Ambiguity and Context

Language works despite ambiguity.

This is remarkable.

Words have multiple meanings.

Sentences can have multiple structures.

Pronouns can have multiple referents.

Speakers imply things they never say explicitly.

Yet most conversations succeed.

Why?

Because language is interpreted in context.

Lexical Ambiguity

A single word may have several meanings.

“Bank.”

Financial institution.

Riverbank.

The sentence usually resolves the ambiguity.

“She deposited money at the bank.”

“She sat on the river bank.”

Context narrows the possibilities.

Polysemy

Sometimes one word has related meanings rather than unrelated homonyms.

“Head” can mean:

• body part,
• leader,
• top of something.

This is polysemy.

Meaning networks are structured rather than simple one-to-one mappings.

Structural Ambiguity

Consider:

“I saw the man with the telescope.”

Two interpretations:

• I used a telescope.
• The man had a telescope.

The words are identical.

The syntax tree differs.

Structure creates ambiguity.

Scope Ambiguity

Logical operators can interact ambiguously.

“Every student read a book.”

Possibility 1:

each student read possibly different books.

Possibility 2:

there is one particular book everyone read.

Quantifier scope changes meaning.

Pronoun Ambiguity

“Anna told Maria that she had won.”

Who is “she”?

Grammar may allow both referents.

Context and discourse history help decide.

Reference is not always encoded locally.

Ellipsis

Human language often omits recoverable material.

“John ordered coffee, and Maria tea.”

We understand:

Maria ordered tea.

The missing structure is reconstructed from context.

Communication relies on inference.

Indexicals

Words such as:

• I,
• here,
• now,
• this

cannot be interpreted without situational information.

Their meaning includes a rule for finding a referent in context.

Language is partly anchored to the speech situation.

Deictic Center

Interpretation often uses a deictic center:

speaker, place, time.

“Tomorrow” depends on when the sentence is uttered.

“Here” depends on where.

Context supplies coordinates.

Pragmatic Ambiguity

The sentence:

“You left the door open.”

could be:

• observation,
• accusation,
• request,
• warning.

Literal semantics underdetermines communicative force.

Pragmatics completes the message.

Sarcasm

“Great job.”

can mean praise.

Or criticism.

The listener uses:

• tone,
• prior events,
• expectations.

Surface meaning can be reversed by context.

Irony

Irony depends on a mismatch between literal expression and understood situation.

Without shared background, ironic meaning can disappear.

This is why irony is difficult for both second-language learners and machines.

Common Ground

Conversation relies on common ground:

what participants mutually assume they know.

Shared knowledge allows compressed communication.

“Same place?”

can be fully meaningful between friends and meaningless to outsiders.

Discourse Context

Interpretation depends not only on immediate sentence but on previous conversation.

A pronoun refers back.

A topic persists.

New information is integrated into an evolving discourse model.

Language has memory.

World Knowledge

Consider:

“The trophy did not fit into the suitcase because it was too small.”

We infer “it” refers to the suitcase.

Why?

Because we know containment relations.

Understanding requires a model of the world.

Winograd-Style Examples

Tiny changes can reverse likely reference.

“The trophy did not fit into the suitcase because it was too large.”

Now “it” likely refers to the trophy.

Such examples became famous in AI because they require contextual reasoning.

Probability

Interpretation is often probabilistic.

A listener considers several meanings.

Context makes one much more likely.

Understanding can be viewed as inference over alternatives.

Bayesian Perspective

In a Bayesian framing:

posterior interpretation

depends on:

prior expectations × evidence from the message.

Context shapes priors.

The utterance provides evidence.

Meaning emerges from constrained inference.

Ambiguity Is Not a Defect

One might think a perfect language should eliminate ambiguity.

But ambiguity makes language efficient.

One short word can serve several related contexts.

Humans resolve most cases effortlessly.

Flexibility saves representational cost.

Precision vs Efficiency

Formal languages reduce ambiguity by increasing explicit structure.

Natural languages rely on intelligent interpreters.

This creates a tradeoff:

more precision

vs

more flexibility.

Human language chooses flexibility surprisingly often.

Ambiguity and Creativity

Poetry exploits ambiguity deliberately.

A phrase can sustain several meanings at once.

Literary richness often depends on semantic openness.

What engineering treats as error, art can use as resource.

Ambiguity in Law

Legal language tries to minimize ambiguity.

Yet disputes still occur.

Words interact with:

• precedent,
• context,
• intent.

Natural language cannot be made perfectly formal without changing its nature.

Ambiguity in Science

Scientific writing reduces ambiguity through:

• definitions,
• notation,
• units.

But conceptual terms can still shift.

Precision requires active discipline.

Shared vocabulary is part of scientific method.

Context in Machine Translation

A word cannot always be translated correctly in isolation.

“Bank” must be interpreted before choosing a translation.

Modern translation systems use surrounding context to improve semantic selection.

Context is computationally valuable.

Contextual Embeddings

Modern language models represent a word differently depending on nearby words.

The vector associated with “bank” in a finance sentence differs from “bank” in a river sentence.

This is a major advance over fixed word representations.

Does Context Solve Everything?

No.

Context can itself be ambiguous.

People can disagree about:

• intention,
• reference,
• implication.

Interpretation remains fallible.

More context reduces uncertainty but does not always eliminate it.

Misunderstanding

Communication fails when sender and receiver use different:

• assumptions,
• conventions,
• background knowledge.

Misunderstanding is not necessarily noise in the signal.

It can be mismatch in interpretive models.

Strategic Ambiguity

Sometimes speakers intentionally leave meaning vague.

Politicians.

Diplomats.

Advertisers.

Ambiguity can preserve flexibility or avoid commitment.

Meaning can be managed strategically.

Context Changes Over Time

A sentence written today may be interpreted differently in a century.

Words shift meaning.

Cultural assumptions change.

Texts survive longer than contexts.

Interpretation is historically situated.

The Philosophical Lesson

Language is not a code where every string maps uniquely to one meaning.

It is a system in which:

• structure constrains,
• context narrows,
• interpreters infer.

Ambiguity is managed rather than eliminated.

The Next Question

Natural language tolerates ambiguity.

Programming languages cannot tolerate the same degree of interpretive freedom.

Their syntax and semantics must be specified far more explicitly.

That brings us to:

programming language syntax and semantics.