A part-of-speech tag is a grammatical label assigned to a word or token in context. Tags identify categories such as noun, verb, adjective and determiner, sometimes using abbreviations such as NOUN, VERB, ADJ or DET. Unlike a dictionary definition, a tag tells us what grammatical job the word is performing in a particular sentence. A tagged corpus makes it possible to search for relevant kinds of word use, not merely matching spellings.
Why does this matter for vocabulary? A word such as record can be a noun in ‘Keep a record’ and a verb in ‘Record the result’. Students who know only its dictionary headword may confuse the patterns that belong to each use. POS tags provide a way to separate the examples, inspect grammatical relationships and choose the expression that accurately communicates the intended event.
This guide explains what parts of speech are, why tagsets differ, how automatic annotation works, what its limitations are and how to apply it in Primary and Secondary English, Mathematics, Science, History, translation and daily life. The Penn Treebank and Universal Dependencies frameworks are used as documented references, with original exercises rather than invented corpus statistics.
Editorial note: All examples written for this guide are illustrative. No automatically generated tag is treated as an infallible grammatical judgment or an assessment of a student’s ability.
Contents: grammar labels as a technology for finding meaning
Open 44 chapters, classroom exercises and practical checks
- 1. A tag identifies grammatical function in context
- 2. Why the same spelling can do different work
- 3. A tag belongs to a tagset
- 4. Nouns name entities and ideas
- 5. Verbs describe actions, states and relations
- 6. Adjectives modify or predicate properties
- 7. Adverbs can modify clauses and actions
- 8. Determiners have their own role
- 9. Pronouns depend on reference
- 10. Auxiliaries and main verbs are different
- 11. Modal verbs change the force of a claim
- 12. Prepositions express important relationships
- 13. Conjunctions organise reasoning
- 14. Numbers and punctuation may receive tags
- 15. A Penn Treebank tag is not an ordinary school grade
- 16. Universal Dependencies uses broader categories
- 17. A tag is different from a dictionary usage label
- 18. A tag is different from a lemma
- 19. Tags are assigned by humans or software
- 20. Ambiguity is normal, not a software scandal
- 21. Gerunds and participles need care
- 22. Proper names are not just capitalised nouns
- 23. Part-of-speech tags help corpus queries
- 24. Tag counts need denominators and definitions
- 25. Sentence position can mislead a tagger
- 26. Singapore English is an important test case
- 27. Other languages may divide categories differently
- 28. A dependency relation is not a POS tag
- 29. A semantic label is also different
- 30. Primary English: identify simple jobs
- 31. Secondary English: repair a sentence
- 32. Composition voice needs more than grammar tags
- 33. Mathematics: a verb tells the student what to do
- 34. Science: nouns and verbs encode different claims
- 35. History: verbs can change interpretation
- 36. Translation: the target language may reorganise grammar
- 37. Workplace procedures need exact grammatical actors
- 38. What an automated tagger can and cannot prove
- 39. Inter-annotator agreement measures consistency
- 40. Protect the provenance of annotated examples
- 41. An effective four-column practice sheet
- 42. The five-check grammar audit
- 43. Do not turn grammatical labels into rankings
- 44. The final transfer task
Two tagsets and what the symbols mean
| Grammatical concept | Universal Dependencies | Illustrative Penn Treebank label |
|---|---|---|
| Common noun | NOUN | NN (singular/mass); NNS (plural) |
| Main verb | VERB | VB (base); VBD (past); others |
| Adjective | ADJ | JJ |
| Adverb | ADV | RB |
| Determiner | DET | DT |
| Auxiliary or modal | AUX | MD (modal); other verb tags for auxiliaries |
| Proper noun | PROPN | NNP or NNPS |
| Punctuation | PUNCT | Punctuation tags vary by mark |
These descriptions simplify richer manuals and are not a complete one-to-one conversion. A universal tag and a Penn Treebank tag can group features differently. Always consult the actual tagset for the corpus being searched.
1. A tag identifies grammatical function in context
A part-of-speech tag labels a word or token according to its grammatical category in a particular context. Familiar categories include noun, verb, adjective, adverb, pronoun and determiner. A tagged corpus can attach codes to each occurrence, allowing a researcher to distinguish the verb record from the noun record.
The key phrase is in context. A visible spelling is not always tied to one part of speech. The sentence determines what job the word is doing. The tag is an annotation of that grammatical role, not an evaluation of how intelligent or appropriate the speaker is.
2. Why the same spelling can do different work
In ‘We will record the result’, record is a verb. In ‘Keep a record of the result’, it is a noun. The sequence of letters is the same, but the words occupy different positions and have different grammatical relationships.
If a corpus query gathers both uses together, a student looking for verb complements could mistakenly learn patterns from noun phrases. A part-of-speech tag helps separate relevant examples, provided the annotation is appropriate and any uncertain instances are checked.
3. A tag belongs to a tagset
A tagset is a defined inventory of grammatical labels and rules for applying them. The Penn Treebank tagset, for example, uses symbols such as NN for certain nouns, VB for a verb base form and JJ for adjectives. Universal Dependencies uses a different universal tag inventory, including NOUN, VERB and ADJ.
These systems should not be mixed casually. A code that means one thing in one dataset may not exist or have exactly the same scope in another. The researcher reads the documentation for the actual corpus rather than assuming every tagger uses the same shorthand.
4. Nouns name entities and ideas
A noun can denote a person, object, place, event or abstract concept. In ‘The decision mattered’, decision functions as a noun. Tagged corpora can help learners find examples of how nouns act as subjects or objects and how they are modified.
But simply spotting a noun does not explain what the noun means in that sentence. Light may refer to illumination in one sentence and have a different role elsewhere. The learner must still interpret sense, reference and grammatical context.
5. Verbs describe actions, states and relations
Verbs are central to clauses because they can express processes, actions, states or relationships. In ‘The students revise their explanation’, revise functions as a verb and takes an object. The grammatical role helps a learner ask what is being revised and by whom.
A writing correction should therefore go beyond replacing one unfamiliar verb with another. The tutor needs to protect the sentence’s participants and logic. Using a verb tag to retrieve comparable examples can help, but the real intended event decides which structure is appropriate.
6. Adjectives modify or predicate properties
Adjectives can modify nouns or describe properties in clauses, as in ‘a careful result’ or ‘the task is difficult’. The first invented phrase may not be the best expression for the intended meaning, which shows why classification alone cannot guarantee natural wording.
A tagged corpus can help locate candidate adjective uses for discussion. The teacher then checks whether the adjective describes a genuine property of the noun and whether the wording suits the genre. A tag identifies a function; it does not certify a collocation.
7. Adverbs can modify clauses and actions
Adverbs cover varied functions, from manner to frequency and degree. In ‘She checked the result carefully’, carefully describes how the checking happened. In ‘Fortunately, the test was repeated’, the initial adverb positions the speaker’s attitude toward the whole situation.
Tagging these occurrences as adverbs may be useful, but it does not show that they are interchangeable. A learner who wants to improve writing should ask what information an adverb adds and whether that information is actually supported by the story or evidence.
8. Determiners have their own role
Words such as the, a, this and those can function as determiners. They help specify how a noun phrase refers to something. Missing or inappropriate determiners can make a sentence unclear even when the main nouns and verbs are correctly chosen.
In some tagsets, words that look similar can be tagged differently depending on whether they determine a noun or function as pronouns. Grammar is therefore a structure to examine, not a table of permanent labels assigned to spellings.
9. Pronouns depend on reference
Pronouns such as she, it and they refer to participants or entities whose identities may be supplied by context. A pronoun tag helps identify the category, but the tag does not reveal which person or object is intended.
In a school composition, the sentence ‘Aisha spoke to Farah because she was worried’ may be ambiguous without surrounding information. Part-of-speech tagging can tell us that she is a pronoun; reference-chain analysis is needed to interpret who was worried.
10. Auxiliaries and main verbs are different
An auxiliary can express tense, aspect, modality or other grammatical information in a clause. In ‘She has finished’, has functions differently from ‘She has a notebook’. Universal Dependencies distinguishes AUX from VERB under its guidelines, although not all tagging systems draw boundaries in identical ways.
A learner should not treat every appearance of have as the same grammatical construction. Look at what follows, what role the word plays and which tagset is being used. This distinction matters for writing accurate statements about time, obligation and possibility.
11. Modal verbs change the force of a claim
Words such as may, might, must and should can change permission, likelihood or obligation. Some English tagsets assign a special modal category; Universal Dependencies may analyse them as auxiliaries according to its rules.
In Mathematics, Science and policy writing, the difference between may and must can alter what is being asserted. A POS tag helps locate candidate clauses but cannot by itself tell the reader whether the obligation is legally valid or the probability is supported by evidence.
12. Prepositions express important relationships
Prepositions often mark locations, times, directions and other relations: in a room, after the lesson, because of the rain. In some systems, a broader adposition category includes both prepositions and postpositions across languages.
Learning a word’s grammatical partners helps prevent errors such as selecting a preposition that gives a different meaning. Yet the correct form may depend on the particular verb, adjective or noun. A tag is a starting point for checking examples, not a universal substitution rule.
13. Conjunctions organise reasoning
Conjunctions link words, phrases and clauses. And can coordinate, while although can introduce a concessive relation. A tagset may distinguish coordinating and subordinating conjunctions, as Universal Dependencies does with CCONJ and SCONJ.
This distinction matters because words connect propositions, not merely decorate sentences. A student who replaces although with because changes the logical relation. Part-of-speech analysis can clarify the structure, but the writer must preserve the intended reasoning.
14. Numbers and punctuation may receive tags
Many annotated corpora label numerals and punctuation. Universal Dependencies includes NUM and PUNCT, along with categories such as SYM for certain symbols. That helps computational tools retain the structure of text rather than treating every visible character as an ordinary word.
For educational language analysis, counting punctuation-tagged tokens as ordinary words can distort the denominator. The researcher should identify what the corpus calls a token and what categories are included in a frequency table.
15. A Penn Treebank tag is not an ordinary school grade
The Penn Treebank inventory includes multiple noun and verb subcategories. For example, NN is used for a singular or mass noun, NNS for a plural noun, VBD for past-tense verbs and VBG for a gerund or present participle form. These are descriptive codes, not quality scores.
Students may encounter these labels in digital language resources or coding lessons. They should first translate the symbols into plain grammar terms and confirm their meaning from the named tagset. There is no advantage in memorising cryptic codes without understanding how sentences work.
16. Universal Dependencies uses broader categories
The Universal Dependencies framework provides a cross-linguistic inventory of universal POS tags. Examples include ADJ, ADP, ADV, AUX, DET, NOUN, PRON, PROPN, VERB and PUNCT. The precise application of categories follows its published guidelines.
This system helps language datasets share a general grammatical framework, but languages can still differ in their constructions and segmentation rules. Universal labels do not mean every language has identical grammar or that language-specific decisions have disappeared.
17. A tag is different from a dictionary usage label
A dictionary’s informal, literary or offensive label describes circumstances or social effects of usage. A POS tag such as NOUN or VERB describes a grammatical category. These answer different questions, even when both are presented as small labels in a digital interface.
Confusing them can lead students to think an adjective is formally superior or a noun is more appropriate for an exam. Grammar tags say what a word does in a sentence; usage labels help readers assess audience, tone and other contextual conditions.
18. A tag is different from a lemma
A lemma groups related inflected forms under a base entry. A POS tag identifies grammatical category at a particular occurrence. A tagged and lemmatised corpus can use both: wrote may be linked to the lemma write and tagged as a past-tense verb under a particular scheme.
The distinction gives researchers finer control. They can search the verb family while excluding unrelated noun uses. But lemmatisation and tagging are automatic processes in many systems and may fail on ambiguous or specialist expressions.
19. Tags are assigned by humans or software
A corpus may be annotated manually, automatically or through a combination of methods. Human annotators use guidelines, while automatic POS taggers learn or apply methods to predict likely categories from context. Both can disagree, especially around unusual constructions.
The proper response is to examine the guidelines and validate a sample. A high accuracy claim from an unrelated benchmark cannot simply be transferred to a new classroom dataset. Annotation reliability depends on the language, genre, method and intended classification.
20. Ambiguity is normal, not a software scandal
Consider the word book. In ‘I read a book’, it is a noun; in ‘Please book a room’, it is a verb. The surface form is ambiguous until placed in a sentence. A tagging tool must use the context to choose.
If the program mislabels an ambiguous token, the student should inspect the clause and ask why the evidence supports a different category. This is not a reason to abandon tagging; it is an opportunity to teach how grammar depends on relationships rather than isolated spelling.
21. Gerunds and participles need care
English forms ending in -ing can play different roles. ‘She is running’ includes a verbal participle, while ‘Running helps me concentrate’ may be analysed differently according to the grammatical framework and the function of the clause.
A rule that every word ending in -ing is simply a verb will misclassify some examples. Teach learners to examine the sentence and consult the appropriate grammar framework. The aim is to make word behaviour predictable enough for writing, not to enforce a crude suffix rule.
22. Proper names are not just capitalised nouns
Proper nouns refer to named people, places and entities. Universal Dependencies distinguishes PROPN from NOUN, but capital letters alone cannot reliably identify them. A word at the start of a sentence is capitalised regardless of category, and brand or place names can resemble ordinary words.
For example, May can be a month or a person’s name, while may can be a modal. Accurate tagging uses grammatical and contextual information. A case-sensitive search is helpful in some situations but is not a substitute for analysis.
23. Part-of-speech tags help corpus queries
A student investigating the verb record can use a suitable tagged corpus to restrict examples to verbs. This may reduce noise from noun occurrences and make the remaining concordance easier to interpret. The query’s exact syntax depends on the corpus and tool.
Filtering by tag is not infallible: some relevant examples may be misclassified and some returned items may be false matches. State the tagset and inspect samples. A good query is a transparent question about observable language, not a promise of automatic truth.
24. Tag counts need denominators and definitions
A corpus may contain different numbers of nouns, verbs or adjectives across genres. Raw totals depend on collection size, tokenisation and annotation. Comparisons should identify the category system, relevant denominator and sample composition.
If one dataset includes punctuation as tokens and another does not, proportions may not be directly comparable. If tags were assigned by different systems, label mappings must be justified. Numbers are useful only after the linguistic units and categories are made clear.
25. Sentence position can mislead a tagger
A word’s meaning and grammatical role can depend on distant context, ellipsis, quotations or nonstandard punctuation. Short fragments may give a tagger insufficient information. Dialogue, poetry and casual digital messages can also contain structures underrepresented in the system’s training material.
When interpreting a surprising label, open the whole sentence or paragraph. Ask whether the example contains missing words, irony, reported speech or an unusual register. Language software should prompt closer reading, not discourage it.
26. Singapore English is an important test case
Singapore English includes standard academic and colloquial forms, alongside multilingual influence and code-switching in some contexts. A tagger trained mostly on another variety may make less reliable predictions on certain local patterns. That is a potential domain mismatch, not a judgement about the speaker.
For school instruction, teach the register required by the task while respecting legitimate language variation. Where an automated tag seems unusual, check the construction and intended meaning instead of assuming that local speech is defective.
27. Other languages may divide categories differently
Not every language expresses tense, number, definiteness and grammatical relations using the same visible forms as English. A universal tagset offers broad comparability while retaining language-specific guidance and morphological features.
A multilingual vocabulary project should therefore use the right language documentation and not blindly apply an English tagger. Good translation preserves the speaker’s event, relationship and force rather than matching grammatical category names mechanically.
28. A dependency relation is not a POS tag
A POS tag describes a token’s grammatical category. A dependency analysis describes a relationship between tokens, such as a noun serving as the subject of a verb or an adjective modifying a noun. These complement one another but are not the same annotation.
Knowing that evidence is a noun does not reveal which verb governs it in every sentence. A word sketch or dependency analysis can add relational information. Students need both the word’s role and the clause’s structure to make an accurate revision.
29. A semantic label is also different
A semantic tag may place a word occurrence in a meaning category such as people, movement, measurement or emotion, depending on the scheme. A POS tag identifies grammatical role. Two noun occurrences may denote very different kinds of things.
Combining grammatical and semantic evidence can improve a search, but the classifications should not be confused. A student’s word may be grammatically sound yet express the wrong concept. A full vocabulary lesson asks both what job the word does and what it means.
30. Primary English: identify simple jobs
Give a Primary pupil three original sentences: ‘The children play’, ‘The play was exciting’ and ‘They play outside.’ Ask the learner which occurrence names an activity as a thing and which describes an action.
This teaches grammatical categories through meaning and sentence structure. The teacher need not introduce database tag codes immediately. Once the child can justify a choice, the labels noun and verb become useful names for already understood patterns.
31. Secondary English: repair a sentence
Suppose a learner writes, ‘The carefully student checked the paper.’ The issue is not that carefully is an unknown word; the adverb does not work as the intended ordinary adjective before student. The writer might mean ‘The careful student checked the paper’ or ‘The student carefully checked the paper.’
The two revisions have different emphases: one describes the student, the other the action. Explaining the role helps the learner choose deliberately. A grammar tag is valuable when it leads to that decision, not simply when the pupil can recite a code.
32. Composition voice needs more than grammar tags
An ornate adjective can be grammatically correct yet unsuitable for a young narrator speaking casually. A regional expression can be meaningful in dialogue yet inappropriate in a formal report. A POS tag does not resolve tone, audience or cultural context.
Combine grammatical analysis with dictionary usage labels, examples and the intended character voice. The goal is wider control over language, not a machine-generated list of forbidden words.
33. Mathematics: a verb tells the student what to do
Mathematics questions use verbs such as calculate, estimate, justify and simplify. Classifying them as verbs is elementary; distinguishing their requested operations is the higher-level skill.
A student may calculate the right numerical value without supplying the justification the question requires. Vocabulary teaching must connect grammar with meaning and mathematical expectation. The problem statement, not a corpus tag, determines the correct working.
34. Science: nouns and verbs encode different claims
Compare measurement, measure, prediction and predict. Derivational relationships can change part of speech while preserving a conceptual family. A student needs to form sentences that keep clear who measured something, what was measured and what remained uncertain.
Tags can help organise the vocabulary family, but they cannot turn an inference into a measured result. Scientific accuracy requires the writer to distinguish data, method, interpretation and evidence strength.
35. History: verbs can change interpretation
A source might say that a government introduced, imposed or reversed a policy. Each form functions as a verb, but the verbs express different relationships and evaluative effects.
A POS tag will not decide whether a historical source is fair. Students should analyse word meaning, source context and corroborating evidence. Grammar is necessary to decode the sentence; historical reasoning is necessary to evaluate it.
36. Translation: the target language may reorganise grammar
A noun in one language may be translated through a verb phrase in another, or the target language may express information with particles or morphology. Preserving the grammatical tag across languages is not always the same as preserving meaning.
The translator should identify what happens, who is involved and what conditions apply. Then choose a natural structure in the target language. Part-of-speech information may aid analysis, but literal category matching can distort the message.
37. Workplace procedures need exact grammatical actors
In a sentence such as ‘The manager reviewed the form’, the verb and its subject identify an action and the person who performed it. If the sentence becomes passive, information about the actor may be omitted or backgrounded.
A POS-tagged search can help compare forms, but process accuracy still requires the real approval trail. Do not infer that a task was completed merely because a sentence uses the appropriate verb. Grammar makes responsibilities visible; evidence establishes whether they were fulfilled.
38. What an automated tagger can and cannot prove
An automated tagger predicts labels using language patterns and whatever model or rules it was built with. A plausible label may still be wrong on a sentence outside its expected domain. A label is not a measure of an author’s talent or competence.
In educational settings, inspect errors before correcting students. The tool can suggest where to look but cannot replace teacher judgement about intended meaning. If a system repeatedly fails on a legitimate language variety, its limitations should be documented rather than blamed on users.
39. Inter-annotator agreement measures consistency
When human annotators tag language, they may disagree about ambiguous cases. Agreement measures can help quantify consistency under defined methods, but raw agreement can be misleading when one category is much more frequent than others.
A school project can make disagreements visible without attaching statistical significance it has not calculated. Ask annotators to explain the rule they used and revise the annotation guide if needed. Consistency matters because tag-based searches depend on a shared classification.
40. Protect the provenance of annotated examples
A tagged corpus may display words, labels, sources and dates. When preparing a worksheet, identify where authentic examples came from and follow the relevant rights or licence conditions. Original constructed sentences can be labelled separately.
Do not publish private student writing simply to demonstrate a tagger. Remove identifying information and follow school rules before processing genuine learner records. Technical sophistication should not undermine ethical handling of language data.
41. An effective four-column practice sheet
Use four columns: sentence, target word, grammatical role and evidence from the sentence. For older students, add the code from a named tagset and a note about uncertainty. This makes the decision inspectable.
For example, compare ‘I will book the room’ with ‘I bought a book’. The learner should be able to justify the categories by explaining the role of each occurrence. A code without a supporting reason is memorisation, not understanding.
42. The five-check grammar audit
Ask five questions: What is the whole sentence? What job does the target word do? Which neighbouring words show that role? Does the chosen tagset define the category this way? Could another analysis be reasonable in this context?
This routine is suitable for debugging software output and for teaching grammar. It encourages students to support a classification with evidence rather than guessing from the word’s spelling. The strongest outcome is a correct sentence, not a perfect-looking tag table.
43. Do not turn grammatical labels into rankings
Nouns are not superior to verbs, and advanced pupils are not automatically better writers because they produce more adjectives. Effective texts use grammatical resources according to their purposes. A clear instruction may contain short verbs and few modifiers; a description may require different structures.
A vocabulary lesson should show how language choices change the reader’s understanding. Counting categories can support analysis, but it must not become a substitute for clarity, accuracy, style and argument quality.
44. The final transfer task
Choose a real sentence with a word that can act as more than one part of speech. Identify its actual role, compare an alternative sentence and explain how the meaning changes. Then write a new example without copying the originals.
This demonstrates the technology behind POS tags: they convert an intuitive difference into a shared description that can be searched, tested, taught and revised. When used carefully, grammatical labels make language more navigable without claiming to replace human understanding.
An original practice set
- Identify the role of book in ‘Please book a room’ and in ‘I finished the book’.
- Identify the role of record in ‘Record your result’ and ‘Read the record’.
- Explain why carefully differs from careful in a sentence about checking a result.
- Distinguish has in ‘She has completed the task’ from ‘She has a notebook’.
- Find a pronoun that could refer to more than one person and explain why the tag alone cannot resolve the reference.
- Write one new sentence using a familiar word in a different grammatical role, then explain how the meaning changes.
Ten frequently asked questions
1. What does POS mean?
Part of speech, the grammatical category a word occupies in context.
2. Is a POS tag a dictionary meaning?
No. It describes grammatical role, not the full meaning of a word.
3. Can the same word have different tags?
Yes. Context can make one spelling a noun in one sentence and a verb in another.
4. What is a tagset?
A documented inventory of grammatical labels and conventions.
5. Are Penn Treebank and Universal Dependencies the same?
No. They use different category schemes and levels of detail.
6. Can tagging software make mistakes?
Yes. It can misclassify ambiguous, unusual or unfamiliar constructions.
7. Do POS tags tell me whether a word is polite?
No. Register and usage labels concern different aspects of language.
8. Why do corpus researchers use tags?
To search or analyse grammatical patterns more precisely.
9. Do tag counts grade good writing?
No. The value of a sentence depends on its accuracy, purpose and reader, not merely the number of each grammatical category.
10. What should a student practise?
Explain the word’s job in a real sentence and use it accurately in a fresh one.
Primary sources and related lessons
- Universal Dependencies: Universal POS Tags — official definitions and category boundaries.
- Universal Dependencies: POS Tag Inventory — the complete set of universal grammatical labels.
- American National Corpus: Penn Treebank Tagset — English tag codes and their meanings.
- Sketch Engine: Concordance Guide — how tagging supports more selective corpus queries.
- Natural Language Toolkit: Processing Raw Text — why tokenisation and annotation belong to an explicit text-processing pipeline.
- Vocabulary | What Is a Corpus Query? — use grammar annotations in searching.
- Vocabulary | What Is a Usage Label? — distinguish grammatical tags from register guidance.
- Vocabulary | What Is a Word Token? — understand the occurrences receiving tags.
- Vocabulary Learning Hub — the wider school-to-life vocabulary curriculum.
A POS tag is valuable because it makes a grammatical distinction visible and searchable. That visibility should lead to better thinking: read the sentence, identify the role, preserve the meaning and write with greater control.