Teaching Recognition Memory with a Word Memory Test: A 45-Minute Class Plan

A word memory test is one of the best ways to show, hands-on, how memory is measured in a cognitive psychology, educational psychology, or learning science class. Words appear one at a time, and students only decide whether each word is new or one they have already seen. Three mistakes end the run, and the score is how many words they lasted. It takes about five minutes, needs no sign-in, and with a group link every student's result lands in an anonymous distribution you can put on screen right away. This guide covers what the test measures, how to run a 45-minute class, and which numbers from which papers to cite.

What it measures

This is a continuous recognition task, a design introduced by Shepard and Teghtsoonian (1961): there is no separate study phase and test phase. As the stream of words goes by, about half are repeats of earlier words (2 to 50 words back), and students judge each one as 'seen' or 'new'. The more words that have come in between, the harder a repeat is to recognize, and the accuracy-by-lag chart on the result screen shows exactly that.

Besides the score, the result screen shows six things: (1) hit rate, the share of repeated words answered 'seen'; (2) false-alarm rate, the share of new words answered 'seen'; (3) discrimination = hit rate − false-alarm rate; (4) the overall share of 'seen' answers (response bias); (5) mean response time; and (6) accuracy by lag. The score alone cannot tell 'good memory' apart from 'pressing seen for everything', so signal detection theory looks at two separate axes, discrimination and bias. The 'hit rate minus false-alarm rate' measure recommended by Snodgrass and Corwin (1988) for recognition memory is exactly this discrimination index.

Running a 45-minute class

1. Before class: use 'Create a group link' on the test page (test = word memory, post-test = in 4 weeks) and post the link in the class chat or on a slide. Students just tap the link; no sign-in. If devices are shared, as in a computer lab, have each next student start with the 'Start next participant' button on the result screen so they count as a new participant.
2. Warm-up, 5 minutes: show 10 words for 30 seconds, hide them, and have students write down what they remember (recall). Then show 20 words and ask which ones were on the list (recognition). Most students find recognition far easier. Today's question: how is pulling a memory out different from recognizing it?
3. The test, about 5 minutes: everyone works alone in quiet. Have them silence sounds and notifications and leave the result screen open when they finish (each score is visible only to its owner).
4. Show the distribution, 5 minutes: put the group hub page on the projector to show the anonymous histogram, median, and number of participants. Agree first that no one reveals where they fall.
5. Concept lecture, 15 minutes: signal detection (hits, false alarms, bias), forgetting as seen in accuracy by lag (is it the passing time or the words in between?), the picture superiority effect, and the testing effect. Use the numbers under 'Key studies' as they are.
6. Discussion, 10 minutes: pick two of the questions below. It helps to point out, in the real distribution, how the same score can hide different discrimination and bias.
7. Wrap-up, 5 minutes: announce the post-test in 4 weeks. Say up front that people usually do a little better the second time just from familiarity with the task (a practice effect); on post-test day, 'did our memory really improve?' becomes the second lesson.

Key studies: numbers to use in class

• Shepard & Teghtsoonian (1961): the original continuous recognition task. Without separating study from test, it asks 'seen before?' within a stream, and showed that the more items come between the two appearances, the lower the chance of recognition. The accuracy-by-lag chart in this test draws the same picture.
• Shepard (1967): right after seeing several hundred items once each, people chose the old item in two-alternative recognition tests 98% of the time for pictures, 90% for words, and 88% for short sentences (as summarized by Standing, 1973). Our ability to recognize something seen once is remarkably large.
• Standing (1973), 'Learning 10,000 pictures': after viewing 10,000 pictures over five days, people averaged 27.2 errors on 160 two-alternative trials (about 83% correct), an estimated 6,600 pictures retained. In the same paper, after 1,000 items the mean errors on an 80-trial test were 9.2 for pictures and 15.4 for words: the picture superiority effect.
• Brady, Konkle, Alvarez & Oliva (2008): after viewing 2,500 object photos for 3 seconds each over 5.5 hours, people were 92% correct when the old object was paired with an object of a different kind, 88% against a different object of the same kind, and still 87% against the same object in a different state (say, a full cup versus a half-full one). Recognition memory holds detail, not just gist.
• Ebbinghaus (1885): he measured forgetting by how much less time it took to relearn a list of nonsense syllables (savings). Savings were 58.2% after 20 minutes, 44.2% after 1 hour, 35.8% after about 9 hours, 33.7% after 1 day, 27.8% after 2 days, 25.4% after 6 days, and 21.1% after 31 days (Table 3 in Murre & Dros, 2015): steep forgetting on the first day, slow after that. Murre and Dros (2015) reproduced the curve with one person who spent 70 hours on it. Be sure to point out that these are not percentages remembered but percentages of relearning effort saved.
• Roediger & Karpicke (2006): after reading a short passage, one group read it again and another took a recall test (with no feedback). After 5 minutes rereading was ahead (81% vs 75%), but after 2 days the tested group led 68% to 54%, and after 1 week 56% to 42%. In the second experiment, one-week recall was 40% after four study periods, 56% after three studies and one test, and 61% after one study and three tests. Practicing retrieval holds memories longer: the testing effect.
• Snodgrass & Corwin (1988): a methods paper arguing that recognition memory should be scored not with a single percent correct but with 'hit rate minus false-alarm rate' (discrimination) and response bias, kept separate. It underlies the discrimination and bias measures on the result screen.

Discussion questions

• Two students both scored 60 words. One has a high hit rate and many false alarms; the other is low on both. Who 'remembered better'? What do we miss when we rank people by a single score?
• If accuracy drops at longer lags, is that because time passed or because other words got in between? How could we design an experiment to separate the two (decay versus interference)?
• If we can recognize about 6,600 of 10,000 pictures seen once, why is material we studied for an exam so hard to bring to mind? Explain it with the difference between recognition and recall.
• Rereading wins after 5 minutes and loses after a week (Roediger & Karpicke, 2006). So why do students prefer rereading? How does the feeling of knowing differ from actual memory?

Variations and extensions

• Prediction game: before the test, have students write down how many words they expect to last, then compare. How accurate our knowledge of our own memory is (metamemory) becomes the topic.
• Recall versus recognition: collect the warm-up results (words recalled and words recognized) on the board and compare the class averages. How far apart the two numbers are for the same list is a natural lead-in to the lecture.
• Mini testing-effect experiment: split the class; one half reads the same short passage twice, the other reads it once and then writes down what they remember on a blank sheet. A recall test in class a week later makes a small-scale replication of Roediger & Karpicke (2006). If your results differ from the original, why they differ is an even better discussion.
• Compare with visual memory: in the same class, make a second group link for the paired associates task (remembering where pictures are hidden) and show the verbal and visual distributions side by side.

• 📚Word memorySeen it before, or new? Keep the streak.Take the test →
• 🗃️Paired associatesWhat was where? Object-location memory.Take the test →
• 🔢Digit spanShort-term memory span for digits.Take the test →