Is lexical ambiguity the same as syntactic ambiguity?

No. Lexical ambiguity concerns possible interpretations of words; syntactic ambiguity concerns possible structures, though they can interact.

Does part-of-speech tagging always resolve it?

No. Different senses can share the same part of speech.

Can context leave several readings open?

Yes. Some short or deliberately ambiguous expressions cannot be resolved from the available text alone.

AI Glossary / Graphs and symbolic AI / Beginner

Lexical ambiguity

Definition

Lexical ambiguity occurs when a word or lexical expression has more than one possible interpretation. For example, crane can refer to a bird or a lifting machine. In language processing, the system must use context to determine which interpretation is intended, or preserve uncertainty when the available information does not resolve the ambiguity.

What it means

The ambiguity lies in the interpretation of a lexical item, rather than solely in how a whole sentence is structured. A sentence can also have structural ambiguity, and the two types can interact. Recognizing the spelling of a word correctly therefore does not guarantee that its meaning has been identified correctly.
Some multiple meanings are closely related, often discussed as polysemy, while others are more distinct, often discussed as homonymy. The boundary and the appropriate number of senses can depend on the linguistic analysis or application. A system needs a suitable interpretation scheme rather than assuming that every dictionary distinction maps neatly to its task.

How it works

A language system can consider surrounding words, grammatical structure, document topic, and wider conversational context. A word-sense disambiguation component may choose among entries in a sense inventory; other systems represent contextual meaning without assigning an explicit dictionary label.
Context provides evidence, not a universal guarantee. A short phrase may support several readings, and domain-specific usage can differ from common usage. When an interpretation affects an important action, the application may need a clarifying question, additional context, or a deliberately cautious output. That behavior should follow the task requirements rather than hiding uncertainty behind fluent wording.

Three practical examples

A document search tool distinguishes reports about lifting cranes from wildlife reports about cranes.
A translation system uses context to choose an appropriate target-language expression for an ambiguous source word.
A customer-service assistant checks whether a request about a “charge” concerns a fee, electrical power, or another meaning relevant to the conversation.

Think of it this way

Think of one label attached to different drawers. The surrounding situation helps determine which drawer is intended. The analogy is limited because word meanings are not always neatly separated boxes: related senses can overlap, and speakers can introduce figurative or novel uses that do not fit a fixed inventory cleanly.

Worked example

Consider the invented sentence “The crane is near the water.” Both the bird and machine readings are plausible. Adding “its operator is checking the lifting cable” strongly supports the machine reading. Adding “it is feeding in the shallows” strongly supports the bird reading.
Now imagine a deliberately simplified classifier with only two candidate senses. In a construction-document collection, suppose it starts with illustrative prior probabilities of 0.90 for machine and 0.10 for bird. For the contextual cue “feeding in the shallows,” suppose the model assigns likelihoods of 0.01 under machine and 0.80 under bird.
The unnormalized weights are 0.90 times 0.01 = 0.009 for machine and 0.10 times 0.80 = 0.08 for bird. Their sum is 0.089, giving a normalized bird probability of 0.08/0.089, approximately 0.899. Strong contextual evidence reverses the initial preference.
These numbers are invented to demonstrate a possible inference calculation, not measured linguistic probabilities. A real system may have more senses, dependencies between cues, or a different model entirely. The example also does not prove the sentence describes an actual bird: interpreting a sentence and verifying its factual content are separate tasks.

Illustrative case study

In a fictional archive, Brook Index retrieves machinery manuals for every query containing crane. Wildlife staff find the results unhelpful, and the team discovers that a keyword rule always selects the collection's most common meaning.
The revised search uses surrounding query terms and offers topic clarification when the query is too short. The invented case illustrates a relevance problem caused by unresolved lexical ambiguity. It does not claim that one contextual cue always disambiguates a word or that the revision has a measured accuracy improvement.

Common misconceptions and mistakes

Lexical ambiguity is not necessarily a spelling error or poor writing. Many ordinary words naturally support several meanings. More context often helps, but not every passage supplies enough information to select one reading confidently.
Do not confuse identifying a word's grammatical category with identifying its sense. Both crane meanings in the example are nouns. Also, a familiar interpretation is not automatically the intended one in a specialist domain. Finally, resolving ambiguity does not establish truth, intent, or permission to act. It only addresses which interpretation the language evidence supports.

Business use

For business NLP, ambiguous terminology can affect search, routing, extraction, and generated responses. Identify words whose alternative meanings lead to different outcomes in the actual workflow. Build evaluation examples that contain both common and less frequent senses, including short inputs with insufficient context.
Use domain terminology consistently where possible, while allowing users to express themselves naturally. When the evidence is unclear, make the system's uncertainty actionable through context gathering or an appropriate clarification path. Review errors by sense rather than relying only on overall accuracy, which can look strong when one meaning dominates the data.

Check your understanding

Why does calling crane a noun not resolve the worked example?

Answer: Both the bird and lifting-machine interpretations are nouns. The system needs semantic and contextual evidence to distinguish those readings.

Related terms

Word sense disambiguation
Homonymy
Syntactic ambiguity
Natural language understanding
polysemy
lexical semantics
contextual representation
sense inventory

Learn next

Choose an ambiguous word from a familiar domain and write two sentences supporting different readings. Then remove context until both readings become plausible. Explain which additional information would resolve the uncertainty and which interpretation claims would remain unsupported.

Sources

Sources: Dan Jurafsky, original Stanford lecture, Language Perception and Comprehension, ambiguity and word-sense interpretation: https://web.stanford.edu/~jurafsky/symbsys100.2005/Lec_5_5.pdf

Previous: Lewandowski-Kurowicka-Joe distribution Next: Lexical category