Natural-Language Properties: A Linguistic and Semantic Taxonomy

Purpose#

This document organizes important properties of natural language from a linguistic and semantic point of view. The focus is on how natural language differs from programming languages, formal languages, and APIs.

We have to define the foundational properties of natural language as an interface protocol, ensuring that system architectures natively accommodate indeterminacy, context-dependence, and pragmatic intent, rather than attempting to enforce formal-language determinism on the user.

Also teams lack sometimes an explicit model of the linguistic mechanisms behind user input, so they misdiagnose failures, assign fixes to the wrong layer, and overestimate what a component can infer or execute safely.

Core thesis#

Natural language is not simply an imprecise version of a programming language. It is a different kind of communicative system.

Programming languages and APIs are designed around explicitness, determinism, stable reference, and well-defined execution semantics. Natural language is designed around situated communication: speakers rely on context, shared background knowledge, pragmatic inference, social convention, and conversational repair.

A compact contrast:

1. Foundational linguistic layers#

Before grouping the properties, it helps to separate several layers of linguistic meaning.

1.1 Syntax#

Syntax concerns the structure of an expression: word order, grammatical relations, phrase structure, and well-formedness.

Example:

“Move the meeting to tomorrow.”

This is syntactically well-formed. It has an imperative structure, a verb phrase, an object, and a temporal modifier.

In programming languages, syntax is usually rigid and formally specified. A syntactic error prevents interpretation or execution.

1.2 Semantics#

Semantics concerns conventional meaning: what words, phrases, and sentences contribute independently of a specific conversational situation.

Example:

“tomorrow” conventionally means the day after the relevant reference day.

However, even this semantic content is not fully interpretable without context: we still need to know the relevant date and timezone.

1.3 Pragmatics#

Pragmatics concerns intended meaning in context: what the speaker is doing with the utterance and what should be inferred beyond literal content.

Example:

“Can you send this to Anna?”

Literally, this is a question about ability. Pragmatically, it is usually a request.

1.4 Discourse#

Discourse concerns meaning across multiple sentences, turns, paragraphs, or conversational moves.

Example:

“Find three options. The second one looks best. Make it cheaper.”

The phrase “it” depends on a previous result. The instruction “make it cheaper” modifies an ongoing task rather than starting a new independent sentence-level interpretation.

1.5 Social and interactional meaning#

Natural language also carries signals about politeness, certainty, authority, urgency, hesitation, preference strength, and interpersonal stance.

Example:

“Maybe we should avoid changing the public API.”

This is not merely a statement about an API. It encodes uncertainty, caution, recommendation strength, and possibly a constraint.

2. Group A: Indeterminacy of meaning#

Indeterminacy covers cases where an expression does not map cleanly to one exact interpretation. These phenomena challenge the assumption that every input has a single precise meaning.

Group A includes:

1. Ambiguity
2. Vagueness
3. Underspecification
4. Prototype-based categories
5. Metaphor

2.1 Ambiguity#

Ambiguity occurs when an expression has more than one discrete interpretation.

Example:

“Book a table.”

Possible interpretations:

Ambiguity can occur at several levels.

Lexical ambiguity#

A single word has multiple meanings.

Examples:

Syntactic ambiguity#

A sentence structure allows multiple parses.

Example:

“I saw the man with the telescope.”

Possible interpretations:

1. I used a telescope to see the man.
2. I saw a man who had a telescope.

Scope ambiguity#

A modifier or operator may apply to different parts of the sentence.

Example:

“Email all managers and engineers in Berlin.”

Possible interpretations:

1. Email all managers, plus engineers who are in Berlin.
2. Email managers and engineers who are both in Berlin.

Referential ambiguity#

A referring expression could point to more than one entity.

Example:

“Send it to Alex.”

Ambiguities:

• Which item is “it”?
• Which Alex?
• Send by which channel?

2.2 Vagueness#

Vagueness occurs when a term has fuzzy boundaries rather than a fixed exact threshold.

Examples:

Vagueness differs from ambiguity.

Vague terms are often gradable. Something can be somewhat cheap, very cheap, relatively cheap, cheap for Tokyo, or cheap compared with competitors.

2.3 Underspecification#

Underspecification occurs when an utterance leaves out information that may be needed for a fully explicit interpretation.

Example:

“Schedule a meeting with Alex.”

Missing information may include:

Underspecification is normal in natural language because speakers often rely on context, defaults, shared routines, and follow-up interaction.

In a programming-language or API context, omitted required parameters usually cause a validation error. In natural language, omission may be intentional, efficient, or socially expected.

2.4 Prototype-based categories#

Many natural-language categories are not defined by strict necessary-and-sufficient conditions. Instead, they are organized around prototypes: central, typical examples of a category.

Example:

“Find startup-like companies.”

There may be no strict definition of “startup-like,” but the phrase suggests a cluster of features:

• relatively young company
• growth-oriented
• technology-oriented
• venture-backed or venture-style
• experimental product culture
• relatively informal operating style
• high uncertainty
• high upside

Another example:

“Make the design feel more enterprise.”

This may imply:

• more formal visual tone
• stronger information density
• clearer compliance/security signals
• conservative color choices
• administrative workflows
• role-based access and permissions

Prototype-based meaning differs from formal classification. A database schema might require a value like:

company_type = "startup"

Natural language often works with resemblance:

“This feels more like a startup than an enterprise vendor.”

2.5 Metaphor#

Metaphor uses one conceptual domain to structure another.

Examples:

Metaphor is not ornamental only. It is often a normal way of reasoning and communicating. Technical domains also use metaphor constantly: threads, locks, queues, pipelines, memory, garbage collection, parent/child processes, deadlocks, branches, forks, roots, leaves.

2.6 Summary of Group A#

These phenomena all show that natural-language meaning is often non-exact, graded, partial, or multiply interpretable.

3. Group B: Context dependence#

Context dependence covers cases where an expression cannot be interpreted from the sentence alone. Its meaning depends on the speaker, addressee, time, place, previous discourse, shared knowledge, or current situation.

Group B includes:

1. Implicit context
2. Deixis
3. Coreference
4. Ellipsis
5. Presupposition

3.1 Implicit context#

Implicit context is unspoken background information that speakers assume is available.

Example:

“Do it like last time.”

This utterance requires several contextual facts:

Implicit context may come from:

• previous conversation
• shared work history
• visible interface state
• current document or object
• user preferences
• organizational conventions
• physical environment
• cultural assumptions
• task history

Implicit context is not a small edge case. It is one of the central features of natural language.

3.2 Deixis#

Deixis refers to expressions whose interpretation depends on the situation of utterance.

Common deictic categories:

Examples:

“I need this by tomorrow.”

Interpretation requires:

Programming languages usually avoid deictic expressions unless they are explicitly bound to runtime variables such as currentUser, now, or selectedObject.

3.3 Coreference#

Coreference occurs when multiple expressions refer to the same entity.

Example:

“Sarah sent the contract yesterday. She said it was ready. Forward it to legal.”

Coreference links:

Coreference can be easy or difficult depending on how many candidate referents exist.

Example with ambiguity:

“Anna sent Maria the report after she revised it.”

Possible questions:

• Who revised the report: Anna or Maria?
• What does “it” refer to: the report or something else?

Coreference is one reason natural-language interpretation is discourse-sensitive rather than sentence-isolated.

3.4 Ellipsis#

Ellipsis occurs when part of an expression is omitted because it can be recovered from context.

Example:

“Find flights to Berlin. Cheapest, not fastest.”

The second sentence omits much of the full structure:

“Find the cheapest flights to Berlin, not the fastest flights to Berlin.”

Other examples:

Ellipsis is efficient because speakers do not need to repeat shared or recently mentioned material.

3.5 Presupposition#

A presupposition is background information that an utterance assumes to be true.

Example:

“Cancel my next meeting with Anna.”

Presuppositions:

1. The speaker has meetings.
2. There is at least one upcoming meeting with Anna.
3. One of them is the next such meeting.
4. The speaker has authority or ability to cancel it.

Another example:

“Stop sending weekly reports to the client.”

Presuppositions:

• Weekly reports are currently being sent.
• There is a relevant client.
• The sending can be stopped.

Presuppositions are different from direct assertions. If someone says:

“The current CEO of the company is leaving.”

The sentence asserts that the CEO is leaving, while presupposing there is a current CEO and a relevant company.

3.6 Relationship inside Group B#

Implicit context is the broadest category. Deixis, coreference, ellipsis, and presupposition are specific ways natural language depends on context.

A useful structure:

Context dependence
├── Implicit context: unspoken background assumptions
├── Deixis: meaning anchored to speaker, time, place, or discourse
├── Coreference: different expressions point to the same entity
├── Ellipsis: omitted material recovered from context
└── Presupposition: background facts treated as already true

3.7 Summary of Group B#

Group B shows that natural language is not self-contained. Meaning is often distributed across the utterance, the situation, and prior discourse.

4. Group C: Pragmatic meaning and speaker intent#

Pragmatics concerns what the speaker intends to accomplish with an utterance in context. It goes beyond conventional sentence meaning.

Group C includes:

1. Pragmatics in the broad sense
2. Speech acts
3. Implicature
4. Relevance filtering
5. Politeness and indirectness
6. Modality

4.1 Pragmatics in the broad sense#

Pragmatics studies meaning as use: how people infer intended meaning from context, shared assumptions, communicative goals, and social norms.

Example:

“Can you open the window?”

Literal semantic content:

Are you able to open the window?

Pragmatic interpretation in most contexts:

Please open the window.

The same surface form can perform different pragmatic functions depending on context.

Example:

“You finished the report?”

Possible functions:

4.2 Speech acts#

A speech act is an action performed through language.

Common speech acts:

In formal systems, operation type is usually explicit. In natural language, speech act type is inferred from form, context, tone, relationship, and task state.

4.3 Implicature#

Implicature is meaning that is suggested or implied rather than directly stated.

Example:

“It’s getting late.”

Possible implicatures:

Another example:

“Some of the tests passed.”

This may imply that not all tests passed, even though the literal statement is compatible with all tests passing. The implication arises from conversational expectations: if all tests passed, the speaker would likely have said so.

4.4 Relevance filtering#

Relevance filtering is the process by which listeners select the interpretation that is most relevant to the current context.

Example:

“Is John available tomorrow?”

Possible meanings of “available” include:

• available for a meeting
• available for work
• available by phone
• available emotionally
• available as a contractor
• available in a calendar system

In a work scheduling context, the relevant interpretation is usually:

Is John available for a meeting tomorrow?

Natural language permits many theoretical interpretations, but humans normally do not enumerate all of them. They select contextually relevant meanings.

4.5 Politeness and indirectness#

Politeness and indirectness allow speakers to soften commands, avoid face-threatening acts, signal uncertainty, or preserve social relations.

Examples:

Indirectness matters because literal interpretation can be misleading.

Example:

“Could you send me the file?”

This is usually not a question about ability. It is a request.

4.6 Modality#

Modality expresses possibility, necessity, permission, obligation, likelihood, and certainty.

Examples:

Modality affects how strongly an utterance constrains interpretation.

Compare:

4.7 Relationship between pragmatics and implicit context#

Pragmatics and implicit context are closely related, but they are not the same.

A useful distinction:

Implicit context supplies background information.
Pragmatics uses that background to infer speaker intent.

Example:

“It’s cold in here.”

Semantic content:

The temperature in this place is low.

Possible pragmatic interpretations:

The utterance itself does not explicitly command any action. Context makes a pragmatic interpretation more likely.

4.8 Summary of Group C#

Group C shows that natural language is not merely about what is said. It is also about what is meant, implied, requested, suggested, permitted, forbidden, or socially negotiated.

5. Group D: Discourse and conversational structure#

Discourse phenomena concern meaning across larger stretches of language: multiple sentences, turns, paragraphs, revisions, and interaction sequences.

Group D includes:

1. Discourse structure
2. Repair and correction
3. Mixed initiative
4. Redundancy and emphasis
5. Incremental refinement

5.1 Discourse structure#

Discourse structure concerns how parts of a text or conversation relate to each other.

Examples of discourse relations:

Natural language often uses discourse markers:

• however
• therefore
• because
• although
• first
• then
• instead
• actually
• for example
• in short
• except
• unless

These markers guide interpretation across clauses and turns.

5.2 Repair and correction#

Repair occurs when a speaker fixes, revises, or clarifies a previous utterance.

Examples:

Repair is central to natural conversation. Speakers often do not formulate the final intended meaning in one pass. Meaning is negotiated and corrected over time.

Programming languages generally do not have conversational repair as part of the language itself. One replaces or edits code. In natural language, repair is part of the interaction.

5.3 Mixed initiative#

Mixed initiative means that either participant can guide the direction of the conversation.

Example:

User: “Help me compare these options.” Other participant: “Do you care more about cost, risk, or speed?” User: “Risk first.”

In many formal interfaces, the user must provide a complete command. In natural conversation, the participants can collaboratively determine the next relevant move.

Mixed initiative appears in:

• clarification questions
• suggestions
• objections
• reframing
• prioritization
• topic shifts
• negotiation of criteria

5.4 Redundancy and emphasis#

Natural language often repeats or restates information. This is not necessarily noise. It may signal importance, urgency, contrast, or emotional force.

Example:

“I need this today, before end of day, ideally by 5.”

The deadline is expressed redundantly. The repetition reinforces urgency and priority.

Other examples:

Formal languages often avoid redundancy. Natural language uses redundancy for robustness, emphasis, and social signaling.

5.5 Incremental refinement#

Incremental refinement occurs when a speaker gradually narrows, revises, or extends an earlier request or description.

Example:

“Find laptops under $1500.” “Only 14-inch ones.” “With 32GB RAM.” “Actually, prioritize battery life.”

Each turn modifies the previous search space. The later utterances are elliptical and context-dependent.

Incremental refinement is common because users often discover their criteria while interacting. Natural language supports this naturally.

5.6 Summary of Group D#

Group D shows that natural language is dynamic. Meaning develops across turns rather than appearing only as isolated complete commands.

6. Group E: Social and communicative framing#

Social and communicative framing covers signals about the speaker’s stance, certainty, authority, emotional positioning, relationship to the listener, and strength of commitment.

This group overlaps with pragmatics, but it is useful to separate because these signals affect how an utterance should be understood socially, not merely propositionally.

Group E includes:

1. Politeness and face management
2. Epistemic stance
3. Affective stance
4. Authority and obligation
5. Urgency and priority
6. Commitment strength

6.1 Politeness and face management#

Natural language often protects the social position of speaker and listener. Speakers may soften criticism, hedge requests, or avoid direct confrontation.

Examples:

These forms do not merely decorate the message. They modify how the message should be interpreted socially.

6.2 Epistemic stance#

Epistemic stance concerns how certain the speaker is and what kind of knowledge claim they are making.

Examples:

Compare:

“This is broken.”

with:

“This seems broken.”

The second version marks weaker certainty and leaves more room for revision.

6.3 Affective stance#

Affective stance concerns the speaker’s emotional or evaluative attitude.

Examples:

Affective stance can be explicit or implicit. It often affects how direct, careful, or explanatory the response should be in conversation.

6.4 Authority and obligation#

Natural language encodes authority relations and obligation strength.

Examples:

Authority may come from social role, institutional context, prior agreement, or task ownership.

6.5 Urgency and priority#

Natural language signals urgency in many ways, not only through explicit deadlines.

Examples:

Urgency is partly semantic and partly pragmatic. “ASAP” does not define a precise time by itself; it depends on context and expectations.

6.6 Commitment strength#

Commitment strength concerns how strongly the speaker commits to a proposition, preference, or instruction.

Examples:

Commitment strength is important because two utterances may have similar propositional content but different force.

6.7 Summary of Group E#

Group E shows that natural-language meaning is not only informational. It is also interpersonal and stance-bearing.

7. Relationship between the groups#

The groups are analytically separate, but real utterances often involve several at once.

Example:

“Could you maybe send that to Sarah before the meeting?”

Relevant phenomena:

Another example:

“Actually, not that one – the cheaper option from yesterday.”

Relevant phenomena:

The groups are therefore not mutually exclusive boxes. They are lenses for analyzing how meaning is produced.

8. Why pragmatics and implicit context should be adjacent#

Pragmatics and implicit context are deeply connected.

Implicit context answers questions such as:

• Who is speaking?
• What is currently being discussed?
• What objects are visible or salient?
• What happened earlier?
• What assumptions are shared?
• What is the current task?

Pragmatics answers questions such as:

• What is the speaker trying to do?
• Is this a request, question, warning, correction, complaint, or suggestion?
• What is implied but not stated?
• Which interpretation is relevant?
• How strong is the speaker’s commitment?

The relationship can be summarized as:

Context dependence supplies the background.
Pragmatics derives intended meaning from that background.
Discourse tracks how that meaning changes over time.
Social framing shapes how the meaning is expressed and received.

Example:

“This is a little too aggressive.”

Context dependence:

• What is “this”?
• Which part is being evaluated?
• What genre is it: email, legal letter, design, negotiation message?

Pragmatics:

• The speaker likely wants revision, not just description.
• “A little” may soften criticism.
• “Too aggressive” implies a desired shift in tone.

Social framing:

• The speaker avoids saying “This is bad.”
• The criticism is mitigated.

Discourse:

• The utterance likely modifies a previous draft or proposal.

9. Comparison with formal languages, APIs, and intermediaries#

This section explicitly separates the internal execution semantics of a programming language (state management and logic) from boundary schemas (APIs) and intermediate domain-specific models.

The contrast with programming languages and APIs helps clarify the distinctive nature of natural language. By observing how systems process meaning across a spectrum of communicative protocols, we can map exactly where systems enforce strict execution bounds and where natural language relies on open-world inference.

The fundamental baseline is that formal languages favor explicit, context-independent interpretation, while natural language favors situated, context-sensitive, pragmatically inferred meaning.

9.1 General-purpose programming languages (Internal determinism)#

Programming languages govern a system’s internal state machine. They require formally valid expressions whose meaning is governed by a specification, compiler, interpreter, type system, or runtime.

Typical properties include:

• explicit variables
• fixed syntax
• defined operators
• scoped bindings
• strict error behavior
• formal execution model

Layer 0 Contrast: General-purpose programming languages are the antithesis of natural language because they completely reject Group A (Indeterminacy). A compiler will halt immediately if faced with ambiguity or vagueness. Furthermore, they reject Group E (Social and communicative framing); social meaning is entirely irrelevant to the execution semantics, and code executes identically regardless of intent. Context is strictly “closed-world,” defined by exact lexical or dynamic scope boundaries.

9.2 Declarative configuration languages (Bounded context dependence)#

Declarative configuration languages (e.g., Terraform, Kubernetes YAML) define a desired infrastructure end-state rather than imperative operations. The system’s execution loop continuously compares the environment to the configuration and mutates the system to match.

Layer 0 Contrast: These languages introduce a mechanized, rigid form of Group B (Context dependence). A configuration file cannot be evaluated in isolation; its execution depends entirely on an external state file or environment. However, unlike natural language–which relies on unbounded situational and cultural knowledge–this context remains strictly defined and bounded. There is no unspoken background assumption; state is mathematically tracked.

9.3 Domain-specific languages and query systems (Deterministic intent)#

Domain-Specific Languages (DSLs) like SQL or GraphQL are highly specialized interfaces optimized for a single operational domain. They are largely declarative, allowing the user to state the “what” while the system handles the “how.”

Layer 0 Contrast: DSLs mimic Group C (Pragmatic meaning and speaker intent) by abstracting execution. Intent is resolved via deterministic solvers. However, they aggressively reject phenomena like vagueness (fuzzy boundaries). A database query planner cannot process a request for values that are “cheap” or “nearby” without a strictly explicit, deterministic parameter.

9.4 Application programming interfaces (Boundary validation)#

APIs govern the external communication boundary between systems. They do not manage internal state directly; instead, they require inputs that conform to a schema.

Typical properties include:

• named endpoints or methods
• required parameters
• optional parameters
• typed values
• validation rules
• defined response formats
• explicit success/failure states

Layer 0 Contrast: APIs strictly reject Group D (Discourse and conversational structure). They are stateless and transactional. While natural language users often omit required parameters–relying on underspecification because it is efficient or socially expected–doing so in an API results in a validation error. APIs do not support mixed initiative or incremental refinement natively.

9.5 Controlled natural languages (The syntactic illusion)#

Controlled Natural Languages (CNLs), such as Gherkin (Cucumber) or AppleScript, are formal execution interfaces disguised with natural language vocabulary (e.g., Given a user is logged in, When they click 'Buy').

Layer 0 Contrast: CNLs present a syntactic illusion. While they use English words, they map directly to rigid, operational schemas. They structurally prohibit conversational repair, implicature, and coreference. Deviating with a non-mapped synonym causes a fatal parsing error, proving that simply using English words does not constitute a natural language interface if indeterminacy is disallowed.

9.6 Natural language (Situated pragmatic interface)#

Natural language is not simply an imprecise version of a programming language; it is a different kind of communicative system. It permits highly underspecified expressions such as:

“Move it to tomorrow.”

This compact utterance fails deterministic parsing and lacks the required parameters of an API schema. To execute it, a system must use implicit context to resolve “it,” deictic framing to resolve “tomorrow,” and pragmatics to determine if “move” means to reschedule, relocate, or reorder. Natural language accommodates missing information as normal, recovering it via pragmatic inference and shared background knowledge.

9.7 [WIP] Synthesis: The interaction paradigm matrix#

Because formal systems and natural language optimize for fundamentally different goals, their primary interaction loops handle missing information and syntactic deviation through completely divergent mechanisms.

This matrix maps communicative systems across a spectrum ranging from maximum internal determinism (PLs) to maximum situated pragmatics (Natural Language). By isolating the “Primary Mechanism” alongside the parameters derived from the Layer 0 taxonomy, we can pinpoint exactly how a system handles intent, context, and failure.

Architectural Takeaway#

A true natural language interface protocol cannot simply be an API wrapped in a speech-to-text parser. It requires an architecture natively capable of Open Inference, Open-World Scope, and Mixed-Initiative Interaction, utilizing Conversational Repair rather than halting execution when faced with indeterminacy.

10. Complete taxonomy table#

11. Compact final model#

A concise model of the groups:

Natural-language meaning
├── A. Indeterminacy of meaning
│   ├── ambiguity
│   ├── vagueness
│   ├── underspecification
│   ├── prototype-based categories
│   └── metaphor
│
├── B. Context dependence
│   ├── implicit context
│   ├── deixis
│   ├── coreference
│   ├── ellipsis
│   └── presupposition
│
├── C. Pragmatic meaning and speaker intent
│   ├── speech acts
│   ├── implicature
│   ├── relevance
│   ├── politeness and indirectness
│   └── modality
│
├── D. Discourse and conversational structure
│   ├── discourse relations
│   ├── repair and correction
│   ├── mixed initiative
│   ├── redundancy and emphasis
│   └── incremental refinement
│
└── E. Social and communicative framing
    ├── epistemic stance
    ├── affective stance
    ├── authority and obligation
    ├── urgency and priority
    └── commitment strength

Constraints#

Irreducibility: Layer 0 properties cannot be engineered away. A system cannot force users to speak in unambiguous, context-free, fully specified API payloads. Therefore, engineering effort must focus on accommodating these properties via Layer 3 system controls (e.g., semantic validators, fallback logic) rather than attempting to “fix” the natural language input.

12. Final summary#

Natural language differs from programming languages and APIs not merely because it is less precise, but because it distributes meaning across several layers:

1. lexical and sentence meaning
2. contextual reference
3. pragmatic intention
4. discourse history
5. social and epistemic stance

The most important rearrangement is to separate implicit context from pragmatics, while keeping them adjacent.

Implicit context provides the background needed to interpret expressions like “this,” “that,” “tomorrow,” “same as before,” and “the other one.” Pragmatics uses that background to infer what the speaker is trying to do: request, suggest, warn, correct, object, confirm, or imply.

A final compact formulation:

Formal languages favor explicit, context-independent interpretation.
Natural language favors situated, context-sensitive, pragmatically inferred meaning.

That difference is the basis for the A-E taxonomy above.