Vocabulary: From Recognition to Contextual Use

Vocabulary is not a set of English = translation pairs. Knowing a word or chunk involves recognising its sound and form, understanding common senses, knowing collocation and register, and producing it in an appropriate setting.

Remove Two Misleading Equations

Earlier editions claimed that 1,000 words give 75% understanding and 7,000 nearly 90%. Such equations mix counting units, corpora, and definitions of comprehension, so this edition removes them.

Before discussing vocabulary size, ask:

• Are we counting word forms, lemmas, or word families?
• Is the material conversation, fiction, news, academic prose, or specialist documentation?
• Are proper nouns and transparent compounds included?
• Does “know” mean recognition or productive use across skills?

Research by Nation and others often discusses 95% and 98% lexical coverage as reading conditions. At 95%, roughly one running word in twenty remains unfamiliar; at 98%, about one in fifty. The higher figure generally supports more independent comprehension, but no fixed vocabulary count guarantees understanding. Background knowledge, text structure, language distance, and the task still matter.

Four Kinds of Vocabulary Evidence

Dimension	Question	Evidence
Receptive listening	Can I segment and understand it in speech?	Hear a sentence without captions and explain it
Receptive reading	Can I understand it from a new context?	Explain its function in a new article
Productive speaking	Can I retrieve it quickly with natural collocation?	Use it naturally in a recording
Productive writing	Can I spell it and choose the right register?	Use it accurately in a new task

Feeling familiar after seeing the answer is only a small part. Receptive knowledge often precedes productive knowledge, so choose the target from the real task.

Word Families Are a Research Tool

help, helps, helped, helpful, helpless may count as one family in some estimates, but learners do not automatically know every derived form. Families help estimate coverage; they do not replace learning morphology, part of speech, and use.

The useful unit is often larger than one word:

• collocation: pose a risk, meet a deadline;
• chunk: What I mean is ..., It depends on whether ...;
• morphology: analyse → analysis → analytical;
• register: kids, children, and minors differ by context.

Do Not Classify Visual and Auditory Learners

Evidence does not support improving learning by matching instruction to fixed visual or auditory labels. Vocabulary itself needs multiple kinds of evidence:

1. hear pronunciation and connected speech;
2. see spelling, morphology, and a real sentence;
3. say it, record it, and compare stress and sounds;
4. write an original sentence or short text;
5. use it in varied but related contexts.

Images suit concrete concepts; definitions and contexts suit abstract ones; pronunciation requires sound. Representation follows the content, not a learner identity.

Select What to Learn

Prioritise items that:

1. recur in current material or work;
2. block the main idea or task when unknown;
3. form useful collocations with known vocabulary;
4. have a clear use in the next two weeks;
5. are essential terms in a specialist task.

Temporarily skip one-off low-frequency detail that does not block meaning.

A High-quality Card

The front should demand retrieval, not recognition.

Context: how do I say that a dependency creates risk for a schedule?
Gap: The dependency may ____ a risk to the schedule.

The back includes:

• pose a risk to;
• one trustworthy source sentence and citation;
• pronunciation or audio;
• a short sense and register note;
• your second example;
• a contrast such as cause versus pose.

Long cards turn review into rereading. Test one decision per card when possible.

Performance-adjusted Spacing

Forgetting is real, but there is no universal review timetable. In Anki or another spaced-repetition tool, let performance control the interval:

• fast, accurate recall plus new use: lengthen it;
• effortful success: keep or slightly lengthen it;
• repeated confusion: add contrast and a new context;
• repeated failure: reduce new cards, split the card, or repair pronunciation/concept foundations;
• correct cards but no real-world use: stop adding volume and switch to speaking or writing.

Anki is one tool, not the method. An old third-party “Macmillan 7000” deck shared here is no longer distributed because its link, maintenance, and redistribution rights cannot be kept current.

Practice in Real Context

For each group, complete at least two transfers:

• alter the people, time, viewpoint, or domain in the source sentence;
• contrast a natural use with a typical misuse;
• give a 60–90 second retelling without the list;
• find three collocations in independent real sources;
• use the chunks in an email, explanation, log, or short essay.

Prefer learner dictionaries for pronunciation, collocation, and examples. AI can generate candidates, but verify definitions, collocations, and examples against dictionaries or real corpora.

A Thirty-minute Session

1. Five minutes: select 5–8 valuable chunks from real material.
2. Eight minutes: check pronunciation, sense, collocation, and register.
3. Seven minutes: create short retrieval cards.
4. Seven minutes: speak or write a different context without the source.
5. Three minutes: record errors and the condition for the next check.

Daily new-word count is not the outcome. Understanding and using items in a new task a week later is.

If you are unsure where the bottleneck is, use the Vocabulary Audit Template to record listening, reading, speaking, writing, and delayed transfer separately.

Seven Days, Thirty Days, Twelve Weeks

Seven days

• collect 20–30 chunks from one real topic;
• include sound, collocation, and source for each;
• complete at least two closed-book retrievals;
• on day seven, record two minutes or write 200 words with the set.

Thirty days

• use parallel material weekly to test listening or reading coverage;
• complete one real output with the week's chunks;
• delete low-value, duplicate, or cue-dependent cards;
• compare comprehension, production, and retention rather than card totals.

Twelve weeks

• build a small corpus around one professional or life domain;
• use high-frequency chunks in meetings, articles, code review, exams, or presentations;
• sample delayed recall and transfer to unfamiliar contexts;
• expand domains only after output becomes stable.

Feedback Criteria

For each target chunk, ask whether you can:

• recognise it in speech and text;
• explain the current sense in your own words;
• use natural collocation, form, and register;
• produce it without a cue;
• reuse it in a new context one week later.

Sources

• Hu & Nation (2000), Unknown Vocabulary Density and Reading Comprehension
• Nation (2006), How Large a Vocabulary Is Needed for Reading and Listening?
• Schmitt, Jiang & Grabe (2011), The Percentage of Words Known in a Text and Reading Comprehension
• Pashler et al. (2008), Learning Styles: Concepts and Evidence

Previous: Learning Principles | Next: Listening