If you’re going to teach vocabulary, you need to organise it in some way. Almost invariably, this organisation is topical, with words grouped into what are called semantic sets. In coursebooks, the example below (from Rogers, M., Taylore-Knowles, J. & S. Taylor-Knowles. 2010. Open Mind Level 1. London: Macmillan, p.68) is fairly typical.
Coursebooks are almost always organised in a topical way. The example above comes in a unit (of 10 pages), entitled ‘You have talent!’, which contains two main vocabulary sections. It’s unsurprising to find a section called ‘personality adjectives’ in such a unit. What’s more, such an approach lends itself to the requisite, but largely, spurious ‘can-do’ statement in the self-evaluation section: I can talk about people’s positive qualities. We must have clearly identifiable learning outcomes, after all.
There is, undeniably, a certain intuitive logic in this approach. An alternative might entail a radical overhaul of coursebook architecture – this might not be such a bad thing, but might not go down too well in the markets. How else, after all, could the vocabulary strand of the syllabus be organised?
Well, there are a number of ways in which a vocabulary syllabus could be organised. Including the standard approach described above, here are four possibilities:
1 semantic sets (e.g. bee, butterfly, fly, mosquito, etc.)
2 thematic sets (e.g. ‘pets’: cat, hate, flea, feed, scratch, etc.)
3 unrelated sets
4 sets determined by a group of words’ occurrence in a particular text
Before reading further, you might like to guess what research has to say about the relative effectiveness of these four approaches.
The answer depends, to some extent, on the level of the learner. For advanced learners, it appears to make no, or little, difference (Al-Jabri, 2005, cited by Ellis & Shintani, 2014: 106). But, for the vast majority of English language learners (i.e. those at or below B2 level), the research is clear: the most effective way of organising vocabulary items to be learnt is by grouping them into thematic sets (2) or by mixing words together in a semantically unrelated way (3) – not by teaching sets like ‘personality adjectives’. It is surprising how surprising this finding is to so many teachers and materials writers. It goes back at least to 1988 and West’s article on ‘Catenizing’ in ELTJ, which argued that semantic grouping made little sense from a psycho-linguistic perspective. Since then, a large amount of research has taken place. This is succinctly summarised by Paul Nation (2013: 128) in the following terms: Avoid interference from related words. Words which are similar in form (Laufer, 1989) or meaning (Higa, 1963; Nation, 2000; Tinkham, 1993; Tinkham, 1997; Waring, 1997) are more difficult to learn together than they are to learn separately. For anyone who is interested, the most up-to-date review of this research that I can find is in chapter 11 of Barcroft (2105).
The message is clear. So clear that you have to wonder how it is not getting through to materials designers. Perhaps, coursebooks are different. They regularly eschew research findings for commercial reasons. But vocabulary apps? There is rarely, if ever, any pressure on the content-creation side of vocabulary apps (except those that are tied to coursebooks) to follow the popular misconceptions that characterise so many coursebooks. It wouldn’t be too hard to organise vocabulary into thematic sets (like, for example, the approach in the A2 level of Memrise German that I’m currently using). Is it simply because the developers of so many vocabulary apps just don’t know much about language learning?
References
Barcroft, J. 2015. Lexical Input Processing and Vocabulary Learning. Amsterdam: John Benjamins
Nation, I. S. P. 2013. Learning Vocabulary in Another Language 2nd edition. Cambridge: Cambridge University Press
Ellis, R. & N. Shintani, N. 2014. Exploring Language Pedagogy through Second Language Acquisition Research. Abingdon, Oxon: Routledge
West, M. 1988. ‘Catenizing’ English Language Teaching Journal 6: 147 – 151
Adaptive Learning in ELT 
Nation also suggests -advocates in fact- a fifth way of organising vocabulary – teach words in frequency order.
Thanks, Jill. In coursebooks, typically, frequency informs other considerations, and us writers usually have to justify the inclusion of any lexical items with reference to frequency lists. Apps, on the other hand, are another matter.
Frequency, however, is problematic. It’s not an exact science: see, for example, the differences between the Oxford 2000 list and the Macmillan 3-starred words (top 2500). Because of this, the best we can really do is to divide words up into frequency tranches of about 500. But even that doesn’t always help. If you take COCA, for example, as your source corpus, ‘apple’ comes in at #1983 while ‘orange’ comes in at #3342. Does this mean that ‘apple’ should be taught so much earlier than ‘orange’?
There’s also the question of which corpus should be used to calculate frequency. Should it be a native-speaker corpus like COCA? If so, which native-speaker corpus? Should it be an ELT corpus? Or should it be a learners’ corpus, such as the one that is behind English Vocabulary Profile? These are not easy questions to answer – much will depend on the objectives of particular learners.
But, even when that is decided, there’s still another problem. Most frequency lists refer to dictionary headwords, but lemmas might make more sense, and word senses would make even more sense (since the primary goal of learners must be to learn word senses rather than ‘words’). But frequency lists rarely differentiate word senses.
If we are to go for word senses, frequency will presumably influence the way we select and organise learning items. But frequency alone is not the only consideration. The most frequent sense of a particular word may be not be its prototypical meaning, which is probably more useful for learners to learn first (since, if you have the prototypical meaning of a word, you can likely work out other meanings – but not the other way round!).
So, frequency is important, but, for the reasons above (and for quite a few others I can think of), it will rarely suffice as the sole organising principle for vocabulary selection and organisation.
Thanks for that, Philip – that vocabulary ‘myth’ seems to be one that is most resistant to change, perhaps because it seems extremely counter-intuitive, and/or because there is such a long tradition of pedagogy associated with it.
However, I don’t think we should take it completely on trust. Even Keith Folse (in Vocabulary Myths 2004), while problematizing this myth, notes that the classic studies which are always cited in this regard, i.e. the two by Tinkham and the one by Waring, ‘used nonsense words as the target words’, matched against ‘equivalents’ in the test subjects’ L1, i.e. ‘none of the studies used real L2 words’. He comments ‘while the use of imaginary words increased the internal validity of the experiments by allowing the researchers to control for the extraneous variable of previous knowledge of the target vocabulary items, it remains to be seen what results will be obtained with L2 students learning L2 words… Future research must be conducted with actual L2 students in actual L2 words to confirm the experimental results’ (p. 55).
Another criticism of these and other more recent studies is that they seldom test long-term recall.There is a suggestion that unrelated vocabulary sets work better for short-term recall, while semantic sets may have more durable effects, not least because they reflect the way that lexical items are stored in long-term memory.
I also have a slight problem with the notion of ‘thematic sets’, particularly in so far as they are different from ‘sets determined by a group of words’ occurrence in a particular text’. How do you define what words belong to a thematic set? One way might be through word association tests (what words do you associate with the word ‘bee’? for example). But these can be highly subjective. Also, when asked to make an association with a word like for example generous, respondents often reply with a word that’s semantically related (a synonym or an antonym) which is exactly the kind of relationship that these studies warn us against.
Another might be by consulting a corpus. But how? One way might be to look for collocations. The largest span available for searching a word’s collocates on COCA is nine words on either side of the node. Here, for example, are the most frequent collocations of the word ‘bee’ (ignoring proper nouns), using COCA:
pollen, bee, sting(s), balm, honey, colonies, species, cave, hive, flower
And, for good measure, for the plural ‘bees’:
honey, bee(s), wasps, butterflies, hive(s), pollen, flowers, pollinate, sting, insects, nectar
This method is obviously limited, though because of the limitations of the span (effectively only 19 words long).
Or you could search a corpus of texts that share a common theme. Any one text in the corpus will include at least some of these words, presumably. But even here, it’s not clear what texts you should include in your corpus and what to leave out.
Here, for example are the keywords (i.e. the words that are statistically significantly frequent compared to a reference corpus) in the Simple Wikipedia entry for ‘bee’:
pollinate wasp insect pollen ancestor sting evolve colony honey queen flower special body collect
And here are the keywords in Sylvia Plath’s poem sequence on bees:
midwife hive swarm veil ivory virgin grove dumb honey mass village queen square flower sweet yellow wing black cell
Note that there is some cross-over (honey, flower) but not a lot. It seems that, as in prototype theory, some words are more ‘typical’ in terms of their associations while others are more peripheral. And, of course, frequency will play a part: words like honey and flower are much more frequent than swarm and hive. Hence, these may be better candidates for selection. Nevertheless, the problems with deciding criteria for thematic set status are not to be underestimated.
It’s little wonder, therefore, that coursebook writers (and app designers) play safe by choosing semantic sets which, for all their problems, are more intuitive and relatively easy to compile.
Thanks, Scott.
All good and interesting points. Perhaps, the Interference Theory is at its most convincing when we consider form-oriented sets, such as the following from ‘English Vocabulary in Use Upper Intermediate and Advanced’ (p. 172-173): bring sb up, bring sth off, bring on sth, bring sth about, bring sth back, bring sth down, bring sth out, bring sb round to sth ?
Yes, Philip, regarding interference, Folse cites one study (Olsen 1999), one conclusion of which ‘was that external factors such as teaching confusing pairs such as sea and see, by and buy, wantand won’t, or loseand loose at the same time actually causes errors.’ By extension, sets of minimally differentiated phrasal verbs must be a recipe for disaster.
How does this apply to studies of lexical/semantic priming that suggest you’re more likely to recall (positively or falsely) a word that belongs to a set of semantically similar words?
Eg tired, dream, bed, nap, cozy (you’d be excused for telling me at a later point in time that I’d showed you the word sleep).
Do these not suggest that facilitating semantic relationships between concepts in the classroom directly mimics our tendency to recall words with related meaning, whether this relationship is subjective or not?
It might also suggest that interference, like making mistakes or falsely recalling words is a normal symptom of learning and recall processes, even if that undermines the initial retention of words in the short term?
My recollection of Hoey’s ‘Lexical Priming’ is that he argues that ‘priming’ (the way that we use words is shaped by our experience of the way they are used) is primarily about collocational and syntactic relationships, rather than semantic ones. But maybe I have got this wrong, and maybe you’re referring to someone else.
Relevant to your comments, I think, is the work of Paul Meara (e.g. ‘Connected Words’, 2009) and his study of word associations in the mind. He discusses early studies of word associations, where these were classified as syntagmatic (syntactical associations), paradigmatic (semantic associations) and klang (phonological associations). If I remember rightly (and I’m not entirely sure that I do), L1 speakers typically display klang associations (‘hairy’ – think ‘fairy’ or ‘airy’) as young children, gradually progress towards making more syntagmatic associations (‘hairy’ – think ‘monster’ or ‘chef’) and later produce more paradigmatic associations (‘hairy’ – think ‘bald’ or ‘shaggy’). L2 learners, as they develop their L2 competence, typically follow a similar pattern. If this is so, it would suggest that semantic relationships between words (words with related meanings and the same part of speech) may be less important, in the early stages of language acquisition, than phonological and syntactic mapping.
Regarding interference, we may be talking at cross purposes. I wasn’t referring to cross-language (transfer) effects, but the idea that if words are very similar to each other (semantically or formally), they may be more difficult to learn. In that sense, it’s not a symptom of memorisation processes, but a problem with input that can, to some extent, be minimised. That said, though, I can see that short-term problems might lead to longer-term gains.
Yes I see what you mean. I suppose it’s a bit like the way we tend to introduce semantic range of lexis at later levels of study, isn’t it?
Yes, I think there’s a parallel there.
[…] How best to teach vocabulary? Organized by semantic set? By theme? This article has some interesting research:"The answer depends, to some extent, on the level of the learner. For advanced learners, it appears to make no, or little, difference (Al-Jabri, 2005, cited by Ellis & Shintani, 2014: 106). But, for the vast majority of English language learners (i.e. those at or below B2 level), the research is clear: the most effective way of organising vocabulary items to be learnt is by grouping them into thematic sets (2) or by mixing words together in a semantically unrelated way (3) – not by teaching sets like ‘personality adjectives’. It is surprising how surprising this finding is to so many teachers and materials writers. It goes back at least to 1988 and West’s article on ‘Catenizing’ in ELTJ, which argued that semantic grouping made little sense from a psycho-linguistic perspective. Since then, a large amount of research has taken place. This is succinctly summarised by Paul Nation (2013: 128) in the following terms: Avoid interference from related words. Words which are similar in form (Laufer, 1989) or meaning (Higa, 1963; Nation, 2000; Tinkham, 1993; Tinkham, 1997; Waring, 1997) are more difficult to learn together than they are to learn separately. For anyone who is interested, the most up-to-date review of this research that I can find is in chapter 11 of Barcroft (2105)." […]