The Best Way to Memorize Vocabulary: The Swahili Experiment That Settled It
Updated: 1 day ago
Most people memorize vocabulary the same way: look at the word, look at the meaning, repeat. Rereading a list feels productive because the words become familiar. A well-known experiment tested whether familiar actually means remembered, using foreign-language vocabulary, and the answer was one of the clearest results in learning research.
The dataset: 40 Swahili words and four ways to practise them
In 2008, Jeffrey Karpicke and Henry Roediger published The Critical Importance of Retrieval for Learning in Science (full text here). College students learned 40 Swahili-English word pairs, such as mashua and boat. Everyone alternated study periods (see the pair) with test periods (see mashua, recall the English).
The twist came once a student recalled a word correctly the first time. From then on, that word was handled one of four ways:
Kept in both study and testing (the standard method)
Dropped from study, but still tested
Still studied, but dropped from testing
Dropped from both, the "learn it, then move on" approach many study guides recommend
A week later, everyone took a final test.
What happened
During the learning phase, the four groups looked identical. Their learning curves showed no differences, and nearly everyone recalled all 40 words by the end. Asked to predict how many they would remember in a week, the groups also gave similar answers, between 20.3 and 22.0 words, or about 50 percent.
The final test was not similar at all:
The two groups that kept being tested recalled about 80 percent of the words.
The group that kept studying but stopped being tested recalled 36 percent.
The group that dropped words entirely recalled 33 percent.
The paper reports that the score ranges did not even overlap. Students who stopped being tested scored between 10 and 60 percent. Students who kept being tested scored between 63 and 95 percent. The effect size was d = 4.03, which the authors describe as about four standard deviations.
The computation
Turning the percentages into words out of 40:
Kept testing: 0.80 × 40 = 32 words
Kept studying only: 0.36 × 40 = 14.4 words
Dropped entirely: 0.33 × 40 = 13.2 words
So repeated recall gave 32 ÷ 14.4 = 2.2 times as many words as repeated studying. Students who reread every word in every study period ended up only about one word ahead of students who stopped practising those words altogether (14.4 versus 13.2).
The paper also counts the work. The test-only group did an average of 236.8 trials, and the study-only group did 243.0, essentially the same effort. Same effort, wildly different results. The authors note that adding about 80 extra study trials produced practically no gain in retention, while adding about 80 extra test trials led to improvements of more than 150 percent.
One more number worth noticing: the students predicted about 50 percent recall in every group. The group that only studied overestimated (36 percent actual). The group that kept being tested underestimated (80 percent actual). People could not feel which method was working.
What this means for learning a language
Producing beats recognizing. The students who did better were the ones who had to pull the word out of memory, repeatedly, even after they already knew it. Rereading a list, or recognizing a word in a multiple-choice question, is closer to the studying condition.
Do not drop words once you know them. The "learn it, then move on" group did worst. A word you recalled once is not finished; the extra recalls are what made it stick a week later.
Trust results, not the feeling. The students' own predictions were useless. The familiar, easy feeling of rereading was misleading.
Caveats: this was one experiment with college students, isolated word pairs and a one-week test. Using words in real sentences is harder and richer than recalling a translation. Still, the gap between 80 and 36 percent is large enough that recall over review is a sensible default.
Recall practice without an app
The easiest way to practise recall is to answer questions in the language, in your own words, every day. You cannot answer without pulling words out of memory, and you keep using the same common words over and over, which is exactly the "keep testing" condition.
Language Textapp does this by text. Once a day, at a time you choose, it sends you one short message in the language you are learning, at your level. You answer in the language. It replies with your corrected sentence, one or two fixes in plain English, and a follow-up that asks you to produce a little more:
App: ¿Qué comiste hoy? (What did you eat today?)
You: Yo comí tacos en la almuerzo
App: ✏️ Yo comí tacos en el almuerzo. Use 'el almuerzo' (masculine). ¡Qué rico! ¿Te gustaron? (Did you like them?)
Beginners get the English meaning in parentheses. For languages with a non-Latin script, such as Japanese, Korean, Mandarin or Russian, beginners and intermediates get the romanization and can answer in romanization. When you are stuck, you can ask it how to say something, then use the word in your next reply.
It is an AI, not a person or a teacher, and it can make mistakes. There is no audio or voice, so this trains reading and writing, not listening.
Try it by text
Text Hi to (385) 525-5598. It asks which language you are learning and whether you are a beginner, intermediate or advanced, then sends your first message. Answer it in the language. After that, a new message arrives each day at 9am, or at a time you ask for.
It works on any phone that can text, with no app and no account. Your first 20 messages are free, then $9.99/month covers every Textapp. Text STOP at any time to end messages. Message and data rates may apply. More at textapp.co/apps/language.
Comments