Word Identification From Spelling to Sound
Van Orden, G. C., Johnston, J. C., & Hale, B. L. (1988). Word identification in reading proceeds from spelling to sound to meaning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(3), 371–386. https://doi.org/10.1037/0278-7393.14.3.371
Each trial shows a category, such as a part of the human body, then a target. Decide as fast as you can whether the target is a member of the category. Some targets sound like a member but are spelled differently: hare, or the nonword brane. They are not members, so the answer is no. Knowing this does not make them easy to reject when you have to answer quickly. Try it.
Run trials
Press F for yes (a member of the category) and J for no, or tap the buttons. Answer before the deadline; a late answer counts as too slow. After an error you will see what the target sounded like.
What the demo shows
In this task, “yes” means you have identified the target as a word whose meaning fits the category. A false “yes” to hare means the meaning of hair was activated, through the sound the two words share, strongly enough to drive the response before the spelling was checked. Compare your false “yes” rate for sound-alikes with the rate for spelling controls such as harp and surt, which are just as close in spelling to a category member but do not sound like one.
Van Orden and colleagues found false “yes” rates of about 21% for word homophones like hare and for nonword homophones like sute, against about 3% for their spelling controls. The nonword result is the key one: sute has no stored pronunciation to look up, so its sound must have been computed from its letters, and that computed sound reached the meaning of suit. Word and nonword homophones were equally error-prone in both experiments, so a stored pronunciation of hare added nothing. They concluded that readers compute sound from spelling as part of identifying printed words, not only after a word is identified.
One caution for this demo: every nonword needs a “no,” so you can reject sute by noticing it is not a word, without thinking about meaning at all. Real-word sound-alikes like hare and rows block that shortcut, which is why most of the sound-alikes here are words.
Your data from this session
Your trials stay in this page only. Nothing is saved; reloading or closing the page erases them. Practice trials are listed but left out of the summary. Rates use trials answered before the deadline; too-slow trials are counted separately.
Van Orden et al.’s rates come from runs without a deadline: subjects were asked to be fast and accurate (Experiment 1), or accurate but very fast (Experiment 2). Experiment 2 had no spelling controls. Your deadline makes errors more likely, so expect higher rates for every type.
All trials
How this demo differs from Van Orden et al.’s experiments
- Their subjects did not know the study was about sound-alikes. You do, so the demo adds a response deadline to keep errors possible, and shows feedback after each error. Their subjects had no deadline and the paper does not describe trial-by-trial feedback.
- About 30% of trials here are sound-alikes or spelling controls, and most sound-alikes are real words. In their Experiment 1, 10% of trials were sound-alikes (half words, half nonwords) and another 10% were spelling controls.
- Sound-alikes and spelling controls come from their Appendix A (Experiment 1) and Appendix C (Experiment 2). Five more word sound-alikes with spelling controls were written for this demo (pair/peak, knows/knots, waste/waits, blew/blow, sun/sin) and are marked “added” in the trial table. The category members and unrelated “no” words were also written for this demo; theirs came from published category norms.
- Each of their subjects saw every sound-alike and every spelling control. Here a run never shows a sound-alike together with its own spelling control, or with the category member it sounds like, and new runs use items you have not seen until the pool runs out.
- Their sessions had 40 practice trials and 200 test trials (about 45 min), and subjects answered by pulling levers. With 5–10 trials per type in a run, your percentages will move a lot from run to run.