Vocabulary Evidence Card: From Card Familiarity to Contextual Retrieval

Copy this card into a private note for one real topic or task. Do not include passwords, customer records, unpublished documents, exam answers, medical privacy, or unauthorised third-party material. A final word list may be public without making source material, recordings, and feedback public.

1. Define the Task and Evidence Conditions

# Vocabulary Evidence - YYYY-MM-DD

Real situation and audience:
Task: listen / read / speak / write / exam / hand over
Material, version, and source:
What should another person understand, decide, or complete?
Time and length limit:
Permitted captions / dictionary / translation / search / AI / people:
Privacy, copyright, confidentiality, exam, and expert boundary:
First-encounter date:
Parallel-task date:

The target is not "items added." It is understanding, choosing, and retrieving under these conditions.

2. Preserve a No-Lookup First Encounter

First-encounter start/end time:
Gist, relationships, and next step before lookup:
Details understood with confidence:
Places that required guessing:
Sentence or timestamp where meaning stopped:
Sense/expression I tried then:
Raw retelling, recording, cloze, or draft location:
Did it complete the real task? yes / partly / no

First check whether the barrier is lexical. Sound segmentation, syntax, background, layout, and attention can also break a sentence.

3. Build an Unknown-Item Decision Queue

Word/chunk	Task impact	Decision	Basis then	Next check
	blocking / important / minor / none	skip / infer / quick lookup / learn / expert check

Five to eight chunks entering deliberate learning:
Items to skip and why:
Items requiring a primary source or expert:
Items deferred because of capacity:

4. Record Eight Dimensions for Each Valuable Chunk

Copy once per chunk:

Word/chunk:
Task function:
Form, inflection, and derivation:
Sound, stress, and connected speech:
Current sense in my own words:
Common collocation or sentence slot:
Grammatical behaviour:
Register, politeness, stance, and identity:
Concept, referent, version, or domain boundary:
Current retrieval time and required cue:
One near-synonym/counterexample and difference:

Do not copy an entire dictionary page to fill the fields. Keep only what can change understanding or use in this task.

5. Verify Sense, Collocation, and Source

Item to check	My candidate	Source, version, location	Result	Unknown
Sense
Pronunciation
Collocation
Register
Technical/institutional concept

Why this dictionary/corpus fits this task:
Primary document / standard / code location:
Check date:
Who decides the high-risk term:
What evidence would change the decision:

AI examples and explanations are candidates, not sources. Frequency or confident model tone cannot decide high-risk terminology.

6. Design Unaided Retrieval

Real intention or situational cue:
Prompt allowed on first attempt:
Prompt removed next time:
Target retrieval time:
Speaking task:
Writing task:
Person / time / topic / tone / mode to change:

Attempt	Condition	Actual expression	Retrieval time	Result
Original context	Answer hidden
Changed context	Source sentence removed
Real task	Audience and consequence

Reproducing only the complete source sentence is not transfer.

7. Ask a Real Reader or Listener to Check It

Meaning retold by the reader/listener:
Relationship, certainty, and next step they heard:
Place that required guessing or confirmation:
Did the lexical choice change fact or responsibility?
Task completed? yes / partly / no
Real response or execution result:

Feedback	Type	Decision	Reason	Used in parallel task?
	necessary repair / acceptable variant / register / style	accept / partly / reject / defer

8. Record the Error and Minimum Repair

Error sample	Type	Likely gap	Change only this	Retest condition
	form / sound / sense / collocation / grammar / register / slow retrieval / transfer

Chunks correct on cards but failed in the task:
Chunks heard but not written:
Chunks read but not heard:
Acceptable variation mislabelled as error:
Items needing a split card, lower volume, or more background:

9. Record AI and Tool Use

Tool / model / version and date:
First encounter and candidate that existed before use:
Request: sense contrast / collocation candidate / cloze / counterexample
         / audience simulation / error classification / parallel task
         / other
Scope generated or rewritten by the model:
Content that may be invented or outdated:
Dictionary / corpus / primary source / human verification:
What I accepted, rejected, or deferred, and why:
Retrieval task after closing the tool:

10. Test Delayed Retention and Transfer

Time	Remove	Change	Output	Result
Same day	Answer and gloss	Original topic	Contextual recall
Day 1	Source sentence	Sentence and people	Spoken/written retrieval
Days 3-7	Recent cue	Mode or audience	Parallel input/output
Day 14	Old cards and feedback	New topic, same task	Timed real output
Day 30	All practice prompts	Real material/consequence	Full delivery

Chunks still retrieved in time:
Chunks successful only in the original item:
Most frequent returning error:
Did the real task require less guessing or rework?

11. Score and Decide What to Keep

Score each item 0-2: 0=no evidence/not completed, 1=cue-dependent or unstable, 2=independent under the current condition.

Item	Same day	Day 7	Day 14	Evidence reason
Recognise it in sound and text
Judge current sense and task impact
Use fitting collocation, form, and register
Retrieve without prompts under time
Deliver intended meaning to reader/listener
Transfer after topic, audience, or mode changes

Chunks to keep learning and why:
Chunks needing a split or new cue:
Chunks to archive:
Chunks requiring expert verification:
One condition to change next cycle:
Next review date:

Related chapters: Vocabulary: From Familiarity to Retrieval in Real Tasks | Listening | Reading | Speaking | Writing | Learning English with AI