A vocabulary page headed fruit often places peach, plum, apricot, nectarine and persimmon in one neat box. The organisation is attractive. For a learner meeting all five forms for the first time, the box can become a source of interference: the meanings overlap, the labels compete, and every answer feels vaguely edible. New words with close meanings should usually gain separate footholds before they are compared as a set.
Imagine a reader who is a Taiwanese secondary student who can memorise 23 English-Chinese pairs for Friday and then swaps three of them during the quiz. The problem may sit in the design of the list, not in the student's effort.
A semantic set creates competitors
Words in a semantic set share a superordinate category. Fork, knife and spoon are utensils. Aunt, cousin and niece are relatives. Coursebooks group them because one picture or substitution drill can contain the whole set. The grouping is convenient for teaching materials. Memory receives a harder discrimination problem.
Thomas Tinkham tested this in a 1993 experiment using artificial second-language forms paired with familiar meanings. Twenty English-speaking participants needed more learning trials for semantically related sets than for unrelated sets. Robert Waring later replicated the design with 20 Japanese-speaking learners and again found an advantage for unrelated groupings. The ERIC record for Waring's replication gives the journal details and central result; Tinkham's original study is indexed under DOI 10.1016/0346-251X(93)90027-E.
These studies used small samples and artificial words. They isolate the mapping problem well, though they cannot reproduce a full English lesson with context, prior knowledge and teacher explanation. Their result should change a default, not create a ban on categories.
Unrelated does not mean random noise
A better first set might contain peach, borrow, narrow, whisper, shelf and damp. Each item can attach to a distinct picture, action or sentence. The words still belong to one lesson if the lesson supplies a meaningful situation. A story about borrowing a book from a narrow shelf in a damp room provides thematic connection without asking six near-neighbours to compete for one semantic space.
This distinction between semantic and thematic grouping matters. A theme gathers words that can participate in the same event: ticket, platform, delayed, suitcase, announce. A semantic cluster gathers coordinate labels such as five kinds of luggage. Tinkham's later work found that thematic groupings could be easier than tightly related semantic sets. A shared situation gives each word a different job.
I would keep the situation and loosen the category. A train-station task can include a noun, a verb, an adjective and a phrase. That mixture gives the learner more retrieval routes than a column of eight near-synonyms.
Classroom evidence points the same way
Erten and Tekin taught 80 selected words to 60 fourth-grade learners, alternating semantically related and unrelated sets, and the learners recalled more from the unrelated sets on immediate and delayed measures while also taking longer to complete tests of related items. The authors report the design, timing and limits in their System article on semantic-set presentation.
One later experiment examined spacing as well as relatedness. The researchers asked whether separating encounters could reduce competition between similar words. Their study of massing, spacing and semantic relatedness reviews six earlier studies that found costs from semantic clustering, then tests how presentation schedules alter the pattern. This line of work warns against a simple slogan. Prior knowledge, spacing, task type and the exact relation among words all matter.
The safe classroom inference is modest: avoid introducing a whole family of unfamiliar coordinate terms in one massed block.
Give each word its own address
Each new word needs a specific first encounter. Teach niece through one family diagram and one sentence: “Mei's brother has a daughter. The girl is Mei's niece.” Retrieve it later from the diagram. Teach cousin in another context, perhaps a wedding invitation, on another day. Once both forms can be recalled, place them side by side and ask learners to explain the difference.
That comparison stage has value because English kinship terms divide family relations differently from 華語, so learners eventually need a map, yet the map works better after each label has an address. Before that point, a five-branch family chart can generate five weak links at once.
Near-synonyms need the same care. Introduce borrow inside a request, “Could I borrow your charger until 4:17?” Bring in lend after borrow is available from meaning. Then contrast the direction of transfer with two arrows and two speakers. The contrast resolves an existing distinction. It no longer asks the learner to build both word forms, both meanings and the relation between them simultaneously.
Retrieval should expose the mix-up
Recognition can hide interference. A multiple-choice task containing one target and three unrelated distractors may look easy even when the learner confuses plum and prune. Use a test that makes the close competitors visible.
Show four fruit pictures without labels and ask for the English. Give one English word and ask learners to choose the exact picture. Put borrow and lend into two short dialogues and require a reason for each choice. Record confusions as pairs rather than marking only right or wrong: niece → cousin tells the teacher more than a red cross.
Then space the troublesome pair. Retrieve one member after 9 minutes, the other near the end of class and both at the start of the next lesson. If the learner still crosses them, return to two distinct contexts before another direct comparison.
This approach fits Choc Education's advice that a vocabulary note should record the words that travel together. Collocations connect a word to its usual partners, such as borrow money or lend support. Coordinate lists connect several possible answers to the same category cue. Those are different networks and deserve different teaching choices.
When a semantic set is useful
Known words can be organised together. If a learner already uses angry, annoyed and furious, a scale can refine intensity and register. Technical study may also require a category because the category itself carries the concept. A biology learner needs artery, vein and capillary in a connected system. The teacher can reduce interference by preteaching the labels separately, using diagrams and returning to them across several sessions.
Very small contrasts can also earn immediate comparison. Two words with a common, damaging mix-up may need to appear together so the learner sees the boundary. The load grows when the page introduces seven unknown neighbours at once.
Before assigning the next單字表, circle every item that is completely new. If four or more circled words share one narrow category, separate their first encounters. Give each a sentence, image or action of its own. Bring the set together only after learners can retrieve the members individually. The tidy box belongs at the end of learning, when peach no longer turns every other orange-coloured fruit into a guessing game.