AI Glossary / Natural language and speech / Beginner
Proper noun
Definition
A proper noun is a noun used as a name, or as part of a name, for a particular person, place, organization, or other entity. Mina and London are familiar kinds of examples. The contrast is with a common noun such as technician or city, which identifies a class rather than functioning as that entity's name.
What it means
Being a name does not make a word a globally unique identifier. Several people can share the name Mina, and several places or organizations can use the same name. Interpreting a proper noun therefore still requires context, just as matching a name in a database may require additional information.
A proper noun is also not identical to a named-entity label. Proper noun describes grammatical behavior; named-entity recognition typically identifies spans and assigns task-specific types such as person or organization. The two analyses can support one another, but they answer different questions.
How it works
A grammatical analysis considers how a word functions in its context. Capitalization can provide a clue in languages and writing systems that use it, but it is not a sufficient rule. The first word of an English sentence is normally capitalized even when it is a common noun, and informal writing can omit expected capitals in names.
Multiword names require care. A name may contain words that retain other grammatical categories under a particular annotation scheme. Universal Dependencies explicitly distinguishes the name as a whole from the tags assigned to its component words and advises against transferring English tagging conventions to other languages without linguistic justification.
After grammatical identification, a separate task may determine which real-world entity the name denotes. This process can compare context, location, dates, and other available evidence. It should also allow unresolved matches. A high-confidence proper-noun tag does not establish a verified identity, and identical spelling does not prove that two mentions concern the same entity.
Three practical examples
1. Contact extraction: A system identifies names in a document, then checks additional context before linking them to existing contact records.
2. Translation review: A reviewer checks whether a name should be preserved, transliterated, or handled according to the project's documented naming convention.
3. Corpus annotation: Annotators distinguish proper-noun tags from the broader spans used by a named-entity task, especially for multiword organization or place names.
Think of it this way
Think of a proper noun as a name on a label rather than a category on a shelf. The label helps identify a particular intended entity. The analogy is limited because labels can be shared, changed, abbreviated, or ambiguous, and a name alone is not a reliable record identifier.
Worked example
Consider the original sentence “Mina visited Paris with Ravi.” Ignoring punctuation, its five tokens are Mina, visited, Paris, with, and Ravi. In the intended ordinary reading, Mina, Paris, and Ravi are three proper-noun tokens. Visited is a verb and with is a preposition.
Now consider “Visitors photographed the city.” Visitors begins with a capital because it starts the sentence, but it is a common noun in this example. A toy rule that labels every capitalized token as a proper noun would identify the three names in the first sentence and incorrectly label Visitors in the second.
Across these two stipulated sentences, the rule predicts four proper-noun tokens, of which three are correct. Its precision is 3/4=75%. It finds all three true proper-noun tokens, giving recall 3/3=100%. These independently checked counts describe only the invented example and its stated reading. They are not a general estimate of capitalization's usefulness, and they do not identify which real people named Mina and Ravi are intended.
Illustrative case study
Fictional case: Rowan Archive indexes correspondence and initially merges every occurrence of the same personal name into one contact. Reviewers discover that two contributors share that name. Both mentions were correctly recognized as names, but the identity merge was unsupported.
The team separates grammatical tagging from entity matching and uses available document context to review possible links. It keeps unresolved cases separate until there is sufficient evidence. This fictional case illustrates why name recognition and identity resolution need different evaluation checks; it does not claim a real archive or measured improvement.
Common misconceptions and mistakes
Do not treat every capitalized word as a proper noun or every lowercase word as a common noun. Writing conventions, sentence position, and informal text can all affect capitalization.
Also avoid assuming that every token inside a named expression receives a proper-noun tag. Follow the annotation framework and language-specific guidance. Finally, a name's grammatical status does not establish whether its referent exists, is unique, or has been correctly linked to an external record. Fictional names can function as names in text too.
Business use
For business data workflows, separate name detection from record matching. Test sentence-initial common nouns, names in inconsistent casing, multiword names, and shared names. Track errors at each stage so a successful grammatical classifier is not credited with identity decisions it never made, and a mistaken merge is corrected at the appropriate level.
People also ask
Does a proper noun always identify one unique entity worldwide?
No. Different entities can share a name, so context is still needed.
Is capitalization enough to recognize proper nouns?
No. It is a useful clue in some writing systems, but sentence position and writing style can mislead.
Is proper-noun tagging the same as named-entity recognition?
No. One assigns grammatical categories; the other identifies and types entity mentions according to a task's scheme.
Check your understanding
Why is Visitors a false positive for the capitalization rule in the worked example?
Answer: It is capitalized because it begins the sentence, but in that sentence it names a class of people rather than functioning as a particular entity's name.
Related terms
- Common noun
- Part-of-speech tagging
- Entity linking
- Noun phrase
- Tokenization
- named entity
- name ambiguity
- capitalization
Learn next
Learn common versus proper nouns in complete sentences, then compare token-level grammatical tags with named-entity spans. Next study entity linking and shared-name ambiguity. Build small examples that isolate capitalization errors from identity-matching errors so the two problems remain distinguishable.
More in Natural language and speech
- Productions - Productions are rules in a formal grammar that specify how grammatical symbols can be expanded or rewritten.
- Pronoun - A pronoun is a grammatical expression that can serve a noun-like role without naming its referent in the same way as a full noun phrase.
- Pronunciation assessment - Pronunciation assessment is the evaluation of how speech is pronounced according to a stated purpose and set of criteria.
- Prosodic phrasing - Prosodic phrasing is the grouping of spoken words into units through intonation, timing, and other speech cues.
- Prosody - Prosody is the intonational and rhythmic organization of speech, including patterns of prominence, phrasing, pitch, and timing.
- PROTO - PATIENT - PROTO-PATIENT is a semantic proto-role: a cluster of properties associated with how a participant is affected in an event.
Sources
Universal Dependencies, PROPN definition and multiword-name guidance: https://universaldependencies.org/u/pos/PROPN.html
Jurafsky and Martin, Speech and Language Processing, chapter 18, word-class discussion: https://web.stanford.edu/~jurafsky/slp3/18.pdf
The example sentences, capitalization-rule scores, and archive case are original teaching material. The grammatical readings are stipulated for the exercise and are not presented as corpus statistics.
Previous: Proper generalized decomposition Next: Proper policy