AI Glossary / Natural language and speech / Beginner
Part of speech
Definition
A part of speech is a grammatical category used to describe a word's characteristic behavior, such as noun, verb, adjective, or adverb. Categories help explain how words combine and what forms they can take. In natural language processing, a part-of-speech tag is a label representing such a category under a specified annotation scheme.
What it means
Familiar explanations connect nouns with things and verbs with actions, but meaning alone is not a reliable definition. A noun can name an activity, and a verb can describe a state. Grammatical behavior, including possible combinations with other words and relevant inflections, provides stronger evidence.
A category is also different from a sentence role. A noun can be a subject in one sentence and an object in another while remaining a noun. Similarly, the written form of a word does not always identify one category: English “record” has ordinary noun and verb uses that context can distinguish.
How it works
An annotation scheme defines its categories and the distinctions it records separately. One scheme may use a broad VERB category plus grammatical features, while another includes finer distinctions directly in tags. Consistent definitions matter more than assuming every tagset divides language identically.
When classifying a word, an analyst considers its usual syntactic behavior, morphology, and the relevant context. A determiner before a word can provide evidence for a noun use, while inflected forms can provide evidence for a verb or adjective. Such clues are interpreted together rather than treated as universal rules without exceptions.
Computational systems learn or apply these distinctions to language data. The resulting category can support parsing, lemmatization, or extraction. It remains one layer of analysis: knowing a word's part of speech does not automatically determine its sense, referent, or exact relationship to every other word in the sentence.
Three practical examples
1. Search normalization: A system distinguishes noun and verb uses before deciding which base form or expansion is appropriate for a search term.
2. Grammatical analysis: A parser uses category information as evidence when proposing phrase structures or dependencies.
3. Language education: A learning tool explains why the same spelling behaves differently in two sentences, using category labels alongside concrete grammatical examples.
Think of it this way
Think of parts of speech as types of components in a language toolkit. The type suggests how a component can combine with others, but does not state its complete purpose in every assembly. The analogy has limits: language categories have exceptions and language-specific behavior, and a single written form can correspond to components of more than one type.
Worked example
Consider the original pair “They record the test” and “The record describes the test.” In the first sentence, “record” is a verb: it expresses the predicate with “They” as its subject. In the second, “record” is a noun within the subject phrase “The record.”
There are two occurrences of the same written form and two different category assignments in this pair. A dictionary that records both uses has identified possible categories; a tagging decision still needs to select the relevant one for each occurrence.
Now compare “The record disappeared” with “Mina found the record.” In both sentences, “record” remains a noun. Its grammatical role changes from subject to object, but its part of speech does not. These four sentences separate three ideas that are easily confused: spelling, category, and sentence role. They are manually constructed examples, not a frequency claim about English or a model evaluation.
Illustrative case study
Fictional case: Birch Manuals develops a terminology tool for workshop documents. Reviewers notice that “seal” can name a component or describe an action. The team stores examples of both uses and checks whether proposed search expansions preserve the distinction.
It also documents the tagset used by its language pipeline so another team does not mistake a broad category for a more detailed grammatical feature. The scenario illustrates resource design and review; it does not claim a real accuracy improvement or production outcome.
Common misconceptions and mistakes
Do not classify a word solely by a rough semantic slogan such as “an action word.” Grammatical categories need grammatical evidence. Also avoid treating subject, object, and modifier as interchangeable with noun, verb, and adjective: the first set concerns functions, while the second concerns categories.
Tagsets can differ, so a category count from one resource may not match another directly. Nor does a shared cross-language tag inventory imply that every language realizes every category in the same way. Consult the scheme and language-specific guidance when interpreting difficult cases.
Business use
Parts of speech provide a useful vocabulary for diagnosing language-system errors. A wrong category can explain an inappropriate lemma, an incorrect phrase grouping, or a misleading extraction, although some errors arise at later levels instead.
For a business application, inspect examples that affect the actual workflow. A broad category label may be sufficient for filtering candidate terms, while another task needs grammatical features and dependencies. Document these requirements so teams exchange compatible annotations and do not overinterpret what a single tag provides.
People also ask
Can one spelling belong to more than one part of speech?
Yes. Many words have established uses in different categories, and context helps distinguish them.
Is a noun always a subject?
No. A noun can participate in different sentence roles, including subject and object, without changing category.
Are all tagsets identical?
No. They can use different inventories or levels of detail, so labels should be interpreted with their annotation guidelines.
Check your understanding
In “Mina checked the record,” does “record” become a verb because it is part of an action description?
Answer: No. It is a noun in the object phrase; “checked” is the verb. The sentence's overall meaning does not change the noun's category.
Related terms
Learn next
Annotate a short sentence with broad categories, then separately identify its subject, predicate, and object. Compare the two layers and examine a word whose category changes across ordinary uses.
More in Natural language and speech
- Parse tree - A parse tree is a tree-shaped representation of a grammatical analysis.
- PARSEVAL - PARSEVAL is a family of measures for evaluating constituency parses by comparing their constituent boundaries, and commonly their labels, with a reference...
- Parsing - Parsing is the process of analyzing an expression's structure according to a grammar or structural model.
- Particle - A particle is a grammatical term for certain small function words whose interpretation depends on association with another word, phrase, or construction.
- PATR-II - PATR-II is a declarative formalism for representing linguistic information in computer-processable grammars.
- Period disambiguation - Period disambiguation is the task of deciding what a period character does in its particular context, especially whether it marks the end of a sentence.
Sources
Sources: Jurafsky and Martin, Speech and Language Processing, Chapter 18, word classes: https://web.stanford.edu/~jurafsky/slp3/18.pdf . Universal Dependencies category inventory: https://universaldependencies.org/u/pos/ . Universal Dependencies morphology and tagging principles: https://universaldependencies.org/u/overview/morphology.html . Relevant primary explanatory passages were read. Examples are original, and the article distinguishes general category concepts from scheme-specific annotation decisions.
Previous: Parsing Next: Part-of-speech tagging