Looking up is not learning

In 2001, Laufer and Jan Hulstijn broke down what a task does for a word into three things. You need the word, you search for its meaning, and you check whether the meaning you found fits the sentence. In 2021, Yanagisawa and Webb worked out across 42 studies which of the three parts explains how well a word is retained. Checking contributed the most, needing came next, and searching contributed nothing at all. So the click itself, the searching, is worth nothing. The learning happens in the moment after it, when you read the sentence again with the meaning in its place. Does it fit? Is ausgerechnet really a reproach here?

In 2009, Elke Peters et al. tested this with 137 Dutch-speaking students who read a German text with an online dictionary. Those who knew that a test was coming looked up more words. Those who had to work on the words again after reading could still recall their meaning later. And the words needed for the comprehension questions stuck better than those that could be skipped. The study concludes that the meagre yield of reading alone rises markedly when you have to look words up, connect the meaning thoroughly with the sentence and work on the word again after reading.

In 2010, Ana Pellicer-Sánchez and Norbert Schmitt had advanced learners read a whole novel and then tested them on 34 unknown words. From one to four encounters nothing happened, at five to eight the curve rose, and at ten to seventeen it jumped. Beyond that, more encounters added hardly anything. But even after ten to seventeen encounters, the readers recognised the meaning of 84 per cent of the words and could give it themselves for only 55 per cent. In 2018, Irina Elgort and colleagues saw the reason in eye movements. After five to ten encounters, the form of a new word is read as quickly as that of a known one. Fitting its meaning into the sentence still takes longer after twenty, sometimes forty encounters. A book gives you the common words. For the others, what counts is where else you meet them and whether they come with a context when you do.

The second mistake after looking a word up is looking at it again. You go through the list, nod and move on. In 2008, Jeffrey Karpicke and Henry Roediger showed in Science what that achieves. As soon as a word pair had been recalled correctly once, looking at it again did nothing for recall a week later, while testing it again did a great deal. The students themselves predicted the opposite. In the meta-analysis by Christopher Rowland from 2014, a test with a cue, where you have to bring up the word yourself, has the largest effect, and feedback afterwards almost doubles it. So you cover the German side, bring the word up from memory, and only then turn the card over.

In 2009, Nate Kornell had students learn vocabulary with flashcards, either as one large stack or as four small ones. The large stack won, and it also beat cramming on the last day. Even so, after the first session 72 per cent believed that the small stacks had worked better. In 2007, Karpicke and Roediger found that it hardly matters whether the gaps between the repetitions grow or stay the same. What matters is that the first test only comes when it is already hard. In 2015, Ulf Schuetze saw the same with beginners in a German course. Growing gaps won the test shortly afterwards and equal gaps won the later one, and the small function words in particular suffered most from the growing gaps. When an old word that you have almost forgotten turns up again, nothing has gone wrong with the plan, because that is how it is meant to work.

It is normal for this to feel wrong. In 2020, Aroline Seibert Hanson and Christina Brown required students in a Spanish course to use a program with spaced repetition. The students complained that words from chapters tested long ago were a waste of time. The number of days on which they used the program anyway predicted their results at the end of the semester, even after their starting level, motivation and beliefs were taken into account. And in the study by Yuichi Suzuki, Tatsuya Nakata and Xuehong He in 2025, what explained retention was not a memory test but how many attempts someone had needed for a word in the program, which accounted for 39 per cent of the differences against 2 per cent for the test.