Rule Semantics

1. Defining rule semantics

Rule semantics concerns the meaning and effect of rule expressions

Semantics is different from syntax. Syntax concerns how words or symbols are arranged. Semantics concerns what that arrangement means. Two sentences may use different syntax while expressing materially the same rule. Conversely, two sentences may look similar while imposing different obligations because one word, definition, threshold, or condition changes the effect.

The Object Management Group's Semantics of Business Vocabulary and Business Rules specification was developed to support the meaning of business vocabularies and rules independently of any single implementation language.¹ OASIS LegalRuleML likewise provides formal features for representing legal norms, policies, guidelines, and their relationships.² These standards illustrate a central Rules Integrity principle: meaning should be capable of preservation even when representation changes.

2. The semantic structure of a rule

Meaning is distributed across several connected elements

Consider the rule: “The Records Manager must preserve approved contracts for seven years after expiration unless a legal hold requires longer retention.” Its meaning depends upon every component:

• Actor: the Records Manager;
• modality: obligation expressed by “must”;
• action: preserve;
• object: approved contracts;
• duration: seven years;
• trigger: contract expiration;
• exception or override: a legal hold requiring a longer period;
• implied scope: the organization, records system, and contract category governed by the policy.

Removing or changing one element changes the rule. “May preserve” converts an obligation into permission. “Seven years after approval” changes the starting event. “Business contracts” may narrow or broaden the object depending on the governing vocabulary. Semantic analysis makes these changes visible.

3. Vocabulary and concepts

A rule cannot be more consistent than the vocabulary it uses

Rules operate through concepts: employee, customer, material breach, confidential information, approved system, emergency, business day, authorized officer, and hundreds of others. When those concepts are undefined, duplicated, or used inconsistently, rule interpretation becomes unstable.

A controlled vocabulary should identify:

• the preferred term for each governed concept;
• its definition and source;
• broader, narrower, and related concepts;
• synonyms, deprecated terms, and abbreviations;
• jurisdictional or departmental variations;
• effective versions and change history;
• rules that use or depend upon the concept.

Definitions should explain the boundary of a concept, not merely repeat the word. “Critical vendor means a vendor that is critical” is circular. A useful definition identifies criteria such as operational dependency, replacement difficulty, information access, financial exposure, or regulatory significance.

Definitions also create dependencies. Changing the definition of “record” may alter retention, privacy, access, discovery, and destruction rules. Vocabulary governance is therefore part of rule governance, not a separate editorial task.

4. Actors, actions, and objects

The rule must identify who is governed and what conduct is directed

The actor may be a person, role, organization, automated system, decision body, contractor, customer, regulator, or category of persons. The actor should be defined by responsibility or status rather than by a temporary individual's name unless the rule is truly personal.

The action should use a verb that identifies observable conduct or a decision: approve, notify, retain, disclose, calculate, inspect, classify, escalate, refrain, verify, or document. Vague verbs such as “handle,” “address,” or “support” may conceal several distinct obligations.

The object identifies what the action concerns: a record, transaction, person, system, request, risk, decision, or item of information. An incomplete object can silently broaden the rule. “Managers must review changes” does not identify which changes, to what, or for what purpose.

The triple is still incomplete without timing, conditions, scope, and exception handling, but it exposes the central relationship that the rule governs.

5. Modality and normative force

The modal expression determines whether conduct is required, permitted, prohibited, or recommended

Normative force is carried by words such as must, may, must not, and should, or by equivalent structures. The Office of the Federal Register uses “must” for a requirement, “should” for a strong recommendation, “may” for an option, and “can” for capability or possibility.⁴ ISO drafting conventions similarly distinguish requirements, recommendations, permissions, possibilities, and capabilities.³

Modal semantics should not be inferred from tone. A sentence can sound forceful while remaining nonbinding. A recommendation can become functionally mandatory if departure is punished. A permission can become an obligation when another rule requires the actor to exercise it under stated conditions.

Rules Integrity therefore examines both expressed modality and institutional effect.

6. Conditions, triggers, and prerequisites

A rule may be dormant until particular facts make it applicable

Conditions determine when a rule activates. They may concern an event, state, classification, threshold, prior decision, missing action, or combination of facts.

Common forms include:

• trigger: “When a complaint is received...”;
• prerequisite: “After identity has been verified...”;
• classification condition: “If the vendor is critical...”;
• threshold condition: “When the amount exceeds $50,000...”;
• negative condition: “If no response is received...”;
• compound condition: “If the account is privileged and access is remote...”

Logical connectors matter. “A and B” requires both. “A or B” may require either or both unless exclusive choice is stated. “Either A or B, but not both” is different again. Long sentences containing several connectors should often be represented in a decision table or structured condition model.

The rule should also address incomplete or disputed facts. What happens when a classification has not yet been made, the threshold value is unavailable, or two systems report different states? Silence at the condition layer often produces inconsistent default behavior.

7. Scope and quantification

Words such as all, any, each, some, and only define the reach of the rule

Quantifiers identify how broadly a statement applies. “Each employee must complete training” imposes an individual obligation. “The department must ensure that employees complete training” assigns responsibility to the department. “At least one reviewer” differs from “every reviewer.”

Scope may be:

• universal: all cases within the defined category;
• existential: at least one case or actor;
• limited: only specified categories;
• conditional: categories meeting stated criteria;
• geographic or jurisdictional;
• organizational;
• technical;
• temporal.

The placement of qualifiers can change meaning. “Managers must review only external disclosures” limits what must be reviewed. “Only managers must review external disclosures” limits who has the obligation. The same words create different rules because the qualifier attaches to a different element.

8. Temporal semantics

Time must be defined as carefully as conduct

Temporal meaning includes triggers, deadlines, duration, recurrence, sequence, effective periods, suspension, and expiration. Many operational contradictions arise not from different actions but from different understandings of time.

“Within five days” is incomplete without a starting event, day-count convention, treatment of weekends and holidays, time zone, and the consequence of a deadline falling on a nonworking day. Precision should reflect materiality; trivial processes may not need elaborate temporal rules, while legal notices and high-value transactions often do.

9. Thresholds and measurement

A threshold is meaningful only when its measurement basis is controlled

Threshold rules depend upon numbers, categories, scores, percentages, durations, or qualitative levels. The comparison operator is only one part of the meaning.

“Transactions above $100,000 require review” leaves several semantic questions:

• Does exactly $100,000 require review?
• Which currency governs?
• Which exchange rate and date apply?
• Are taxes, fees, or renewals included?
• Are related transactions aggregated?
• Which system is the authoritative source of the amount?
• What happens when the amount changes after approval?

Qualitative thresholds need similar discipline. “High risk,” “material change,” and “substantial delay” may be legitimate categories, but the rule should identify criteria, decision authority, evidence, and review expectations.

10. Negation and prohibitions

The location of a negative can reverse or narrow the rule

Negation is a frequent source of semantic error. Compare:

• “Employees must not disclose restricted data.”
• “Employees are not required to disclose restricted data.”
• “Not all employees must disclose restricted data.”
• “Employees must disclose data that is not restricted.”

These statements contain similar words but express a prohibition, absence of obligation, partial limitation, and positive obligation.

Negative conditions also require care. “Access is denied unless approval is present” establishes denial as the default. “Access is permitted unless a restriction is present” establishes permission as the default. Incomplete data will produce opposite results under those designs.

Direct positive expression is usually easier to test, but prohibitions are often necessary. The engineering record should make default states and the scope of negation explicit.

11. Exceptions, overrides, and priority

An exception changes which rule controls under specified conditions

Exceptions are semantic operators. They narrow a general rule, replace its result, suspend it, or authorize a different decision path. Their meaning depends upon the relationship to the general rule.

Common forms include:

• exclusion: the general rule does not apply to a defined category;
• conditional exception: the rule yields when specified facts exist;
• override: another rule has priority;
• waiver: an authorized actor may suspend application for a case;
• emergency rule: temporary conditions activate a different regime;
• specific-over-general rule: a specialized rule controls its narrower subject.

The exception should state whether it removes the obligation entirely, substitutes another obligation, changes the deadline, or grants discretion. “Except in an emergency” is semantically incomplete unless emergency, authority, alternative conduct, evidence, and duration are governed elsewhere.

12. Ambiguity and vagueness

Ambiguity offers competing meanings; vagueness leaves uncertain boundaries

The concepts are related but different.

Ambiguity exists when language supports more than one distinct interpretation. “The manager approved the report with the exception” may mean the report contained an exception or the manager used an exception in approving it.

Vagueness exists when a concept has uncertain boundaries. “Promptly,” “reasonable,” “significant,” and “high risk” may have clear central cases but disputed edge cases.

Not all vagueness is defective. Standards such as reasonableness can preserve judgment where circumstances vary. The integrity question is whether the uncertainty is intentional, governed by relevant factors, assigned to a legitimate decision-maker, and subject to evidence and review.

Ambiguity is more often avoidable through structure, definitions, punctuation, sentence division, explicit references, and examples.

13. Semantic equivalence

Different wording may express the same rule—and similar wording may not

Semantic equivalence matters when rules are consolidated, translated, paraphrased, automated, or compared across documents. Two statements are materially equivalent when they produce the same normative result across the relevant cases.

“A supervisor must approve every refund above $500” may be equivalent to “No refund exceeding $500 may be issued without supervisor approval,” provided the same actor, amount basis, timing, and scope apply. The second form combines a prohibition with an approval condition; implementation may differ even when the intended outcome is the same.

Equivalence should be tested, not assumed. A paraphrase may alter modality, default, threshold inclusion, time, actor, or exception. Translation across jurisdictions or technical systems may introduce concepts that do not map one-to-one.

14. Implicit meaning and hidden assumptions

Readers routinely supply missing meaning from context

Rules often rely upon assumptions that are obvious to the author but invisible to later readers. Examples include:

• which organizational unit owns the action;
• which time zone applies;
• whether electronic approval counts as a signature;
• which record system is authoritative;
• whether a permission includes authority to delegate;
• whether a deadline applies to receipt or completion;
• whether an obligation survives contract termination.

Not every assumption belongs in the rule text. Excessive detail can make rules unreadable. But material assumptions should be governed through definitions, architecture, procedures, implementation specifications, or referenced conventions.

Hidden assumptions become especially dangerous when rules cross organizational or technical boundaries. Shared context disappears, leaving the words intact but the intended meaning incomplete.

15. Human and machine representations

Meaning must survive movement between prose, models, data, and code

Human-readable language carries nuance, purpose, and legal authority. Structured representations support consistency, comparison, testing, and automation. Neither form should be treated as universally superior.

Controlled natural language can reduce variation by using defined terms, consistent modal verbs, and standard sentence patterns. Decision tables can expose condition combinations. Formal rule languages can represent normative relationships and support automated reasoning. Software can enforce selected decisions at operational speed.

The danger arises when transformation is not traceable. A programmer may choose a default not stated in policy. A decision table may omit an exception described in a footnote. A plain-language summary may weaken a legal obligation. A machine rule may require binary input where the source relies upon professional judgment.

Each transformation should retain a trace to the prior form, identify interpretive choices, and be tested for semantic preservation.

16. Common failure cases

Small semantic defects can produce system-wide inconsistency

17. Semantic review

Twenty questions for semantic integrity

1. 01

   Vocabulary

   Are key concepts defined, governed, and used consistently?

2. 02

   Actor

   Is every responsible, permitted, prohibited, or protected actor identifiable?

3. 03

   Action

   Does the rule use an observable or decision-relevant action?

4. 04

   Object

   Is it clear what the action concerns?

5. 05

   Modality

   Is normative force explicit and consistent with the drafting convention?

6. 06

   Conditions

   Are triggers, prerequisites, classifications, and logical connectors clear?

7. 07

   Scope

   Are population, jurisdiction, organization, system, and activity boundaries controlled?

8. 08

   Quantification

   Are words such as all, any, each, only, and at least used deliberately?

9. 09

   Time

   Are trigger, day-count method, deadline, duration, recurrence, and effective period defined where material?

10. 10

    Thresholds

    Are operator, unit, source data, aggregation, and boundary inclusion explicit?

11. 11

    Negation

    Is the scope of every negative expression unambiguous?

12. 12

    Defaults

    What happens when data is absent, conditions are unknown, or no explicit result applies?

13. 13

    Exceptions

    Is the relationship between the general rule and each exception explicit?

14. 14

    Priority

    Can competing rules be resolved through hierarchy, specificity, timing, or authorized override?

15. 15

    Ambiguity

    Does any wording support materially different interpretations?

16. 16

    Vagueness

    Where boundaries require judgment, are factors, authority, evidence, and review governed?

17. 17

    Assumptions

    Which material meanings are supplied only by local knowledge or custom?

18. 18

    Equivalence

    Have paraphrases, translations, summaries, and technical expressions been tested against the source?

19. 19

    Examples

    Do examples clarify the rule without silently narrowing or expanding it?

20. 20

    Change impact

    Can a changed term, threshold, condition, or exception identify every affected rule and implementation?

18. Worked examples

Making semantic choices explicit

The controlled expression resolves several uncertainties but raises additional engineering questions: currency, aggregation, amendments, emergency purchases, delegated authority, and electronic approval. Semantic discipline does not imply that one sentence should carry the entire system.

Conclusion

Semantic integrity allows a rule to remain the same rule across contexts and representations

A rule carries meaning through vocabulary, actors, actions, modality, conditions, quantification, scope, time, thresholds, negation, exceptions, priority, and context. Weakness in any one element can produce materially different decisions even when everyone believes they are applying the same text.

Rule semantics does not demand that every rule become rigid or formal. It demands that uncertainty be recognized. Some uncertainty should be eliminated through definitions and structure. Some should be assigned to legitimate judgment. Some should be preserved because future conditions cannot be responsibly predetermined.

The central responsibility is to know which kind of uncertainty exists and to preserve the rule's intended meaning as it moves through interpretation, implementation, automation, review, and change.