Corpus, Dictionary and Reference Literacy: how to verify English instead of guessing

Reference literacy is the ability to choose the right language evidence for the question you actually have. A dictionary can tell you a word’s senses, pronunciation, grammatical behaviour and usage labels. A collocations dictionary can show which words commonly occur together. A corpus can show patterns across thousands or millions of authentic examples. A historical dictionary can trace earlier senses and dated evidence. A style guide can tell you what one institution prefers. An examination specification can tell you what a board requires. A search engine can reveal what exists on the web but not automatically what is representative or reliable. An AI system can generate a plausible answer without proving that the answer is supported.

These tools solve different problems. Confusion begins when a learner treats them as interchangeable.

A student asks, “Can I say do a decision?” A general dictionary may define decision correctly but fail to make the collocational problem obvious. A corpus or collocations dictionary will show that English strongly prefers make a decision. Another student asks whether colour is wrong because an American spellchecker changed it to color. The question is not collocation at all; it is national spelling convention. Another learner asks what awful meant in an eighteenth-century text. A present-day learner dictionary may be insufficient; historical evidence matters. Another asks whether a phrase is appropriate in an academic paper. Frequency alone will not answer that unless the evidence is drawn from the relevant register.

This is SETC’s guide to corpus, dictionary and reference literacy. It teaches learners, tutors, writers and researchers how to verify English usage systematically: definitions, senses, pronunciation, grammar patterns, word families, collocations, register, regional variation, frequency, discourse patterns, historical development and style conventions.

It sits between several SETC owners. Vocabulary in Use teaches usable lexical control. Advanced English Vocabulary Mastery develops C1–C2 precision and phrase control. World Englishes & Variation explains regional and social variation. History of English & Language Change explains longitudinal evidence. Academic English applies source and evidence literacy to study and research. This owner provides the tool-selection system underneath all of them.

The first rule: define the uncertainty before choosing the reference

“Is this English?” is too broad. A better question identifies the uncertainty:

• What does this word mean here?
• Which sense is active?
• How is it pronounced?
• Which syllable is stressed?
• What word class is it?
• Which preposition or grammatical pattern follows it?
• Which words commonly occur with it?
• Is it formal, informal, approving, disapproving, dated or regional?
• Is it common in academic writing or mostly conversation?
• Is this British, American, Singaporean or another regional form?
• Was this sense used historically?
• Is the phrase grammatical but unusual?
• Is it common because of one narrow internet community?
• Does an examination or publisher require a specific convention?
• Is this usage changing?

Each question points toward different evidence.

A reference-tool map

Dictionaries are databases of decisions

A dictionary entry looks simple because enormous editorial work is compressed into a small space. Editors decide how to separate senses, which examples illustrate them, which labels matter, how pronunciation is represented, what grammar information belongs in the entry and when a new word or sense has enough evidence to include.

Different dictionaries make different choices because they serve different users. A historical dictionary and a learner dictionary should not look the same.

Learner dictionaries

Learner dictionaries are designed to help people who are learning English understand and produce language. They often provide controlled definitions, grammar patterns, pronunciation, usage labels, collocations and carefully chosen examples.

This makes them especially useful for productive questions such as “Which preposition follows this adjective?” or “Is this word formal?”

General dictionaries

General dictionaries serve broad audiences and may include wider vocabulary, etymology or usage information but less pedagogical scaffolding.

Historical dictionaries

Historical dictionaries organise words through time. They can trace dated senses, earlier spellings and historical quotations. They answer questions that current dictionaries cannot: when a sense is attested, how meaning developed, and how earlier forms differed.

Bilingual dictionaries

Bilingual dictionaries connect languages and can accelerate access to meaning. Their danger is false equivalence. Two translation equivalents may overlap without matching in collocation, register or grammar.

Use the bilingual dictionary to enter the concept; verify productive English use in a strong monolingual learner dictionary or corpus when precision matters.

Specialist dictionaries

Law, medicine, linguistics, engineering and other fields develop specialised dictionaries and glossaries. A general dictionary may give an everyday sense where the discipline uses a technical definition.

Pronunciation dictionaries

Pronunciation references focus on phonemic form, stress and sometimes multiple variants. They should be read alongside World Englishes awareness: one dictionary may privilege selected British or American models and cannot represent every legitimate accent.

Read the whole entry, not only the first definition

Learners often type a word into a dictionary, copy the first definition and leave. That wastes most of the entry.

A strong lookup checks:

• word class;
• sense number;
• pronunciation and stress;
• countability for nouns;
• transitivity for verbs;
• complement pattern;
• common prepositions;
• collocations;
• regional or register labels;
• example sentences;
• word family;
• cross-references.

The relevant information depends on the question.

Sense selection

Polysemous words have multiple related senses. The correct dictionary sense is determined by context, not by whichever definition looks familiar.

Suppose a learner encounters issue. It can refer to a topic or problem, an edition of a publication, an act of distribution, offspring in legal contexts and several specialised uses. The surrounding grammar and domain narrow the choice.

A good dictionary lookup is therefore a reading task.

Labels are part of meaning

Oxford Learner’s Dictionaries, for example, uses labels such as formal, informal, approving, disapproving and figurative to help users understand the situations and attitudes associated with words. These labels are not decoration. They tell the learner what social work the word may do.

A definition can be semantically correct while the word remains pragmatically wrong for the context.

Examples are evidence of pattern

Dictionary examples show how a word behaves inside sentences. Look for:

• position;
• preposition;
• article use;
• typical object;
• register;
• collocation;
• meaning in context.

Do not copy the example as a template without understanding the relationship.

Collocation: why possible English can still sound wrong

Collocation refers to words that occur together with greater-than-random or conventionally strong association. English speakers make a decision, pose a question, conduct research, strongly recommend, heavy rain and deep concern.

Another combination may be grammatically possible but unusual. Collocational control is one reason advanced learners can produce error-free sentences that still sound non-idiomatic.

Oxford’s Collocations Dictionary is explicitly designed for productive use and groups collocations by grammatical category and meaning, with examples in context. Corpora provide an even wider empirical view of how combinations distribute across real texts.

The corpus idea

A corpus is a structured collection of language data used for analysis. The texts or recordings are selected according to some design: country, period, register, genre, speaker population or research purpose.

A corpus turns usage questions into searchable evidence.

Instead of asking “Does this sound right to me?”, a learner can ask:

• How often does this phrase occur?
• What words usually appear before or after it?
• Which grammatical patterns occur?
• Is it more common in conversation or academic writing?
• Does it appear more in British or American data?
• Has it increased over time?
• Which meanings dominate?

A corpus is a sample, not English itself

Every corpus is built from particular data. A billion-word corpus can still be wrong for your question if it contains the wrong population or register.

The Corpus of Contemporary American English (COCA), for example, contains more than a billion words across multiple genres and is designed to allow comparisons across a large contemporary American English dataset. That makes it powerful for many usage questions, but it does not automatically represent Singapore English, historical English or every specialised professional domain.

The first corpus question should therefore be: What language does this corpus contain?

Representativeness

Representativeness is not a binary property. A corpus can be representative of one defined domain and unrepresentative of another.

Ask:

• spoken or written?
• which countries?
• which dates?
• which genres?
• which speakers?
• edited or spontaneous language?
• balanced across categories?
• how large is each subcorpus?

Frequency

Frequency answers “how often in this dataset?” It does not automatically answer “is this correct?”, “is this formal?”, “is this appropriate?” or “should I use it?”

A frequent informal phrase may be unsuitable in a research paper. A rare technical term may be exactly correct in one discipline. A highly frequent error in learner writing remains an error relative to the target system.

Raw frequency and normalised frequency

If two corpus sections contain different numbers of words, raw counts cannot be compared directly. Researchers often normalise frequencies per million words or another common denominator.

That makes the comparison proportional.

Concordance lines

A concordance displays search results with the target word or phrase aligned inside surrounding context, often called KWIC: key word in context.

Concordance reading is a form of pattern recognition. Instead of reading each sentence as a separate story, look vertically:

• which prepositions repeat?
• which nouns appear as objects?
• which modifiers recur?
• which grammatical frames dominate?
• which senses can be separated?

Collocates

Corpus tools can calculate words that occur unusually often near a target. Statistical association measures help distinguish meaningful co-occurrence from words that are merely frequent everywhere.

Learners do not need advanced statistics to benefit. They do need to understand that a collocate list reflects both frequency and the corpus design.

N-grams

An n-gram is a sequence of n items, often words. Searching recurrent two-word, three-word or longer sequences can reveal phraseology such as as a result of, on the other hand or discipline-specific frames.

Frequent phrase frames are valuable because fluent language is partly built from recurring multiword patterns, not invented word by word every time.

Corpus evidence and learner intuition

Intuition improves with exposure but remains limited by personal experience. A British editor, Singapore teacher and American student may have different intuitions about the same phrase because their input differs.

Corpus evidence does not make intuition useless. It gives intuition something to test.

Quick Read

• Reference literacy means choosing the right evidence for the language question you actually have.
• Use dictionaries for senses, grammar, pronunciation and usage labels; corpora for frequency, collocation, register and attested patterns; style guides and official specifications for institutional conventions.
• Frequency is not correctness. Common does not automatically mean standard, formal or appropriate; rare does not automatically mean wrong.
• Read the whole dictionary entry: sense, word class, countability, complementation, collocation, labels and examples all matter.
• Search engines and AI are useful discovery tools, but they do not replace controlled corpora, dictionaries, primary authorities or source-level verification.
• When references disagree, compare scope, date, region, register, evidence base and whether the source is describing usage or recommending a convention.
• The final decision belongs to the real context: meaning, audience, register, institution and purpose.

The goal is not reference dependence. The goal is to verify efficiently, make a proportionate decision, use the language, and eventually perform without needing to check every familiar pattern again.

Dictionary literacy: read entries as language systems, not definition vending machines

A dictionary entry is a compact model of how a lexical item behaves. Strong users do not merely ask, “What does this word mean?” They ask which sense is active, how the word combines, what grammar it requires, what register it belongs to, how it is pronounced, what region uses it, which related forms exist and what examples reveal about context.

Headwords and lemmas

Dictionaries normally organise inflected forms under a headword or lemma. A learner searching for went is directed conceptually to go; children belongs under child; better may be connected to good or well depending on sense.

Understanding lemma structure matters when using corpora too, because a corpus search may distinguish one surface form from all forms of the same lemma.

Word class

The same spelling can operate as more than one part of speech. Record can be a noun or verb, with different stress in many standard pronunciations. Light can be a noun, adjective or verb. Google began as a proper name and also became a verb in widespread usage.

Before copying a definition, identify which grammatical role the word has in the sentence.

Countability

English nouns may be count, uncount or both depending on sense. Experience can refer to accumulated knowledge as an uncount noun or to individual events as countable experiences. Paper can be material or a document. A learner who ignores countability can know the meaning and still produce unidiomatic grammar.

Learner dictionaries often mark this information explicitly.

Transitivity

Verb entries can indicate whether a verb takes an object and what patterns follow it. Discuss usually takes a direct object in Standard English: discuss the problem, not *discuss about the problem* in the same structure. Depend commonly selects on. These patterns are lexical-grammatical knowledge, not optional style.

Complementation

Verbs, adjectives and nouns can select different complement patterns:

• decide to leave;
• avoid leaving;
• interested in learning;
• the possibility of failure.

A dictionary with grammar coding can save a learner from attempting to generalise one pattern across near-synonyms.

Sense numbering

Dictionaries separate senses because one word can serve several related or unrelated meanings. Sense ordering may reflect frequency, historical development or editorial design depending on the dictionary.

Do not assume sense 1 is always the historically oldest or universally most common. Read the dictionary’s method where precision matters.

Sub-senses

Some entries contain fine-grained sub-senses. Whether two uses deserve separate senses is partly an editorial analysis of meaning. Lexicography is not merely copying facts from language; it is classifying patterns in evidence.

Polysemy and homonymy

Polysemy refers to related senses of one lexical item. Homonymy refers to forms that look or sound the same but have distinct historical or semantic identities. Dictionaries may structure entries differently depending on that analysis.

Learners do not need to solve every theoretical boundary. They do need to recognise that one spelling can lead to very different meanings.

Definitions

A definition is a compressed paraphrase, not the word itself. It describes enough semantic information to distinguish the sense but may not encode every collocational, pragmatic or cultural feature.

That is why memorising definitions rarely produces mastery.

Defining vocabulary

Some learner dictionaries use a restricted defining vocabulary so explanations remain accessible. This can make an advanced word understandable without forcing the learner to chase several new words inside the definition.

Examples

Good examples show behaviour. Read them vertically across an entry and ask what repeats. Does the adjective occur before nouns or after linking verbs? Which objects follow the verb? Which prepositions recur? Is the word usually modified?

Examples become a miniature corpus.

Labels for register and attitude

Usage labels such as formal, informal, literary, humorous, offensive, approving or disapproving tell the learner that lexical meaning includes social positioning.

Oxford Learner’s Dictionaries, for example, explicitly documents labels for situational formality and speaker attitude. A learner who ignores those labels can produce semantically accurate but socially inappropriate language.

Regional labels

Labels such as British English, North American English, Indian English or other regional identifiers mark distribution. A regional label is not a warning that the word is illegitimate. It tells the reader where the form is especially associated.

This links directly to the World Englishes & Variation owner.

Dated, old-fashioned and historical

Words can remain understandable after their productive use declines. Dictionaries may label forms as dated, old-fashioned, archaic or historical. Those labels differ in nuance and dictionary policy.

For literature and historical texts, a historical or unabridged dictionary may provide more context than a learner dictionary.

Offensive and sensitive language labels

Some entries contain warnings about offensive, derogatory or sensitive usage. These labels need careful reading because social evaluation changes over time and across communities.

A learner should understand both the denotational meaning and why use can cause harm or signal prejudice.

Pronunciation notation

Many learner dictionaries use the International Phonetic Alphabet or a related phonemic system. Learning a small set of IPA symbols can make dictionary pronunciation information portable across words.

Pay particular attention to:

• primary stress;
• secondary stress where shown;
• weak vowels;
• variant pronunciations;
• British/US or other reference differences.

Audio

Audio models are useful but should not replace notation or listening comparison. One recording is one performance by one speaker. It does not define every legitimate pronunciation.

Word families

Dictionaries often cross-reference related forms: analyse, analysis, analytical, analytically. Word-family study should include changes in stress, grammar and collocation.

A noun and adjective from the same root do not automatically combine with the same neighbouring words.

Derivatives

Suffixes can change word class and meaning. But learners should not invent unattested forms simply because morphology seems possible. Dictionaries and corpora verify whether the derived word is established.

Compounds

Dictionaries differ in whether compounds are separate headwords, subentries or searchable phrases. Spelling can vary between open, hyphenated and closed forms and may change historically.

Phrasal verbs

Phrasal verbs often deserve separate treatment because particle choice can create meanings not predictable from the base verb alone. Take off, take over, take up and take in illustrate why word-by-word interpretation is insufficient.

Idioms

Idioms may be stored under a key lexical item, and dictionary interfaces differ. Search the full phrase when possible. Read labels because idioms can be informal, regional, humorous or dated.

Collocations dictionaries

A collocations dictionary answers productive questions that ordinary definitions do not. Oxford describes its collocations dictionary as intended especially to support writing, grouping combinations by grammatical category and meaning.

This is useful for advanced learners because the problem is often not “What does evidence mean?” but “Do English writers say strong evidence, heavy evidence, powerful evidence or something else?”

Thesauruses

A thesaurus groups words by similarity, but synonyms are almost never fully interchangeable. Before substituting a thesaurus suggestion, check:

• sense;
• register;
• collocation;
• grammar;
• connotation;
• strength;
• regional distribution.

A thesaurus generates candidates. It does not approve the final sentence.

Etymological dictionaries

Etymology traces word origins and historical development. It is useful for language history and sometimes memory, but origin does not determine modern meaning.

A common reasoning error—the etymological fallacy—assumes that a word’s earliest meaning is its true meaning. Modern usage is determined by current language evidence.

Historical dictionaries

Historical dictionaries such as the Oxford English Dictionary organise evidence diachronically, using dated quotations to document senses and forms across time. They are essential for questions such as “Was this sense used in the nineteenth century?” or “When does the record first attest this form?”

Even then, first attested does not necessarily mean first ever spoken. Archives are incomplete.

Specialised terminology databases

Technical fields may maintain official term banks, glossaries or standards. If a term has a regulated definition, use the authoritative domain source rather than general web frequency.

Style dictionaries and usage manuals

Some reference works focus on disputed usage, editorial preferences or style. They can be helpful when the target is edited prose, but their recommendations should be identified as recommendations rather than universal grammar unless supported by broader linguistic evidence.

Dictionary comparison is a literacy skill

When a word is complex, compare entries across two strong dictionaries. Differences can reveal editorial choices:

• one splits a sense that another combines;
• one marks a phrase formal;
• another gives more regional evidence;
• examples foreground different collocations;
• pronunciation variants differ.

The goal is not to declare one dictionary wrong. It is to understand the evidence model behind the entry.

A dictionary disagreement protocol

1. Check that both entries refer to the same sense.
2. Read labels and examples.
3. Check publication/update date where relevant.
4. Consult corpus evidence.
5. Check whether the disagreement concerns description or recommendation.
6. Choose according to the target context.

What not to ask a dictionary to decide

A dictionary cannot decide whether your essay argument is strong, whether a statistic proves causation, whether a school rubric awards the mark, whether a sentence is persuasive to a specific audience or whether a usage is ethically appropriate in every community.

Reference literacy includes knowing when to stop looking at the dictionary.

Corpus literacy: turn usage questions into evidence questions

A corpus is useful because language memory is selective. People remember striking phrases, recent encounters and forms used by their own communities. A corpus gives the learner access to a much larger sample.

But a corpus does not answer questions automatically. The user must define the query, understand the dataset and interpret the results.

Corpus design comes before corpus size

A larger corpus is not automatically better. A billion words of contemporary American media cannot answer a question about seventeenth-century English better than a much smaller historical corpus. A corpus of academic writing cannot describe casual family conversation.

Before searching, identify the population of language you need.

Balanced corpora

A balanced corpus deliberately includes several text types so no single genre dominates the whole picture. Balance can support broad language description, but the design always reflects choices about what counts as an important category.

Monitor corpora

Some corpora are designed to track language over time with continuing additions or comparable slices. These are useful for change questions.

Specialised corpora

A specialised corpus may contain one discipline, profession, learner group, genre or community. Smaller specialised corpora can answer precise questions more directly than general corpora.

Learner corpora

Learner corpora collect language produced by learners. They can reveal recurring patterns, overuse, underuse and error types. They should not be used as a direct model of target usage without comparison.

Parallel corpora

Parallel corpora align texts with translations. They support translation studies and multilingual research, but translation language can have patterns different from originally composed texts.

Comparable corpora

Comparable corpora contain similar kinds of texts in different languages or varieties without being direct translations. They support cross-linguistic and cross-variety comparison.

Historical corpora

Historical corpora make change measurable by providing dated texts. Their limitations include uneven survival, changing genres, spelling variation and uncertain authorship or metadata.

Spoken corpora

Spoken corpora capture features often underrepresented in writing: hesitation, repair, discourse markers, contractions, turn-taking and spoken grammar.

Transcription choices matter. A transcript can omit prosody or normalise pronunciation.

Corpus metadata

Metadata describes texts or speakers: date, genre, region, age, gender, source, publication or other attributes. Good metadata lets the researcher ask more precise questions.

Missing or unreliable metadata reduces what the corpus can support.

Search terms

Exact-string search

Search the precise phrase when you want that exact sequence. This is simple but can miss inflected variants or spelling alternatives.

Lemma search

A lemma search groups forms such as write, writes, wrote, written, writing where the corpus supports morphological annotation.

Part-of-speech search

Tagged corpora can distinguish record as noun from record as verb or search grammatical sequences such as adjective + noun.

Wildcard search

Wildcards can find families of forms, but broad queries can generate noise. Define the pattern narrowly enough that results remain interpretable.

Phrase-frame search

Variable phrase frames such as the * of or it is * that can reveal productive patterns when used carefully.

Regular expressions

Advanced corpus tools may support regular expressions for complex searches. This is powerful but can create false precision if the user does not understand what the pattern matches.

Concordance reading

The concordance is one of the most valuable learning interfaces because it turns many examples into a visual pattern. Read down the centre and immediate neighbours before reading every full sentence.

Suppose the target is evidence. A learner might notice recurring combinations such as strong evidence, evidence suggests, evidence of, evidence for, provide evidence, lack of evidence.

Those patterns reveal grammar and discourse at once.

Sort the concordance

Some tools can sort by the first word to the right or left. Sorting makes collocational patterns visible.

Expand context

KWIC lines can be too short to identify sense. Expand selected lines after noticing a pattern.

Sample if results are huge

Reading the first fifty results can be biased if ordering is not random. Use random samples or representative slices when the tool supports them.

Frequency is not one number

Raw count

The number of occurrences inside the corpus.

Relative frequency

The count adjusted to corpus size, often per million words.

Document frequency

How many different texts contain the form. A phrase occurring 500 times in one document is different from a phrase appearing across 400 independent texts.

Dispersion

Dispersion describes how evenly occurrences are spread. A high-frequency word concentrated in one subgenre may not be generally common.

Range

Range asks across how many categories or registers a word appears. Broad range often matters for general vocabulary priorities.

Frequency bands

Learner resources sometimes organise vocabulary by broad frequency. These are useful for prioritisation, not precise predictions of what every learner will encounter.

Association and collocation

Raw co-occurrence can be dominated by common words. Corpus tools use association measures to identify combinations that occur more strongly than expected.

Different measures emphasise different patterns. Some reward strong exclusive associations; others favour frequent combinations. The learner should not treat one score as a universal collocation truth.

MI score

Mutual Information often highlights combinations that are statistically distinctive, including some relatively rare pairs.

T-score

T-score tends to favour frequent reliable co-occurrence and can surface common phraseology.

LogDice and other measures

Modern corpus interfaces may use measures designed to compare association across corpora more conveniently. The exact statistic matters less for most learners than understanding that “top collocate” depends on the measure.

Collocation window

Corpus tools define how many words to the left and right count as nearby. A five-word window answers a different question from immediate adjacency.

Grammar often needs narrow windows; semantic association may use wider windows.

Grammatical relations

Advanced corpus tools can search subject–verb, verb–object or modifier relationships rather than simple linear distance. This can distinguish evidence supports claim from unrelated words merely appearing near one another.

Semantic preference

A word may repeatedly collocate with a semantic class. For example, a verb may often take nouns related to problems or benefits. This recurrent meaning environment is called semantic preference.

Semantic prosody

Some words acquire positive or negative evaluative associations through typical co-text. Advanced vocabulary learners can miss this if they study definitions without usage patterns.

SETC’s advanced vocabulary lessons already develop semantic prosody; corpus literacy supplies the evidence method.

Register comparison

Many usage disputes disappear when register is separated. A form can be common in conversation and rare in academic prose.

Compare frequencies across:

• spoken conversation;
• fiction;
• news;
• academic writing;
• web or social media where available.

Do not average those categories if your real question concerns one.

Academic phraseology

A phrase that occurs often in academic corpora can be more useful for academic writing than a globally high-frequency conversational phrase.

This connects to Academic English.

Regional comparison

Comparable corpora can test whether a form differs across British, American, Singaporean or other Englishes.

Regional frequency should be interpreted with register. If one national corpus contains more journalism and another more conversation, apparent national differences can actually be genre differences.

This connects to World Englishes & Variation.

Historical comparison

Corpus evidence across time can show frequency change, new collocations, semantic change and grammatical competition.

Spelling variation can hide older evidence. Historical searches often need alternative forms.

This connects to History of English & Language Change.

Keyness

Keyness identifies words unusually frequent in one corpus relative to a reference corpus. It can reveal what distinguishes a genre, author, topic or community.

Keyness does not mean conceptual importance automatically. A procedural word may be statistically distinctive because of genre.

Keywords in context vs search-engine keywords

Corpus keywords are statistical features of a dataset. SEO keywords are queries or phrases associated with search behaviour. The two concepts should not be confused.

Clusters and lexical bundles

Recurring multiword sequences such as on the basis of or it should be noted that can reveal discourse patterns. Some are highly register-specific.

Learners should understand function rather than copy bundles mechanically.

Corpus-driven vs corpus-based work

Researchers sometimes distinguish using a corpus to test or illustrate an existing category from allowing recurrent corpus patterns to suggest categories. Both approaches can be legitimate.

For learners, the practical version is:

• sometimes search a known question;
• sometimes look at examples and discover a pattern you had not predicted.

Corpus annotation

Some corpora add linguistic information automatically or manually: parts of speech, lemmas, syntax, semantic categories or speaker metadata.

Annotation can contain errors. Automated tags are useful at scale but should not be treated as infallible.

Corpus cleaning

Web corpora can contain duplicated text, boilerplate, malformed pages, machine-generated material or spam. Corpus builders use cleaning procedures, but users should still understand the source.

Search-result bias

Google hit counts are not corpus frequency. Search engines rank, deduplicate, personalise and continually change indexes. Exact counts can be unstable and include duplicated or machine-generated pages.

Search engines are excellent for discovery; they are weak substitutes for controlled corpus evidence.

Google Ngram and book corpora

Large digitised book datasets can reveal broad historical trends, but they represent published books, not the full language. OCR errors, metadata and changing publication composition matter.

Use them for exploratory historical evidence and verify important claims with stronger sources.

One corpus example does not prove a rule

An attested sentence proves that someone used a form in the corpus. It does not prove the form is common, standard, formal or appropriate.

Evidence interpretation needs scale.

Zero results do not prove impossibility

A phrase may be absent because the corpus is too small, the spelling differs, the genre is missing or the query is wrong. Absence is evidence only relative to corpus coverage.

Rare does not mean wrong

Creative writing, specialist terminology and new coinages can be rare and fully legitimate.

Common does not mean standard

Frequent non-standard forms can be common in speech while formal writing maintains another convention.

Corpus workflow for a usage question

1. Define the target form.
2. Choose the corpus matching region, period and register.
3. Search exact and relevant variants.
4. Inspect concordance lines for sense.
5. Compare frequency and dispersion.
6. Inspect collocates or grammatical patterns.
7. Compare another register or variety where useful.
8. Check dictionary labels and grammar information.
9. Return to the real sentence and choose according to purpose.

This is evidence-guided judgement, not corpus obedience.

Reference authority: choose the source that owns the question

Reference literacy depends on matching the question to the authority. A school syllabus is authoritative for what the syllabus requires, not for the full grammar of English. A dictionary is authoritative evidence about lexicographic analysis and documented usage, not about how one university wants citations formatted. A corpus provides evidence of attested use, not an automatic verdict on what an examination will accept.

Strong users ask: Who has the right kind of authority for this claim?

Primary authority

Where an institution controls the rule, use the institution itself. Examples include:

• an examination board for current paper format and assessment requirements;
• a university for its academic-integrity or referencing policy;
• a publisher for its house style;
• a standards organisation for an official technical definition;
• a software provider for current product behaviour;
• a government body for official terminology or regulation.

Secondary explanations can help interpret the rule, but they should not replace the controlling source.

Linguistic authority

Language is not governed by one global academy. Evidence comes from several sources:

• dictionaries;
• grammars;
• corpora;
• linguistic research;
• style guides;
• observed community usage.

Their authority depends on the question.

Descriptive authority

Descriptive sources answer what speakers actually do, which forms exist, where they occur and how they vary. Corpora and linguistic studies are especially important.

Prescriptive authority

Prescriptive sources recommend a target. A style guide may prefer one capitalisation or punctuation pattern. An examination may require Standard English. These recommendations can be legitimate without becoming universal descriptions of all English.

Pedagogical authority

Learner dictionaries and teaching grammars simplify complex language knowledge to make it learnable. Their explanations can be excellent for students even when a research grammar would describe the phenomenon in more technical detail.

Historical authority

Historical dictionaries, dated corpora and manuscript/text scholarship are required when the question is about earlier use. A current dictionary cannot always resolve whether a sense existed in 1750.

Domain authority

Technical fields define their own terminology. Significant in statistical reporting, consideration in contract law or stress in phonetics may require domain-specific references.

Style guides: rules for a document community

Style guides standardise editorial choices. They may govern spelling, punctuation, capitalisation, numbers, dates, headings, abbreviations, citations and tone.

A style guide answers: How should documents in this system be edited? It does not necessarily answer: What forms exist in English?

House style

Organisations often maintain internal conventions. Two newspapers can make different punctuation choices while both use grammatical Standard English.

Academic citation styles

APA, MLA, Chicago, Harvard-style systems, Vancouver and discipline-specific formats organise citations differently. The differences are editorial conventions designed for different scholarly communities.

Do not turn a citation-format preference into an English grammar rule.

Consistency vs correctness

Many editorial choices are matters of consistency. If both organise and organize are allowed in a target standard but the publication chooses one, the editing goal is consistency with house style.

Examination specifications

Students often search the internet for “what examiners want”. The most reliable starting point is the official current specification, syllabus, rubric or examiner documentation where available.

Blogs and tutor advice can add strategy, but current formal requirements belong to the board.

Specifications age

Examination formats change. A page written five years ago can be accurate historically and wrong now. Check dates.

Rubric interpretation

A rubric is not a dictionary of English. It tells you what the assessment values. If it uses terms such as coherence, development or control, pedagogical guidance can unpack them, but the official rubric remains the scoring authority.

Usage disputes

Some language questions attract strong opinions:

• singular they;
• split infinitives;
• sentence-final prepositions;
• less vs fewer;
• who vs whom;
• literally as intensifier;
• new verb forms;
• changing punctuation conventions.

Reference literacy prevents the debate from collapsing into “my teacher said”.

A usage-dispute protocol

1. Define the exact construction.
2. Check current dictionaries and grammars.
3. Check corpus distribution by register and region.
4. Check historical evidence if claims about tradition appear.
5. Check style guides for target-context preferences.
6. Separate grammaticality, frequency, prestige and editorial recommendation.
7. Choose the form appropriate to the audience.

Singular they as a model question

If someone claims singular they is a recent invention, historical evidence is relevant. If the question is whether a particular journal accepts it, the journal style policy matters. If the question is how common it is in conversation, a contemporary spoken corpus helps. One source cannot answer all three questions.

Split infinitives

The construction is grammatically possible in modern English and has a history of prescriptive controversy. A style guide may still prefer alternatives in some contexts. The user should distinguish grammatical structure from editorial tradition.

Sentence-final prepositions

Again, the existence of the construction and its appropriateness in a particular formal style are separate questions.

Grammar references

Comprehensive grammars describe language systems at greater depth than dictionaries. They are useful for questions involving clause structure, complementation, tense/aspect, modality, information structure and variation.

Learners may need pedagogical grammars first, then advanced grammars for difficult questions.

Grammar example problem

Suppose a learner asks whether I suggested him to leave is acceptable in Standard English. A general dictionary definition of suggest will not settle the complementation pattern unless the entry contains grammar coding. A learner grammar or dictionary examples can show patterns such as suggest doing, suggest that… or suggest something to someone.

The reference must expose grammar.

Pronunciation references

Pronunciation questions can involve:

• phonemes;
• stress;
• variant pronunciations;
• regional differences;
• proper names;
• technical terms.

A learner dictionary may show one or two reference accents. For a global variation question, broader evidence is required.

Names and proper nouns

People have authority over how they want their own names pronounced where that information is available. Place-name pronunciations may be locally conventional and resist spelling-based guesses.

Reference literacy includes recognising when the person or community is the best source.

Wikipedia and encyclopedias

General reference works can be excellent orientation tools. Their value lies in mapping concepts, terminology and sources. For high-stakes claims, follow references to primary or specialist material.

Do not reject a source merely because it is tertiary; use it for the right job.

Blogs and forums

Community discussions can reveal real usage questions, attitudes and niche expertise. They can also repeat myths. Treat them as leads or evidence of what people say about language, not automatic linguistic authority.

Search engines

A search engine answers “What pages match this query under the engine’s ranking system?” It does not answer directly:

• How frequent is this phrase in English?
• Is it standard?
• Is it formal?
• Is the source reliable?
• Which variety uses it?

Use search to find references, not to replace them.

Quoted search

Searching an exact phrase in quotation marks can reveal examples, but result counts are unstable and pages are not a controlled sample.

Query wording

Search engines respond to the words you choose. A biased query can return a biased evidence set. Compare alternative formulations.

AI systems

AI can explain dictionary entries, generate examples and propose distinctions. It can also invent citations, overstate rules, mix national conventions and produce plausible sentences that are rare or unnatural.

Treat AI as a hypothesis generator and language practice partner unless the answer is independently verified.

AI verification workflow

If AI says, “This phrase is formal and mostly British”:

1. check a current dictionary for labels;
2. check British and other corpus evidence;
3. inspect authentic examples;
4. check whether the relevant sense matches;
5. then decide.

The AI answer becomes a query, not the conclusion.

Conflicting reference sources

References disagree for several reasons:

• different populations;
• different publication dates;
• different editorial policies;
• different national standards;
• different sense analysis;
• description vs prescription;
• different corpus evidence.

Conflict-resolution protocol

1. Restate your real question.
2. Identify what each source is designed to answer.
3. Check dates and scope.
4. Compare examples.
5. Add corpus evidence.
6. Check institutional requirements.
7. Choose the answer matching the target context.
8. Preserve uncertainty if evidence remains mixed.

Authority is not infallibility

Dictionaries, corpora and style guides are created by humans and systems. They can lag behind change, contain annotation errors, differ theoretically or make editorial compromises.

Reference literacy does not mean replacing personal judgement with reference obedience. It means improving judgement through evidence.

Practical reference workflows: investigate the exact question

Reference literacy becomes valuable when learners can move from uncertainty to a defensible decision. The workflows below use several tools in sequence because most advanced language questions have more than one layer.

Workflow 1: What does this word mean here?

Suppose the sentence contains critical.

The learner should not copy the first definition. Critical can mean involving careful evaluation, expressing disapproval, extremely important, or relating to a crisis point, among other senses.

1. Read the surrounding sentence and paragraph.
2. Identify the word class.
3. Open a strong learner or general dictionary.
4. Compare the numbered senses.
5. Check examples for grammar and domain.
6. Select the sense that fits the context.
7. If the text is specialised, check a domain reference.
8. Paraphrase the sentence using the selected sense.

The final step tests whether the learner actually understood the entry.

Workflow 2: Can I use this synonym?

A thesaurus suggests important, significant, crucial, substantial and consequential.

Do not treat these as interchangeable.

1. Define the exact sense needed.
2. Check each candidate in a dictionary.
3. Read register labels.
4. Check collocations in a corpus or collocations dictionary.
5. Test the candidates inside the real sentence.
6. Choose the word whose meaning and phrase behaviour match.

For example, statistically significant is a conventional technical phrase. Replacing it with statistically crucial is not a stylistic upgrade.

Workflow 3: Which preposition follows this word?

Suppose the learner writes responsible of when they mean having responsibility.

1. Check the learner dictionary entry for responsible.
2. Inspect example patterns.
3. Search responsible for and alternatives in a corpus.
4. Check whether another preposition creates a different meaning.
5. Write three original examples.

The learner should record the phrase pattern, not the adjective in isolation.

Workflow 4: Which verb collocates with this noun?

Suppose the learner wants to write about a decision.

Search collocations. Likely productive patterns include make a decision, reach a decision, reverse a decision, announce a decision, justify a decision.

The learner then chooses based on intended meaning. A collocations dictionary supplies options; context selects among them.

Workflow 5: Is this phrase formal enough for academic writing?

Consider a phrase such as a lot of.

The question is not whether a lot of is English—it clearly is. The question is distribution and academic register.

1. Check dictionary labels.
2. Compare spoken and academic corpus frequencies.
3. Inspect academic alternatives such as many, much, a substantial number of or a more precise quantity.
4. Choose the alternative that matches the claim rather than simply sounding formal.

Sometimes a lot of may be acceptable in less formal academic communication. Corpus evidence helps the learner see the register pattern instead of memorising a prohibition.

Workflow 6: Is this British, American or global?

Take at the weekend and on the weekend.

1. Check regional labels in dictionaries.
2. Compare British and American corpora.
3. Inspect current examples.
4. Check the target institution’s convention if consistency matters.

The evidence may show distributional preference rather than absolute exclusivity.

Workflow 7: Is this phrase common in Singapore English?

Do not use Google alone.

1. Check whether a Singapore dictionary or linguistic source documents it.
2. Search a relevant Singapore English corpus where available.
3. Check spoken vs written use.
4. Identify age, register and social context.
5. Compare with other Englishes.

This prevents local stereotypes from becoming fake rules.

Workflow 8: Was this word used in Shakespeare’s time?

1. Check a historical dictionary.
2. Find dated citations around the period.
3. Check Early Modern English corpora where possible.
4. Distinguish the word form from the specific modern sense.
5. Do not say “Shakespeare invented it” merely because the earliest famous quotation is Shakespearean.

The History of English & Language Change owner explains why first surviving evidence and first actual use are different claims.

Workflow 9: Is this grammar pattern changing?

Suppose two forms compete.

1. Search both forms in contemporary corpora.
2. Compare by date if the corpus allows it.
3. Compare spoken and written registers.
4. Compare age or region where metadata exists.
5. Check recent dictionaries/grammars.
6. Review historical evidence.
7. Describe the pattern proportionately: “increasing”, “common in conversation”, “still dispreferred in formal edited prose”, etc.

Language change is rarely captured by one yes/no answer.

Workflow 10: Can I trust this AI language correction?

AI changes different than to different from and says the original is incorrect.

Do not accept the rule automatically.

1. Check current dictionaries/usage references.
2. Compare corpora by region and register.
3. Check the target style guide.
4. Decide whether the issue is grammar, variation or editorial preference.

The tool may have converted a variation question into a false binary.

Workflow 11: Which word is more common?

“More common” must specify where.

Compare:

• overall corpus frequency;
• spoken frequency;
• academic frequency;
• regional frequency;
• historical trend;
• dispersion across texts.

A word can be globally less frequent but more useful in the target domain.

Workflow 12: Is this collocation strong?

Search the node word and collocate. Inspect both frequency and association. Then read examples. A high association score from three occurrences is not the same learning priority as a frequent productive phrase across hundreds of texts.

Workflow 13: Is this usage offensive?

Check current dictionary labels, recent usage guidance, community context and reliable social-language evidence. Historical dictionaries can explain earlier meanings but should not override current harm or context.

If the term refers to a community, current community preference may be especially important.

Workflow 14: Which pronunciation is correct?

Replace the question with:

• Which reference accent?
• Are several variants recorded?
• Which pronunciation is expected in my context?
• Will listeners understand me?

Check pronunciation dictionaries and audio, then World Englishes evidence where variation matters.

Workflow 15: Which spelling should I use?

Check national standard and house style. Recognise major alternatives receptively. Edit for consistency.

Workflow 16: Is this word C1 or C2?

Level labels in learner resources can help prioritise but are not intrinsic properties of words. Difficulty depends on sense, phrase, learner language background, frequency and task.

Use CEFR-related resources as orientation, then teach actual usage.

Workflow 17: What does this technical term mean?

Start with the specialist source, not a general dictionary. Then use a general dictionary if you need the everyday contrast.

For academic writing, record the operational definition used by the source because different fields can use the same term differently.

Workflow 18: Does this phrase exist?

Existence is a low bar.

Search a corpus. One example proves attestation in that corpus, not naturalness across English. Inspect frequency, dispersion, register and authorship.

Workflow 19: Does this sentence sound natural?

Naturalness combines several systems:

• grammar;
• collocation;
• information structure;
• register;
• idiom;
• context.

Compare with authentic examples that express the same relationship. Do not simply search the entire sentence.

Workflow 20: Which source should settle the question?

Sometimes no single source should.

A difficult usage decision may require:

dictionary + corpus + grammar + style guide + target-context evidence.

Reference literacy is the ability to coordinate those sources without confusing their roles.

Worked investigation: “strongly recommend”

A learner wants an adverb meaning “very much” before recommend.

A dictionary defines recommend but may not foreground adverb choice. A corpus query shows frequent patterns. Strongly recommend appears repeatedly in reviews, advice and formal recommendations. The learner checks examples and sees that highly recommend also appears, especially in reviews and endorsements.

Now meaning matters. Strongly recommend can sound forceful and institutional. Highly recommend often expresses very positive evaluation. Corpus contexts help differentiate.

The learner records both as phrases, not one “correct answer”.

Worked investigation: “researches”

A learner writes, “Many researches show…”

The dictionary marks research commonly uncountable when referring broadly to systematic investigation. It may also document countable uses in some contexts or varieties. Academic corpora show common alternatives:

• much research;
• many studies;
• several research projects.

The repair depends on meaning. If the learner means published studies, many studies is usually transparent.

Worked investigation: “comprise of”

Usage sources may disagree in recommendation. Corpus evidence shows attestation; traditional edited style often prefers comprise without of or be composed of. A strict style guide may reject comprised of, while contemporary usage dictionaries may describe its spread.

The correct conclusion for a formal assessment can be: use the convention expected by the target while recognising the broader usage evidence.

Worked investigation: “data is” vs “data are”

The answer varies by discipline, editorial style and whether data is conceptualised as a mass aggregate or plural of datum. Scientific style guides can preserve plural agreement more strongly than general usage.

Dictionary + corpus + disciplinary style evidence is the correct combination.

Worked investigation: “literally”

Historical and current dictionaries show that intensifying uses of literally have a longer history than internet debate suggests. Corpus evidence can show current distribution. A style guide can still recommend avoiding ambiguity in formal prose.

The interesting question is not whether a social-media complaint is loud. It is what the evidence says about sense, history and register.

Worked investigation: “impact” as a verb

A learner hears that impact should never be a verb. Dictionaries and corpora document verbal use. A style editor may still prefer affect for clarity in some prose.

This is a perfect reference-literacy example: existence, frequency and stylistic preference are different claims.

Worked investigation: “irregardless”

Dictionaries may include a word because it is attested, while labelling it nonstandard or otherwise marked. Inclusion does not equal recommendation for formal Standard English.

This illustrates why “It’s in the dictionary” is not the end of a usage discussion.

Worked investigation: “they” for one person

Historical dictionaries show long use of singular they; current corpora show extensive contemporary usage; style guides and institutions now vary in details of recommendation. A learner can use the sources to separate historical fact from current editorial policy.

Worked investigation: one word in three registers

Take a common word such as kids. Dictionaries may mark it informal. Conversation corpora show high natural use. Academic corpora may prefer children or more specific demographic terms.

The word is not wrong. Register determines fit.

Worked investigation: a Singapore English term

Take a local term such as void deck. A Singapore reader may need no explanation; an international reader may. A local corpus or Singapore-focused dictionary can confirm usage. A general international dictionary may omit or label the term.

Omission from the general dictionary does not erase local legitimacy.

Worked investigation: a new technology term

When a new AI term appears, general dictionaries may lag. Use current technical documentation, specialist reporting and corpus/web evidence. Over time, lexicographic sources may catch up.

This is why reference hierarchy changes with lexical age.

Teaching reference literacy: move from answer-giving to evidence-seeking

Many learners become dependent on teachers because every uncertainty is answered externally. “Is this word okay?” “Can I say this?” “Which one is correct?” The teacher can answer quickly, but the learner does not acquire a verification process.

Reference literacy turns the question back into an investigation.

The tutor’s job becomes:

• identify the type of uncertainty;
• choose an appropriate reference tool;
• model how to read the evidence;
• make the learner decide;
• test the decision in new language;
• fade the support.

The first-referral rule

When a learner asks a question that a reference can answer, do not automatically answer from memory. Ask which tool would settle it.

Examples:

• meaning → dictionary;
• collocation → corpus/collocations dictionary;
• regional difference → dictionaries + comparable corpora;
• exam requirement → official specification;
• historical sense → historical dictionary/corpus;
• style preference → style guide;
• technical definition → domain authority.

This habit creates metalinguistic independence.

Do not force young learners into advanced tools too early

A Primary student may not need concordance statistics. They can still learn reference literacy through simpler habits:

• look beyond the first dictionary meaning;
• check the example sentence;
• notice part of speech;
• compare two words;
• check pronunciation audio;
• write a fresh sentence;
• ask whether the sentence matches the example pattern.

The concept scales with age.

Reference literacy progression

Stage 1: locate

The learner can find a word or phrase in an appropriate dictionary.

Stage 2: select

The learner can choose the relevant sense and identify basic grammar/labels.

Stage 3: compare

The learner compares near-synonyms, regional variants or phrase patterns.

Stage 4: verify

The learner checks authentic examples or corpus evidence.

Stage 5: interpret

The learner understands frequency, register and corpus limitations.

Stage 6: resolve conflict

The learner coordinates dictionary, corpus and style evidence.

Stage 7: transfer

The learner applies the same verification process to an unfamiliar language question.

Stage 8: critique the reference

The learner can explain what the chosen source cannot prove.

Diagnostic matrix

Three-student reference-literacy tutorial

A three-student tutorial makes evidence comparison visible.

Iona is the dictionary reader. She identifies sense, grammar, labels and examples.

Maren is the corpus reader. She searches authentic patterns, compares frequency and register, and checks collocation.

Leonie is the decision controller. She asks what the real task requires and whether the evidence is sufficient.

Then roles rotate.

Example question: “Can I say very essential?”

Iona checks dictionaries: essential can mean completely necessary, which is already absolute in many contexts.

Maren searches corpora and finds actual examples of very essential, showing that speakers do use it.

Leonie asks the target: formal academic writing. She may choose simply essential, or a gradable alternative such as very important, depending on intended meaning.

The group learns that attestation does not automatically settle style.

Reference notebook

A strong learner can maintain a reference notebook organised by question type instead of alphabetically by word.

Possible fields:

• uncertainty;
• sentence/context;
• dictionary evidence;
• corpus evidence;
• register/region;
• decision;
• new example;
• future check.

This records the reasoning process.

Do not record everything

Keep entries for recurring problems, high-value words and useful patterns. Reference literacy should reduce future lookup cost, not create another giant database the learner never revisits.

Spaced reference retrieval

After solving a problem, remove the reference and test later.

If the learner always needs the dictionary for responsible for, the lookup has not yet become productive knowledge.

Retrieve:

• meaning;
• phrase;
• register;
• one original sentence.

Reference literacy and vocabulary learning

Every vocabulary item can be expanded through a “word profile”:

• core meaning;
• other important senses;
• pronunciation;
• grammar;
• word family;
• collocations;
• register;
• regional notes;
• example;
• contrast word.

This profile is richer than definition memorisation but should be scaled to the learner’s needs.

Reference literacy and writing

Writers need tools at different revision passes.

Meaning pass

Dictionary and technical references.

Phrase pass

Collocations dictionary and corpus.

Register pass

Dictionary labels and register corpus.

Consistency pass

Style guide.

Historical/literary pass

Historical dictionary and textual notes.

Citation pass

Institutional citation/style rules.

The tool should follow the revision problem.

Reference literacy and reading

Excessive lookup destroys reading flow. Learners need a threshold.

When a word is unfamiliar, ask:

• Can I infer enough to continue?
• Is the word central to the argument?
• Does it repeat?
• Could the wrong interpretation damage the passage model?

Look up high-value uncertainty, not every unknown word.

Delayed lookup

Mark a word and continue, then verify after the paragraph. This preserves global meaning while still correcting inference.

Reference literacy and listening

When spoken language contains an unfamiliar term, the listener may not be able to consult a dictionary immediately. Record an approximate form, context and meaning guess. Verify later through transcript, audio search or dictionary pronunciation.

This trains the ability to recover a word from partial acoustic evidence.

Reference literacy and speaking

Speakers cannot consult references mid-sentence. Productive control therefore requires prior phrase learning. Use dictionary/corpus work during preparation, then rehearse without the tools.

Reference literacy and oral examinations

Do not introduce new advanced vocabulary hours before an oral examination merely because a dictionary makes it look impressive. Reliable retrieval and natural collocation matter more than novelty.

Reference literacy and editing

Editors should distinguish:

• objective error;
• valid variant;
• house-style choice;
• register mismatch;
• unclear but grammatical wording;
• unusual but purposeful style.

Reference evidence reduces overcorrection.

Overcorrection

Writers sometimes replace valid forms after grammar-checker warnings. Reference literacy asks whether the warning reflects:

• real grammar;
• a statistical style preference;
• one national standard;
• software limitation;
• sentence ambiguity.

Do not accept corrections without understanding them.

Reference literacy and teacher feedback

Teachers can cite the type of evidence:

“Formal Standard English here uses this complementation pattern; see the learner-dictionary examples.”

“Both forms exist, but the exam style requires this spelling.”

“This collocation is possible but rare in academic writing; compare these corpus examples.”

This makes correction teachable.

When the teacher and dictionary disagree

Do not turn the question into a contest. Identify what each is answering. The teacher may be enforcing curriculum convention. The dictionary may be describing broader English.

If the teacher claims the form never exists and evidence shows otherwise, the learner can still use the required exam form while understanding the wider system accurately.

When two teachers disagree

Use the same evidence process. Some disagreements are genuine style preferences. Others come from different national standards or outdated rules.

Reference literacy as intellectual independence

The long-term goal is not that learners consult more references. It is that they ask better questions and need fewer arbitrary rules.

A learner who can verify language becomes less vulnerable to:

• viral grammar myths;
• AI hallucinations;
• prestige prejudice;
• random thesaurus substitution;
• outdated exam advice;
• overconfident corrections;
• search-result popularity masquerading as evidence.

The release test

A learner is becoming reference-independent when they can:

1. name the uncertainty;
2. choose a suitable tool;
3. read the evidence accurately;
4. recognise source limits;
5. make a context-sensitive decision;
6. use the form later without reopening the reference.

The reference has done its job when the learner no longer needs it for the same problem.

Advanced reference literacy: triangulate evidence across English systems

Advanced language questions rarely resolve through one lookup. The strongest method is triangulation: combine sources that answer different parts of the problem.

For example, to investigate whether a new phrase belongs in formal academic writing:

• a dictionary can show documented meaning and labels;
• an academic corpus can show attestation and frequency;
• a general corpus can show broader distribution;
• a style guide can show editorial preference;
• recent research can show whether usage is changing.

Agreement across these sources strengthens the conclusion. Disagreement tells the user which dimension is contested.

Triangulation is not majority voting

Three weak sources do not outweigh one source that directly owns the question. An official current examination specification should outrank several tutor blogs about examination format. A specialist dictionary can outrank general web pages about a technical term.

Weight sources by relevance and authority, not count.

Reference literacy in Academic English

Academic writers regularly need to verify:

• technical definitions;
• disciplinary collocations;
• reporting verbs;
• stance language;
• citation conventions;
• source identity;
• historical meaning;
• terminology consistency.

The Academic English owner explains research communication; reference literacy supplies language evidence.

Academic vocabulary check

Suppose a student wants to write The study accentuates the relationship between sleep and memory.

The dictionary shows that accentuate means make more noticeable or prominent. Corpus examples may show it is possible, but academic collocation with study may be much less natural than highlights, shows, identifies or demonstrates depending on evidence strength.

The student learns that advanced vocabulary should be chosen by discourse function, not rarity.

Reporting verbs

Use corpus and dictionary evidence to distinguish:

• report;
• argue;
• claim;
• suggest;
• demonstrate;
• observe;
• speculate.

Each positions the source differently.

Academic phrase frames

Corpora can reveal recurrent frames such as the results suggest that, consistent with previous research, a possible explanation is. Learners should analyse the rhetorical job rather than copy bundles into unrelated contexts.

Reference literacy in World Englishes

Variation claims require regional evidence. A global corpus may blur the exact variety. A learner asking whether a term is Singapore English should consult Singapore-focused sources where possible.

Regional dictionary labels

Regional labels are useful but do not tell the whole frequency pattern. One dictionary may label a word “Singapore English” while another omits it. Corpus data can show actual contemporary distribution.

Accent evidence

Pronunciation dictionaries can document reference pronunciations, but phonetic studies and recordings are needed to describe community variation accurately.

Register within region

A form may be locally widespread in informal speech and rare in formal writing. Regional evidence without register can produce misleading conclusions.

Reference literacy in language history

Historical claims are especially vulnerable to myths because attractive stories travel faster than archive caveats.

First citation

A historical dictionary can show the earliest citation in its evidence set. Say earliest recorded example or first attested in this source rather than automatically the word was invented in…

Semantic history

Track sense numbers through time. A word can exist centuries before the sense relevant to your text appears.

Spelling history

Historical corpora may contain variant spellings. Search tools can miss evidence if the user searches only the modern form.

Etymology

Use etymological evidence to explain origin and development, not to dictate modern usage.

The History of English & Language Change owner develops these principles fully.

Reference literacy in evidence and data work

The Evidence, Data & Multimodal English owner asks whether claims match evidence. Reference literacy asks whether the language evidence itself is well sampled and interpreted.

A corpus table is data. It therefore needs the same questions:

• What population?
• What sample?
• What denominator?
• What uncertainty?
• What comparison?
• What does the measure actually mean?

Frequency claims need denominators

“Phrase A appears 1,000 times and Phrase B 500 times” is meaningless if A was searched in a corpus twice the size.

Use normalised counts for fair comparison.

Small counts

When a form occurs three times, percentage differences can look dramatic. Report raw counts alongside rates where useful.

Corpus composition can create false trends

If later decades contain more social-media text, an increase in an informal form may reflect dataset change rather than language change. Time-comparable corpora try to control this problem.

Token vs type frequency

Tokens count every occurrence. Types count distinct forms. A text can contain 10,000 word tokens but far fewer unique word types.

Vocabulary studies need to know which measure is being reported.

Lemma frequency

Combining inflected forms can change ranking. run as lemma includes runs, ran, running where the annotation groups them appropriately.

Sense frequency

Raw word frequency can hide sense distribution. A common word may have a rare technical sense. Dictionaries and sense-tagged corpora can help.

Frequency vs familiarity

A learner can know a rare word because of a hobby and fail to know a common academic phrase. Frequency predicts exposure imperfectly at the individual level.

Frequency vs usefulness

A specialised word can be essential for one course despite low general frequency. Learning priorities should combine frequency, domain relevance and communicative need.

Reference literacy and AI-generated text

AI makes verification more important because fluency no longer signals evidence.

Generated examples

An AI-generated sentence may be grammatical but unattested or stylistically odd. Verify important phraseology in corpora or dictionaries before teaching it as a model.

Generated dictionary-style definitions

AI can produce clear definitions but may merge senses or omit labels. Use it for explanation after verifying the entry.

Generated usage rules

AI often turns probabilistic style patterns into absolute rules. Ask for the evidence and check independently.

Generated citations

Never treat a bibliographic-looking string as proof the source exists. Verify title, author, publication and content.

Generated etymologies

Language-origin myths are easy for AI to reproduce because they are common online. Check historical dictionaries or scholarship.

Generated regional English

Requests such as “rewrite this in Singlish” can produce stereotypes or mixed features. Authentic community evidence matters.

Reference literacy and grammar checkers

Grammar checkers use rule systems and statistical models. Their suggestions can be valuable but should be classified.

A correction might be:

• a genuine grammar error;
• a likely typo;
• a readability suggestion;
• a style preference;
• a false positive;
• a change in meaning.

Accept with explanation

If the learner accepts a correction, they should be able to explain what relationship improved.

Reject with evidence

If the learner rejects it, they should have a reason beyond preference: target style, corpus evidence, intended meaning or legitimate variation.

Reference literacy and plagiarism/attribution

Dictionaries and corpora are references too. Definitions copied verbatim in academic work may require quotation/attribution depending on context and institutional expectations. Corpus examples can be copyrighted texts depending on source.

Reference use should respect licensing and academic integrity.

Corpus quotation

Corpus interfaces can display sentences from copyrighted or licensed texts. Researchers should follow the corpus terms for citation and quotation.

Reference notes

When using a reference to support a public SETC article, record:

• source;
• edition/version;
• date accessed where relevant;
• what exact claim it supports;
• scope/limitations.

This allows later review when usage changes.

Volatility classes for language-reference claims

Stable

Basic grammar structures and established historical facts usually change slowly.

Usage-sensitive

Register, regional distribution, connotation and new vocabulary require periodic review.

Institutional

Exam, citation and style rules should be rechecked when the institution updates them.

Fast-moving

AI terminology, platform language and internet slang can change rapidly.

This connects reference literacy directly to SETC’s Atlas governance model.

Reference versioning

Online dictionaries update continuously; printed editions do not. Corpora can release new versions. If a precise claim depends on the dataset, note which version was used.

Reproducible corpus queries

For serious research, record:

• corpus name/version;
• query string;
• filters;
• date range;
• register;
• region;
• normalisation;
• association measure;
• sample method.

Another researcher should be able to understand how the result was produced.

Corpus screenshots are not enough

A screenshot can show a result but often hides the query settings. Preserve the method.

Statistical significance vs linguistic importance

A large corpus can make small differences statistically detectable. The language-learning question may be whether the difference is large enough to change teaching or editing.

Effect size and practical relevance matter.

Reference literacy for editors

An editor needs a hierarchy:

1. meaning and factual accuracy;
2. grammar and syntax;
3. terminology consistency;
4. target standard;
5. house style;
6. rhythm and preference.

Do not spend editorial authority on personal taste while missing meaning errors.

Reference literacy for lexicographers

Advanced learners can reverse the perspective and ask how dictionaries are built:

• How is evidence collected?
• How are senses split?
• How are examples chosen?
• When is a new word included?
• How are labels assigned?
• How are pronunciation variants selected?

This helps users read dictionaries critically without undermining trust in them.

Reference literacy for teachers

A teacher should maintain a small trusted toolkit rather than improvise every usage answer from memory. The toolkit can include:

• a strong learner dictionary;
• a general/historical dictionary where available;
• a collocations resource;
• a contemporary corpus;
• relevant official curriculum/exam documents;
• a grammar reference;
• discipline-specific sources when teaching Academic English.

The exact brands matter less than source quality and fit.

Reference literacy for parents

Parents do not need to adjudicate every grammar dispute. They can help children learn the habit:

“What kind of question is this, and where could we check?”

That response models inquiry rather than anxiety.

Reference-literacy practice labs

Reference literacy improves through repeated decisions, not by reading about tools. The labs below are designed to make the learner formulate a question, gather evidence, interpret it and then use the result without the reference open.

Lab 1: one word, five senses

Choose a polysemous word such as issue, point, case, run or field.

1. Find five different authentic sentences.
2. Match each to a dictionary sense.
3. Explain what contextual evidence selected the sense.
4. Create one new sentence for each sense.

This trains semantic discrimination rather than definition memorisation.

Lab 2: grammar inside the dictionary

Choose five verbs often misused by learners: for example suggest, discuss, explain, recommend, depend.

For each one, record:

• transitivity;
• common complement patterns;
• prepositions;
• one corpus example;
• one original sentence.

Then hide the notes and write a fresh paragraph using all five naturally.

Lab 3: thesaurus trap

Start with a simple sentence:

The result was important because it changed the next decision.

Use a thesaurus to find five alternatives for important. Check each in dictionaries and corpora. Rank them by fit for the sentence and explain why the top choice wins.

The point is to discover why synonym lists require evidence.

Lab 4: collocation map

Choose an academic noun such as evidence, argument, risk, decision or method.

Build categories:

• adjective + noun;
• verb + noun;
• noun + preposition;
• noun + noun;
• common phrase frames.

Use a collocations dictionary and corpus. Then write a paragraph using only combinations that serve real meaning.

Lab 5: spoken vs academic English

Choose ten phrases common in conversation. Search them in spoken and academic subcorpora. For each:

• compare normalised frequency;
• inspect examples;
• identify a more typical academic alternative where one exists;
• decide whether the difference is formality, genre or meaning.

This teaches register empirically.

Lab 6: British vs American English

Choose five known differences and five uncertain ones. Compare dictionaries and regional corpora. Classify each as:

• spelling convention;
• vocabulary preference;
• grammar preference;
• pronunciation difference;
• false stereotype.

Do not choose examples only because they are famous. Include at least one case where the difference is smaller than expected.

Lab 7: Singapore English evidence

Take one Singapore English term or structure. Gather:

• a dictionary/linguistic description where available;
• five authentic examples;
• register information;
• formal Standard English equivalent where relevant;
• one international explanation.

The learner should be able to say what is local, what is formal and what is globally transparent without ranking speakers.

Lab 8: historical word detective

Choose a word whose modern meaning seems surprising.

1. Check etymology.
2. Find historical senses.
3. Identify semantic broadening, narrowing, metaphor or another change where appropriate.
4. Find dated examples.
5. Write a short history without claiming more than the evidence.

Lab 9: first-attestation myth

Find an online claim that a famous writer “invented” a word. Check a historical dictionary and corpus. Rewrite the claim precisely:

“The earliest citation currently recorded in X is…”

or

“The writer helped popularise the form, but earlier evidence exists…”

Lab 10: usage dispute dossier

Choose one disputed form. Build a one-page evidence dossier containing:

• dictionary entries;
• grammar commentary;
• corpus frequency;
• register distribution;
• historical evidence;
• two style-guide positions;
• your recommendation for three contexts.

The learner must separate description from recommendation.

Lab 11: AI correction audit

Give an AI system ten sentences containing a mixture of:

• true grammar errors;
• valid regional variants;
• informal but grammatical language;
• deliberate stylistic fragments;
• rare but correct collocations.

Ask it to “correct everything”. Then verify every change using references.

Classify false positives.

Lab 12: grammar-checker audit

Run the same paragraph through two grammar checkers. Compare suggestions. Which are shared? Which conflict? Which change meaning?

Use evidence to accept or reject each suggestion.

Lab 13: corpus sampling

Search a common word with thousands of results. Read:

• the first 20;
• a random 20;
• 20 from academic writing;
• 20 from conversation.

Compare the impressions each sample creates. This makes sampling bias visible.

Lab 14: dispersion

Find two phrases with similar total frequency. Check whether one appears across many texts while the other is concentrated in one source or subgenre.

Explain why the first may be a better general-learning priority.

Lab 15: collocation statistics

Compare a frequent collocate and a high-association but rare collocate. Decide which one should be taught first to an intermediate learner and why.

Lab 16: corpus query repair

Write a query that accidentally produces irrelevant results. Diagnose why. Narrow by lemma, part of speech, register or context.

Corpus literacy includes debugging searches.

Lab 17: dictionary-definition comparison

Compare one complex word across:

• a learner dictionary;
• a general dictionary;
• a historical dictionary;
• a specialist dictionary where relevant.

Explain what each entry is optimised to do.

Lab 18: style-guide conflict

Choose a punctuation or capitalisation issue where two style guides differ. Write two correct versions, each consistent with its system.

The learner discovers that editorial consistency can produce multiple valid outputs.

Lab 19: examination authority check

Find a popular tuition claim about an exam format. Verify it against the official current specification. Record:

• whether it is correct;
• whether it is outdated;
• whether it is advice rather than a rule;
• what date the official source applies to.

Lab 20: corpus vs Google

Search the same phrase through a corpus and web search. Compare:

• counts;
• duplicate results;
• register;
• source quality;
• geographic distribution;
• machine-generated pages.

Explain why the tools answer different questions.

Lab 21: proper-name pronunciation

Select names from several languages. Compare dictionary/media pronunciation guides with first-person or official pronunciation evidence where available.

Discuss why spelling-based guessing can fail.

Lab 22: technical terminology

Choose a word with everyday and technical meanings: work, stress, significant, force, validity.

Compare general and specialist definitions. Use both in original sentences.

Lab 23: phrase strength

Compare suggests, indicates, supports, demonstrates, proves in academic corpora and dictionaries. Arrange them by typical claim strength cautiously, then find examples that complicate your ordering.

The complication is the lesson: stance depends on context.

Lab 24: offensive-language labels

Choose a historical or socially sensitive term. Compare current dictionary labels, historical senses and modern community guidance. Write a note explaining why a definition alone is insufficient.

Lab 25: lexical change watch

Choose a new technology or social term. Revisit it monthly using current dictionaries, corpora or news datasets. Track:

• spelling stabilisation;
• sense development;
• new collocations;
• register shift;
• dictionary inclusion.

This turns language change into observable evidence.

Practice design principles

Start with a real uncertainty

Reference work is more memorable when the learner needs the answer for actual reading or writing.

Require a decision

Looking up evidence without choosing how to write or interpret leaves the work incomplete.

Require explanation

The learner should state why the evidence supports the decision.

Remove the tool

Retest later without reference access where the goal is internalised language.

Vary the surface

Use the same verification skill on a different word, variety or genre.

Increase ambiguity gradually

Early tasks can have clear answers. Advanced tasks should include mixed evidence and context-dependent choices.

A reference-literacy assessment

Instead of asking students to memorise dictionary abbreviations, give them five language problems and access to several references. Assess:

• tool selection;
• query design;
• evidence reading;
• source limitation;
• decision quality;
• explanation.

This tests the capability the real world requires.

Reference myths that make English harder than it needs to be

“If it is in the dictionary, it is correct.”

Dictionaries record language. Entries can include informal, regional, nonstandard, offensive, archaic and disputed forms. Inclusion tells you that lexicographers have evidence worth documenting. It does not guarantee suitability for your examination, journal or formal report.

“If it is not in the dictionary, it is not a word.”

No dictionary contains every proper name, technical coinage, regional item, newly emerging form or productive compound. Dictionaries have scope and inclusion criteria.

“The first definition is the main meaning.”

Sense order depends on editorial policy. Use context.

“A thesaurus gives synonyms.”

It gives related candidates. Full synonymy is rare. Grammar, collocation, register and connotation can differ.

“Google result counts show frequency.”

Search-engine counts are not controlled corpus frequencies. They can include duplicates, search-engine estimation, machine-generated text and uneven regional/genre representation.

“More corpus hits means more correct.”

Frequency and correctness answer different questions. Common informal or nonstandard forms can have many hits; rare specialist language can be correct.

“Corpus evidence is objective, so interpretation is automatic.”

Corpus construction, annotation, query design, sampling and statistical measures all involve choices.

“A native speaker knows whether a phrase is correct.”

Native-speaker intuition is valuable evidence about familiar varieties and registers, but no individual has experienced all English. Corpus and dictionary evidence can reveal patterns outside one person’s repertoire.

“Oldest meaning is the true meaning.”

Modern meanings are established through current usage. Etymology explains history, not semantic authority.

“A style guide tells you grammar.”

Some advice concerns grammar; much concerns editorial consistency or preference. Classify the rule.

“The OED tells you modern learner usage.”

A historical dictionary is extraordinary for diachronic evidence but is not identical in purpose to a learner dictionary designed for current productive English.

“Learner dictionaries are simplified, so advanced users do not need them.”

Advanced learners often benefit enormously from clear grammar coding, collocations, register labels and examples.

“AI has read all dictionaries, so asking AI is equivalent.”

No. AI output is generated from model patterns and may not reproduce a current dictionary’s analysis accurately. Verify contested claims.

“Grammar checker = grammar authority.”

Grammar checkers combine grammar rules with statistical and stylistic suggestions. They can be wrong or change intended meaning.

“If two dictionaries disagree, one must be bad.”

They may have different sense boundaries, user populations, evidence, national orientation or update dates.

“Formal means better.”

Formal is a register description. The best language fits the task.

“Rare means sophisticated.”

Rare can mean technical, archaic, regional, specialised or simply uncommon. Sophistication is precision and control.

“Collocations are fixed idioms.”

Some collocations are strongly fixed; others are probabilistic preferences. Collocational strength exists on a continuum.

“A concordance line is a complete sentence.”

KWIC displays often truncate context. Expand when sense or discourse matters.

“Corpus examples are all good models.”

Corpora contain real language, including mistakes, quotations, unusual style and nonstandard forms. Authentic does not mean pedagogically ideal.

“Historical corpora show how everyone spoke.”

Surviving written records represent literacy, genre and preservation patterns. Spoken historical English is reconstructed indirectly.

“Frequency lists tell me exactly what to learn.”

They are one prioritisation signal. Personal goals, domain, school stage and existing knowledge matter.

“One corpus is enough.”

Only if its population matches the question adequately. Advanced claims often need comparable datasets or triangulation.

Frequently asked questions

Which English dictionary is best?

There is no single best dictionary for every task. Choose by purpose. Learner dictionaries are excellent for productive use; historical dictionaries for diachronic questions; specialist dictionaries for technical domains; general dictionaries for broad coverage. Advanced users often compare more than one.

Should students use monolingual or bilingual dictionaries?

Both can be useful. Bilingual dictionaries provide fast conceptual access; monolingual learner dictionaries expose English sense boundaries, grammar, collocations and register. A strong learner can move between them.

When should I use a corpus?

Use a corpus when the question concerns actual distribution: frequency, collocation, phrase pattern, register, regional variation, historical change or authentic context.

Is a corpus difficult to learn?

Basic corpus literacy is straightforward: search, read concordance lines, filter by register and compare examples. Advanced statistics can come later.

What is the difference between a corpus and Google?

A corpus is a designed language dataset with known or documented composition. Google is a web search and ranking system. Google is broader for discovery; a corpus is stronger for controlled language evidence.

What is the difference between corpus and corpora?

Corpus is singular; corpora is the traditional plural, though corpuses appears in some general contexts. In linguistics, corpora is standard terminology.

What is a concordance?

A display of occurrences of a search term with surrounding context, often aligned so patterns become visible.

What is KWIC?

Key Word In Context: a common concordance format that centres the search term with left and right context.

What is a collocation?

A recurrent combination of words whose association is conventional or statistically stronger than chance, such as heavy rain or make a decision.

What is colligation?

Colligation refers to association between lexical items and grammatical patterns or categories, for example a word’s tendency to appear in a particular syntactic structure.

What is semantic preference?

A tendency for a word to occur with items from a particular semantic class.

What is semantic prosody?

An evaluative association that emerges from recurrent co-text, such as a word tending to appear around negative or positive situations.

What does “attested” mean?

Documented in evidence. An attested form has been observed in the relevant dataset or record. Attestation does not automatically imply commonness or standardness.

What does “nonstandard” mean?

Outside the codified standard targeted by the reference in that sense. It does not necessarily mean random or unused.

What is a usage label?

A label that tells you where, how or with what attitude a word or sense is used: formal, informal, regional, offensive, approving, dated and so on.

What is a lemma?

The dictionary/corpus base form under which related inflected forms are grouped.

What is a lexeme?

An abstract lexical unit underlying word forms and senses in linguistic analysis. The exact theoretical treatment can vary.

What is lexicography?

The practice and study of dictionary making.

What is lexical frequency?

How often a word or lexical unit appears in a defined dataset.

Why do dictionaries add words that people dislike?

Because dictionaries document language evidence. Inclusion is not endorsement.

Who decides what a word means?

Meaning emerges through community use over time. Lexicographers analyse and document it; institutions can define technical meanings inside specific domains.

Can a dictionary be wrong?

Entries can become outdated, omit variation, make debatable sense divisions or contain errors. Strong dictionaries have editorial processes and update evidence over time.

Why do dictionaries disagree about pronunciation?

They may represent different reference accents, record different accepted variants or use different transcription conventions.

Why do dictionaries disagree about word class?

Some constructions sit near category boundaries and linguistic analyses can differ. Dictionary granularity also varies.

How do new words enter dictionaries?

Editors gather evidence that a form has sufficient sustained use, identifiable meaning and distribution according to their inclusion policy. A viral coinage may be monitored before inclusion.

Can corpora predict future English?

They can show trends, not guarantee future outcomes. Many innovations rise and disappear.

How large should a corpus be?

Large enough and appropriate enough for the question. Rare phenomena need more data; specialised questions may need smaller targeted corpora.

Can I build my own corpus?

Yes. A class can collect essays, articles, speeches or transcripts for a defined question. Ethical, copyright and privacy constraints must be respected.

What is corpus bias?

Systematic distortion caused by which texts/speakers are included, how they are sampled, annotated or processed.

Can corpus results prove grammar?

Corpus evidence shows attested patterns and distribution. Grammar analysis also involves structural relationships and sometimes judgements about possible but rare forms.

What if my correct sentence has zero corpus hits?

Language is productive. A novel but grammatical sentence may never have occurred in the corpus. Search the underlying pattern, not only the full sentence.

Can corpus frequency replace teacher judgement?

No. Teaching decisions also consider developmental level, task, meaning, transfer and curriculum target.

Should I memorise corpus statistics?

No. Learn to interpret them when needed.

What is a style guide for?

Consistency and editorial decisions within a community or publication.

What if the style guide conflicts with the dictionary?

Follow the style guide for that document if it legitimately governs the task, while understanding the broader dictionary evidence.

What should I trust when AI and a dictionary disagree?

For dictionary-type facts, the dictionary and relevant corpus/grammar evidence should carry more weight than an unsupported AI assertion.

Can AI help me search corpora?

It can help formulate queries or explain results, but the actual corpus output should remain visible and verifiable.

Should children learn IPA?

A small amount can be useful, especially for older learners using dictionaries independently. Full phonetics training is not necessary for every child.

What reference skills matter most for PSLE or Secondary students?

Sense selection, word class, grammar pattern, collocation, register, pronunciation, formal-standard target and efficient lookup.

What reference skills matter most for JC/GP students?

Academic phraseology, evidence language, source verification, dictionary/corpus comparison, register and current factual/terminological references.

What reference skills matter most for university students?

Disciplinary terminology, corpus/reference evaluation, citation rules, source provenance, technical definitions, academic phraseology and AI verification.

Trusted external reference routes

The exact tools available to a reader can change. SETC’s system is therefore tool-independent, but several high-quality current resources illustrate the categories well.

• Oxford Learner’s Dictionaries — current learner-focused definitions, pronunciation, grammar and usage information.
• Oxford dictionary labels — explanation of formal/informal, attitude and other usage labels.
• Oxford Collocations Dictionary introduction — explains productive collocation use.
• Cambridge Dictionary — current dictionary, examples and corpus-linked usage material.
• Oxford English Dictionary — historical lexical evidence and sense development, subject to access conditions.
• Corpus of Contemporary American English (COCA) — large contemporary American corpus across multiple genres.
• British National Corpus interface — British English corpus access through English-Corpora.org.

These tools are examples of reference categories, not a claim that one provider must answer every question.

Current corpus scale example

English-Corpora.org describes the expanded COCA as containing more than one billion words with balanced yearly genre sampling across 1990–2019. That scale is useful, but the design still defines the population: contemporary American English across the included genres.

Size does not remove scope.

Current collocation example

Cambridge Dictionary examples are explicitly drawn from corpora and web sources, illustrating how modern lexicography can connect dictionary entries to observed usage. Oxford’s collocations resource is explicitly designed to support productive writing. These are complementary forms of evidence rather than interchangeable interfaces.

Mastery standards

Foundation

The learner can use a dictionary to choose the correct sense, word class and pronunciation.

Developing

The learner reads grammar, register and regional labels and checks examples.

Independent

The learner can investigate collocations and phrase patterns using corpora or collocation tools.

Advanced

The learner compares registers, regions and time periods while interpreting frequency responsibly.

Expert reference literacy

The learner coordinates dictionary, corpus, grammar, historical, style and institutional evidence; documents methods; resolves conflicting sources; and knows what the evidence cannot establish.

Reference-literacy release test

Give the learner an unfamiliar disputed usage and access to several tools. Do not tell them which source to use.

A strong performance should:

1. define the question precisely;
2. select appropriate references;
3. search efficiently;
4. distinguish sense and register;
5. interpret corpus scope and frequency;
6. identify conflicting evidence;
7. check target institutional requirements;
8. make a proportionate recommendation;
9. state uncertainty honestly;
10. use the final form correctly in a new sentence.

The learner has not merely found an answer. They have demonstrated a method for finding future answers.

The institutional standard for SETC

SETC should model the same discipline publicly.

When an article claims that a form is:

• common;
• formal;
• regional;
• historical;
• new;
• rare;
• preferred in academic writing;
• required by an examination;

the evidence should match the claim.

“Common” should not come from memory alone. “Historical” should not come from an unsourced meme. “Required by the exam” should not come from an outdated tuition page. “AI says” should never function as final linguistic evidence.

This reference standard is part of the reason SETC can become a long-lived English institution: the site does not only teach answers. It teaches and practises the process by which English claims are verified.

How dictionaries are built: lexicography as evidence work

Dictionaries can feel authoritative because the finished entry is concise and polished. The underlying work is much less simple. Lexicographers collect evidence, identify patterns, decide where one sense ends and another begins, select examples, assign labels, choose pronunciation models, track new uses and revise earlier analyses when the language changes.

Understanding this process makes learners better dictionary users. It explains why entries change, why dictionaries disagree and why no dictionary can be treated as the language itself.

Evidence collection

Modern lexicography draws on corpora, published texts, spoken data, specialist sources, historical records and editorial databases. Different dictionaries use different evidence sets.

A learner dictionary may prioritise high-frequency contemporary usage and pedagogical clarity. A historical dictionary may prioritise dated attestations across centuries. A specialist dictionary may prioritise expert terminology.

Headword selection

Editors decide whether a form deserves a separate entry, subentry, phrase entry or cross-reference. Productive compounds and transparent derivations may not all receive independent headwords.

This is why absence from the headword list is not proof that a form is impossible.

Sense division

Meaning often forms a continuum. Editors need to decide whether recurrent uses are:

• one broad sense;
• separate related senses;
• figurative extension;
• specialist use;
• idiom;
• homonym.

Two dictionaries can reasonably divide the same evidence differently.

Definition writing

A definition must be broad enough to cover typical instances but narrow enough to exclude neighbouring concepts. Learner dictionaries also aim for accessible defining language.

Definitions are therefore models, not perfect substitutions.

Example selection

Examples need to demonstrate sense and language behaviour without introducing unnecessary complexity. Some dictionaries edit corpus examples for length or clarity; others use authentic examples more directly.

The user should understand the source policy where examples are being treated as evidence.

Usage labels

Labels require judgement. Editors examine distribution and social meaning. Whether a word is labelled informal, dated, offensive or regional depends on current evidence and editorial policy.

Labels can change as usage changes.

Pronunciation entries

Pronunciation editors select reference variants and represent them through phonemic transcription and audio. No finite entry can capture all individual and regional variation.

Etymology

Etymological work uses historical forms and comparative evidence to reconstruct origins and pathways. Popular stories can be attractive but unsupported. Historical dictionaries revise etymologies when better evidence becomes available.

New-word monitoring

Lexicographers watch emerging forms across news, social media, specialist fields and corpora. They look for sustained evidence, stable meaning and distribution before deciding whether inclusion is justified.

A viral word can disappear before it becomes lexicographically durable.

Sense change

An existing word may acquire a new sense faster than a new headword appears. Dictionary revision therefore involves re-analysing old entries.

Regional inclusion

Global dictionaries increasingly document English beyond a small set of traditional national standards. Regional lexical evidence requires local expertise and suitable corpora.

Editorial neutrality

Documenting a word does not endorse the behaviour or ideology associated with it. Dictionaries describe offensive, hateful, illegal and controversial language because users need to understand what exists.

Dictionary dating

Printed dictionaries freeze an editorial state at publication. Online dictionaries can update continuously. When a usage claim depends on current status, edition and update context matter.

Advanced corpus reasoning

Basic corpus literacy asks what occurs. Advanced corpus literacy asks how robust the pattern is.

Dispersion in practice

Imagine Phrase A appears 1,000 times, but 900 examples come from one bestselling novel repeated across an archive. Phrase B appears 600 times across 400 independent texts. For general usage, Phrase B may be more broadly established despite lower raw frequency.

Dispersion protects learners from being misled by concentration.

Genre imbalance

Suppose a corpus contains far more news than conversation. Overall frequency can reflect the corpus composition. Compare within genre or use balanced/normalised measures.

Author imbalance

A prolific author or source can dominate results. Document-frequency and source filters can reveal this.

Temporal imbalance

If one decade contains twice as many words, raw historical counts can create false growth. Compare rates or matched samples.

Regional imbalance

A global corpus can contain much more data from one country. Apparent “global” patterns may therefore be dominated by that region.

Annotation error

Automatic part-of-speech taggers can misclassify ambiguous forms. For low-frequency patterns, inspect examples manually.

Tokenisation

Corpus systems decide what counts as a token. Contractions, hyphenation, punctuation and multiword expressions can be tokenised differently. This matters for precise frequency comparisons.

Lemmatisation

Automatic lemma grouping can make mistakes with ambiguous forms or irregular words. Check the interface documentation.

Case sensitivity

Searching Apple vs apple may distinguish company and fruit contexts. Query design should match the question.

Homographs

Frequency for a spelling such as lead can mix noun and verb or different pronunciations. Part-of-speech filters or context are needed.

Multiword expressions

Some expressions behave as units even when the corpus tokenises them separately. Search phrase sequences and variants.

Negative evidence

A pattern’s absence becomes more meaningful when the corpus is large, appropriate and searched through several plausible variants. Even then, phrase productivity means “not found” should be phrased carefully.

Confidence and uncertainty

For language teaching, exact statistical confidence intervals are rarely needed. But the mindset matters: ten examples support a weaker generalisation than ten thousand well-dispersed examples.

Effect size

When comparing two registers, ask how large the difference is. A statistically detectable difference can be too small to justify a strong pedagogical rule.

Association vs prediction

A high collocation association score tells you the words occur together unusually often. It does not mean the phrase is the most likely next phrase in every context.

Corpus causality

Corpus co-occurrence does not prove why a language pattern exists. Historical, social or psycholinguistic explanation may require other evidence.

Corpus triangulation

For a strong usage claim, compare:

• at least two corpora where feasible;
• dictionary labels;
• grammar/reference descriptions;
• the target register or institution.

Manual coding

When a word has several senses, researchers may manually classify samples. This produces richer analysis but introduces human judgement. Coding guidelines and agreement checks improve reliability.

Corpus building in a classroom

Students can build a small corpus from:

• school model essays;
• news articles;
• science explanations;
• speeches;
• their own drafts;
• peer writing.

Then ask questions such as:

• Which reporting verbs dominate?
• Which transitions are overused?
• Which adjectives appear in strong essays?
• How does student writing differ from published prose?

Privacy and consent matter when using student work.

A mini lexicography project

Advanced learners can create a dictionary entry for a local or emerging word.

Required fields:

• headword;
• pronunciation;
• part of speech;
• definition;
• sense distinctions;
• register/region label;
• five authentic examples;
• common collocations;
• etymology where known;
• evidence of current use;
• one note on uncertainty.

The exercise exposes how much analysis sits behind one entry.

A mini corpus research project

Question: How do writers use “may” and “might” differently in academic English?

1. Choose an academic corpus.
2. Search both modal verbs.
3. Normalise frequency.
4. Sample 100 examples of each.
5. Classify functions: possibility, permission, concession, tentative suggestion, etc.
6. Compare grammatical environments.
7. Check a grammar reference.
8. Report patterns with limits.

The learner discovers that frequency alone is not the research question; function matters.

A mini regional corpus project

Question: How is “revert” used in Singapore/Indian business English compared with general American English?

Use regional/business data where possible. Separate the sense “reply/get back to” from the general sense “return to an earlier state”. Compare register and region. Then write guidance for an international audience.

This integrates World Englishes and reference literacy.

A mini historical dictionary project

Question: How did “broadcast” move from agriculture to media?

Trace earlier senses, dated examples and semantic extension. Explain the metaphorical pathway without claiming the later sense appeared instantly.

Reference literacy and knowledge graphs

A future English knowledge system should not store isolated definitions only. Concepts need relationships:

• synonym;
• antonym;
• hypernym;
• hyponym;
• part–whole relation;
• collocation;
• word family;
• register;
• regional variant;
• historical predecessor;
• genre association.

SETC’s Atlas is already moving in that direction at page scale. Reference literacy provides the lexical evidence layer beneath it.

Reference literacy and retrieval systems

Search and AI systems increasingly retrieve fragments from sources rather than presenting whole documents. This makes provenance important. A quotation or definition extracted from a page should retain enough source identity to be verified.

Machine convenience should not break the evidence chain.

The expert standard

An expert language user does not know every word or rule. They know how to find strong evidence quickly, recognise uncertainty and stop when the evidence is sufficient for the task.

That is a more durable form of expertise than memorising thousands of isolated usage verdicts.

Reference-literacy operating manual by stage and task

The same verification principles should look different for a ten-year-old, a GP student, a university writer and an editor. Tool complexity should increase only when the learner’s questions require it.

Early Primary: dictionary confidence without lookup overload

The first goal is not speed. It is learning that a reference can answer a language question.

Useful tasks:

• find a word alphabetically or through search;
• listen to pronunciation;
• identify noun/verb/adjective where appropriate;
• choose between two simple senses;
• read one example;
• use the word in a new sentence.

A parent or teacher can ask: Which part of the entry helped you?

That question shifts attention from answer to evidence.

Upper Primary: meaning, grammar and phrase

By upper Primary, learners can begin noticing:

• countability;
• common prepositions;
• word families;
• register labels;
• collocations;
• pronunciation stress.

They do not need corpus statistics. They can still compare several dictionary examples and notice repeated phrase patterns.

PSLE: efficient reference use under preparation conditions

PSLE students need strong independent control of vocabulary, grammar, editing and oral preparation, but actual examination conditions may limit reference access. Therefore, reference work belongs mainly in learning and review.

Useful system:

1. collect errors from practice;
2. classify each as meaning, grammar, collocation, spelling or register;
3. verify through a reference;
4. write the corrected pattern;
5. retest later without the reference.

The dictionary should reduce repeated error, not become a permanent crutch.

Secondary 1–2: compare words and registers

Secondary learners can investigate near-synonyms and begin basic corpus-style thinking through authentic examples.

Questions:

• Which word is more formal?
• Which noun does this adjective usually modify?
• Which verb fits the meaning?
• Is this phrase conversational or written?
• Is this a national variant?

This stage is ideal for breaking the “one word = one definition” model.

Secondary 3–4: usage evidence and editing

Upper-secondary learners can use dictionary labels, collocation tools and simple corpora directly.

They should learn to distinguish:

• grammar error;
• valid variant;
• informal register;
• rare but possible form;
• formal exam target.

This improves editing because the learner stops changing sentences randomly.

Secondary examination preparation

When checking a disputed answer or teacher correction:

1. identify the exact construction;
2. check current Standard English references;
3. check official examination expectations where relevant;
4. record the exam-safe form;
5. retain broader variation knowledge separately.

This lets examination strategy coexist with accurate linguistics.

JC / General Paper

GP students benefit from advanced reference literacy because they work with argument, evidence and topic-specific vocabulary.

High-value skills:

• checking precise definitions of issue terms;
• distinguishing reporting verbs;
• testing academic collocations;
• checking whether evidence language overstates certainty;
• using corpora to inspect formal phraseology;
• verifying current institutional or factual terminology.

The learner should be able to explain why a phrase is appropriate, not merely that it “sounds academic”.

GP comprehension

Reference work during preparation can illuminate:

• polysemy;
• semantic prosody;
• tone words;
• register shifts;
• historical or regional senses;
• authorial lexical choice.

This deepens reading without turning every passage into dictionary work.

University / Academic English

Academic learners should combine dictionaries with corpora and discipline-specific references.

Before submitting important prose, investigate:

• technical terminology;
• reporting verbs;
• collocations;
• stance markers;
• citation style;
• discipline-specific phraseology;
• regional/institutional conventions.

A corpus should support the writing decision; it should not replace conceptual understanding.

Research students

Researchers need reproducible reference practice. Record exact reference editions, corpus query settings and institutional standards where claims depend on them.

If a dissertation makes a claim such as “X is increasingly used in academic English,” the reader should be able to see what evidence supports increasingly and academic.

Teachers and tutors

Teachers should build an explicit correction taxonomy:

This reduces arbitrary correction and helps learners generalise.

Editors

Editors should decide whether they are working for:

• meaning;
• standard grammar;
• house style;
• international accessibility;
• brand voice;
• legal/technical precision.

Different goals require different references.

Writers

Writers can use references in three phases:

Drafting: quick meaning and terminology checks.

Revision: collocation, phraseology, register and consistency.

Final verification: style, spelling convention, citations and high-risk terms.

Do not interrupt every sentence with lookup during first-draft idea generation unless the term is essential to the argument.

Translators and multilingual writers

Bilingual dictionaries provide entry points, but productive English should be verified through monolingual evidence. Pay special attention to false friends, collocations and register.

The Multilingual English & Mediation owner supplies the broader translation system.

Parents

Parents can encourage reference independence by replacing “I’ll tell you” with:

“What exactly are you trying to find out?”

Then help the child choose the source.

Task-based reference menus

Writing a narrative

Useful references:

• dictionary for precise sense;
• thesaurus for candidate generation;
• corpus for natural phraseology;
• pronunciation only if dialogue/oral performance matters.

A formal style guide is usually less important than voice and meaning unless required by the task.

Writing an argument

Useful references:

• dictionary for key-term definitions;
• academic corpus for reporting/stance language;
• factual sources for evidence;
• style guide for citations;
• official definitions where policy/legal terms matter.

Editing grammar

Use learner dictionaries, grammars and corpus examples. Avoid relying only on spellcheck or AI correction.

Preparing oral communication

Use pronunciation audio, stress notation, collocation evidence and authentic recordings of the relevant register/variety.

Reading literature

Use historical dictionaries, literary notes, etymology and period corpora when modern senses mislead.

Reading news

Use current dictionaries, institutional/background references and source verification. Corpus frequency is secondary unless the language itself is the object.

Writing research

Use specialist dictionaries, academic corpora, citation standards, source databases and current scholarly literature.

Checking a viral language claim

Use dictionaries, corpora, historical evidence and linguistic scholarship. Do not let meme repetition function as proof.

Checking a school answer key

Use the target curriculum, question wording and Standard English evidence. An answer key can contain mistakes; a dictionary can also fail to capture the intended test construct. Analyse both.

Checking a tutor’s rule

Ask whether it is:

• a grammar rule;
• exam-safe advice;
• style preference;
• memory shortcut;
• myth.

Then verify with the matching authority.

Reference-literacy curriculum: twelve moves

Move 1: name the uncertainty

Do not begin with the tool.

Move 2: select the reference type

Dictionary, corpus, grammar, style, historical, specialist, official.

Move 3: read the source scope

Who/what does it represent?

Move 4: query precisely

Sense, lemma, phrase, register, period.

Move 5: inspect examples

Do not trust numbers without language.

Move 6: compare

Near-synonyms, variants, corpora, references.

Move 7: identify labels

Formal, regional, dated, technical.

Move 8: interpret statistics

Frequency, dispersion, association.

Move 9: check the target institution

Exam, journal, school, workplace.

Move 10: decide

Write the recommendation.

Move 11: use

Produce the word/pattern in a real sentence.

Move 12: remove the reference

Retest later.

Reference-literacy failure states

Lookup addiction

The learner checks every uncertain word and loses reading fluency.

Authority addiction

The learner cannot decide without one favourite dictionary/teacher/AI.

Frequency addiction

The learner chooses whatever has more corpus hits regardless of context.

Formality addiction

The learner replaces clear ordinary language with formal-looking synonyms.

Correction addiction

The learner accepts every grammar-check suggestion.

Etymology addiction

The learner explains current meaning through origins even when usage changed.

Search addiction

The learner treats web search as the answer to every language question.

Tool sophistication addiction

The learner uses advanced corpus statistics when a simple dictionary example would settle the problem.

The cure in every case is the same: return to the actual uncertainty and choose the smallest sufficient evidence.

Efficiency: stop when the evidence is sufficient

Reference literacy can become procrastination. A writer can spend forty minutes comparing dictionaries for a sentence whose ordinary wording was already clear.

Use stop rules:

• one strong dictionary is enough for an uncontested basic sense;
• add a second reference when labels or senses are uncertain;
• use corpus evidence when distribution/collocation matters;
• use official sources when an institution controls the rule;
• stop when additional sources no longer change the decision.

Leonie’s execution principle applies: verification must protect the work, not prevent completion.

A complete worked reference decision

Maren writes:

The policy will greatly alleviate educational inequality.

She is unsure about alleviate inequality.

Step 1 — uncertainty: collocation and claim strength.

Step 2 — dictionary: alleviate means make suffering, a problem or negative condition less severe. Examples often involve pain, poverty, suffering, symptoms or problems.

Step 3 — corpus: search alleviate inequality. The phrase is attested but less common than combinations such as reduce inequality or address inequality.

Step 4 — meaning: does the evidence justify saying the policy definitely makes inequality less severe? Perhaps the paper only argues it may improve access.

Step 5 — revision:

The policy may reduce one source of educational inequality by expanding access to…

The final sentence is less impressive lexically and more precise intellectually.

That is reference literacy serving writing.

A complete worked disagreement

A teacher says, “Never start a sentence with And.”

The learner checks current dictionaries/grammars and edited prose and finds sentence-initial And widely attested. The teacher may be using a pedagogical rule to prevent immature sentence chaining.

The learner should not respond, “Teacher is wrong, dictionary wins.” A more precise understanding is:

Sentence-initial “And” is grammatical in Standard English, but this class may discourage it as a style strategy at this stage or in this task.

The learner gains both accuracy and context control.

A complete worked corpus trap

Iona searches very unique and finds thousands of examples. She concludes the phrase must be formally ideal.

Maren checks style guidance and the semantics of unique; some editors treat it as absolute and prefer avoiding degree modification. Corpus evidence shows speakers do modify it, especially in promotional and conversational contexts.

Leonie asks the real task: a formal analytical essay. The group chooses distinctive, highly unusual or simply unique depending on intended meaning.

The corpus answered “does it occur?” The task required “what should we write here?”

A complete worked historical trap

A website claims that nice “really means foolish” because of its Latin history.

The historical dictionary shows earlier senses, but current dictionaries and corpora show modern meanings such as pleasant, kind or agreeable. The historical information explains semantic change; it does not invalidate current English.

The learner identifies the etymological fallacy.

The long-term outcome

A learner with strong reference literacy becomes harder to mislead. They can hear a grammar myth and ask for the target and evidence. They can see an AI correction and verify it. They can meet an unfamiliar regional word without assuming error. They can read an old text without forcing modern meaning onto it. They can write academic English with collocations drawn from actual usage rather than thesaurus prestige.

More importantly, they can keep learning after the tutor is gone.

The reference system is not there to replace judgement. It is there to make judgement evidence-aware, efficient and transferable.

From evidence to wording: how to write a language claim at the strength the references support

Reference work is incomplete until the evidence is converted into a proportionate statement. The strongest research process can still produce a misleading article if the final sentence is stronger than the data.

SETC therefore needs a vocabulary for evidence strength.

Attested

Attested means the form has been documented in the evidence set.

Safe claim:

The phrase is attested in contemporary American English.

Unsafe leap:

The phrase is common in American English.

Attestation does not establish frequency.

Frequent

A frequency claim should identify the dataset or comparison where possible.

Better:

In this contemporary spoken corpus, the form occurs substantially more often than the alternative.

Weaker:

Everyone uses this form.

Widespread

Widespread implies distribution across speakers, texts, regions or contexts. Check dispersion rather than raw count alone.

Typical

Typical suggests a recurrent pattern characteristic of the target context. This may require collocational, register or genre evidence.

Preferred

Preferred can mean statistical preference, stylistic recommendation or institutional rule. Name which one.

British academic writing in this corpus prefers X.

is different from:

The style guide recommends X.

Standard

Standard should refer to codified conventional usage within an identified Standard English system, not simply “the form I was taught”.

Nonstandard

Nonstandard does not mean meaningless, random or ungrammatical within every dialect. It means outside the codified standard in the relevant description.

Formal and informal

These are register claims. Use labels, distribution and examples. Do not infer formality from word length.

Regional

Regional claims should specify where. “British” can still hide England/Scotland/Wales/Northern Ireland and internal dialect diversity. Use the level of precision the evidence supports.

Historical

If a historical dictionary records a form in 1600, write:

The form is attested by the early seventeenth century.

Do not automatically write:

The form was invented in 1600.

Emerging

Emerging requires evidence of recent growth, new distribution or increasing social recognition. A new-looking word is not necessarily emerging; it may be old but newly visible to you.

Declining

A decline claim needs comparable time evidence. One modern learner’s unfamiliarity is not evidence of historical decline.

Rare

Rare relative to what? Corpus, genre, population and period should be clear.

Acceptable

Acceptability can refer to speaker judgements, formal-standard status or institutional acceptance. Those are different forms of evidence.

Natural

Naturalness is difficult to reduce to one metric. A phrase can be grammatical and attested yet feel marked in a particular register. Use authentic comparable examples and speaker evidence.

Grammatical

Grammaticality is a structural claim, not a frequency claim. A productive grammar can generate sentences never previously attested.

Unclear

If references conflict and evidence is insufficient, say so. “Usage is mixed” or “dictionaries differ in treatment” can be the most accurate conclusion.

Claim templates for SETC writers

Dictionary-supported meaning

Current learner dictionaries distinguish at least two relevant senses: X and Y. In this sentence, the surrounding context supports sense X because…

Collocation claim

Corpus and collocation evidence show X recurring strongly with Y, while Z is attested but much less typical in the target register.

Regional claim

Dictionaries label X mainly British, while corpus evidence shows Y more frequent in American data; both are intelligible and established variants.

Historical claim

Historical dictionary evidence records this sense by the nineteenth century, although surviving citations cannot establish the exact moment the sense first appeared in speech.

Style claim

Both forms are grammatical, but the target style guide recommends X for edited documents.

Exam claim

The current examination specification requires X. Broader English usage may include Y, but learners should use X for this assessment context.

Changing usage claim

Recent corpus slices show the form increasing in the sampled register, but the evidence is not sufficient to conclude that it has replaced the older variant generally.

Publication audit for language claims

Before SETC publishes a sentence containing words such as always, never, common, rare, formal, incorrect, regional, new, old-fashioned or preferred, the writer should ask:

1. What evidence class supports this adjective?
2. What population or register is implied?
3. Is there a stronger or narrower wording?
4. Does another recognised variety differ?
5. Could the claim change over time?
6. Does the target institution control the answer?

High-risk words in language advice

Always / never

Use only for genuinely categorical constraints or explicitly bounded rules.

Wrong

Specify whether the problem is grammar, meaning, target standard, spelling, register or house style.

Proper

Usually too vague. Replace with formal Standard English, target pronunciation, academic register or another precise category.

Native

Do not use native-speaker status as a shortcut for every usage judgement.

Modern

Specify time range when making historical contrasts.

British / American

Use as broad standards where appropriate, while recognising internal variation.

Academic

Academic English varies by discipline and genre. A phrase common in humanities essays may be rare in laboratory reports.

SETC internal evidence tags

For complex future editorial work, SETC can internally classify claims:

• DICT — dictionary/lexicographic evidence;
• CORP — corpus distribution evidence;
• HIST — historical dictionary/corpus evidence;
• GRAM — grammar/reference evidence;
• INST — institutional/examination authority;
• STYLE — style-guide recommendation;
• REG — regional/World Englishes evidence;
• RES — research literature;
• DOMAIN — specialist technical authority.

These tags need not appear publicly. Their purpose is to ensure the writer knows what kind of evidence is carrying the sentence.

When one source class is not enough

A statement such as “X is an increasingly common informal British usage” contains several claims:

• increasingly — time evidence;
• common — frequency/dispersion evidence;
• informal — register evidence;
• British — regional evidence.

One dictionary label may support part of the sentence but not all of it.

This decomposition is one of the most powerful habits in reference literacy.

Reference literacy as editorial quality control

When SETC becomes very large, not every page can be reread manually every month. The architecture therefore needs review triggers.

Foundational owner

Review when scholarship or internal conceptual architecture changes.

Usage owner

Review when dictionaries, corpora or current usage materially shift.

Institutional application

Review when the official syllabus, exam or policy changes.

Fast-moving digital page

Review frequently and date claims.

Reference literacy tells the maintenance system what evidence should trigger revision.

Reference literacy and federation

When a language question needs deeper investigation, use this SETC guide for dictionary and corpus verification and word use. Use eduKateSingapore’s vocabulary resources for larger word lists and eduKateSG’s guides for career and adult applications. Return here when you need to check disputed language evidence.

Federation works when each site knows which evidence world it owns.

Continue through SETC

• English Atlas — locate reference literacy in the whole English system.
• How English Works — understand the language mechanisms behind the reference evidence.
• Vocabulary in Use — apply dictionary and corpus evidence to usable vocabulary.
• Advanced English Vocabulary Mastery — build C1–C2 precision, collocation and phrase control.
• World Englishes & Variation — apply reference evidence to regional, social and register variation.
• History of English & Language Change — use historical dictionaries and corpora responsibly.
• Academic English — use reference literacy inside research, source use and academic language.
• Evidence, Data & Multimodal English — interpret quantitative and multimodal evidence proportionately.

The Reference Literacy standard

A learner has mastered reference literacy when they no longer ask only, “What does the dictionary say?” They ask what kind of question they have, which evidence can answer it, what population the evidence represents, what the source cannot establish and what choice the real task requires.

A tutor has mastered it when corrections become explainable and transferable rather than personal verdicts.

An institution has mastered it when its public English claims remain traceable to the right evidence and are revised according to the volatility of the claim.

Do not memorise every verdict. Learn how to investigate. Define the uncertainty, choose the evidence, read its limits, decide for the context, and carry the pattern forward without the tool.

The language-evidence dossier: one reusable template for any difficult usage question

When a language question matters enough to affect teaching, assessment, publication or a canonical SETC owner, build a small evidence dossier. The dossier prevents one source from quietly carrying more authority than it deserves and creates a record that can be reviewed later.

1. The exact question

Write the uncertainty as a sentence. Avoid broad prompts such as “Is this correct?”

Better questions include:

• Is this verb–preposition pattern accepted in formal Standard English?
• Which sense of this word is active in this nineteenth-century passage?
• Is this phrase characteristic of Singapore informal speech or wider English?
• Which spelling convention does this examination require?
• Is this collocation typical in academic writing?
• Has this usage increased in the last two decades?

2. Target context

Record the actual task:

• Primary/Secondary/JC examination;
• academic paper;
• creative writing;
• international presentation;
• local conversation;
• formal institutional document;
• historical interpretation.

The same evidence can support different choices under different targets.

3. Dictionary evidence

Record the relevant sense, grammar, labels, pronunciation and examples. If dictionaries disagree, preserve the disagreement rather than hiding it.

4. Corpus evidence

Record corpus name, region, period, register, query and enough context to identify sense. Note raw/normalised frequency where useful and whether results are well dispersed.

5. Grammar or linguistic evidence

For structural questions, consult an appropriate grammar or linguistic source. Corpus presence alone does not fully analyse grammar.

6. Historical evidence

If anyone makes a claim about age, tradition, origin or change, record historical dictionary/corpus evidence and the exact strength of the dating claim.

7. Regional/World Englishes evidence

If a form is described as British, American, Singaporean, Indian or another variety, check evidence specific to that population and register.

8. Institutional evidence

If the task belongs to an examination, journal, school, employer or publisher, record the current governing rule separately.

9. Style-guide evidence

Record recommendations as recommendations. Note which guide, edition or institution they belong to.

10. Conflicting evidence

Write the conflict explicitly:

Dictionary A labels the form informal; Dictionary B gives no label; corpus evidence shows strong spoken use and very limited academic use.

This sentence is more useful than pretending the sources agree.

11. Decision

Write the recommendation for the target context:

Use X in the formal examination. Recognise Y as a common conversational alternative.

or:

Both forms are established; follow the publication’s American spelling standard.

12. Confidence

Classify the conclusion internally:

• High confidence: strong agreement across relevant sources.
• Moderate confidence: clear pattern but some variation or limited data.
• Low confidence: sparse, conflicting or rapidly changing evidence.

Confidence is not a public score of truth; it reminds the writer how forcefully to phrase the claim.

13. Review trigger

Specify what should make SETC revisit the dossier:

• new examination specification;
• new dictionary treatment;
• new corpus release;
• major usage shift;
• new linguistic research;
• change in institutional style.

Worked dossier: “different than”

Question: Is different than incorrect?

Context: formal world-facing prose.

Dictionary/grammar evidence: major contemporary references document several constructions involving different, with regional and syntactic variation.

Corpus evidence: different from is widespread; different than is well attested, especially in American usage and in constructions where a following clause can make than convenient.

Institutional evidence: a specific style guide may prefer different from.

Decision: do not describe different than as nonexistent English. Use different from as a broadly safe formal default where no syntactic reason favours another construction, and follow the target style where one is specified.

Confidence: high on the existence/variation distinction; target preference remains context-dependent.

Worked dossier: a new AI phrase

Question: Has a new AI-related term become established English?

Dictionary evidence: perhaps absent or newly included.

Corpus/news evidence: check sustained use across independent sources, not one vendor’s marketing.

Technical evidence: check whether specialists use the term consistently or whether several competing labels exist.

Historical evidence: archive early examples cautiously.

Decision: if usage is still unstable, define the term on first use and avoid claiming universal acceptance.

Why the dossier matters for a ten-year website

SETC will eventually contain thousands of statements about English. Some will be stable for decades; others will age. A dossier allows future editors or AI systems to understand why a wording was chosen and what evidence would justify changing it.

The same method can work across the federation. eduKateSingapore can use institutional evidence for Singapore examination terminology; eduKateSG can use domain references for adult and career applications; local sites can inherit the deep English verdict without recreating the investigation.

This is how evidence becomes infrastructure.

The minimum sufficient evidence test

Good reference work should be strong enough for the decision and no larger than necessary. Before stopping, ask five questions.

1. Meaning: Have I identified the correct sense or structure?
2. Distribution: If I am making a claim about frequency, register, region or period, do I have evidence from the relevant population?
3. Authority: If an institution controls the rule, have I checked the current primary source?
4. Conflict: Have I noticed any credible source that would materially change the conclusion?
5. Transfer: Can I now use or explain the pattern without reopening the reference?

If the answer is yes to all five, more searching may add volume without changing the decision. Stop.

If one answer is no, gather the specific missing evidence rather than starting the investigation again from zero.

Too little evidence

One remembered rule, one Google example or one AI answer is often too weak for a broad claim.

Too much evidence

Twenty dictionaries are unnecessary for a basic uncontested sense. Reference literacy includes efficiency.

The right amount

The correct evidence burden grows with the claim. “This dictionary labels the word informal” needs one accurate dictionary citation. “This form is becoming more common across global academic English” needs substantially more evidence across time, region and register.

This proportionality rule protects both learners and SETC from two opposite failures: careless certainty and endless research.

Ask only as much of the evidence as the evidence can carry—and gather only as much evidence as the real decision requires.

Apply reference literacy to vocabulary and grammar

These existing SETC lessons turn the verification system above into repeated language decisions.

• Lesson 014 — Learner Dictionaries & Corpora — investigate senses, grammar, collocations, register and frequency.
• Lesson 003 — Collocations, Lexical Bundles & Phrase Frames — move from single words to attested phraseology.
• Lesson 022 — Colligation & Verb Patterns — verify grammar frames when near-synonyms change the syntax.
• Lesson 023 — Semantic Prosody — use corpus evidence to detect hidden evaluative patterns.
• Lesson 028 — Intensity & Scalar Meaning — calibrate degree and avoid overstatement with usage evidence.
• English Prepositions — connect grammar rules with collocation and lexical frames.