interchangeable with arbitrary relations: their mathematical behavior differs. Finally, adding a new symbol does not automatically connect it to existing symbols; explicit linking axioms or transformations may be required.

Business use

Vocabulary design affects whether an AI knowledge system can answer the questions the organization actually asks. It also influences data exchange, maintenance, and the clarity of explanations.
Start from representative queries and domain examples rather than compiling a large list of names without a purpose. Document each symbol's meaning, arity, argument order, and relevant constraints. Review changes when the business process evolves. A carefully chosen small vocabulary can be more useful than a broad one whose terms overlap or leave crucial distinctions implicit.

Check your understanding

Why does adding ApprovedByAt require more than renaming ApprovedBy?

Answer: The new relation has an additional time argument and expresses a different kind of fact. Existing data needs a defined temporal interpretation.

Related terms

Knowledge representation
Predicate (logic)
Arity
Ontology
Knowledge engineering
Predicate
Constant symbol
Function symbol
Signature
Token vocabulary

Learn next

Next, extend the review example to distinguish approval of a document revision from approval by a reviewer at a time. Propose the arguments needed, then write one fact and one query. Explain which earlier ambiguities your representation resolves and which assumptions still need rules.

Sources

Primary sources: Russell and Norvig, AIMA, Knowledge Engineering in First-Order Logic: https://aima.cs.berkeley.edu/4th-ed/pdfs/newchap08.pdf ; Stanford Introduction to Logic, vocabulary and arity explanation, indexed excerpt: https://logic.stanford.edu/intrologic/lessons/lesson_07_02.html . The document-review vocabulary is original.

Previous: Vivification Next: Vocabulary-based transformation