Lexical decision task with your own frequency curve

The task looks trivial and is the workhorse of visual word recognition for one reason: the yes-responses are not all the same speed. A word you encounter every day is recognized faster than one you meet twice a year, the ordering holds across languages and across decades of hardware, and it is measurable inside a single person’s eighty trials. That is what makes it a demonstration rather than a survey. Everything interesting about running it well is downstream of a single design question — what the non-words are — because the non-words set the difficulty of the decision. A string of random consonants can be rejected without consulting the lexicon at all, so a set built that way produces fast rejections and a small word-versus-non-word difference. A string that obeys English spelling forces a real search that has to end in a failure. The forty here go one step further: each is a real word with one letter changed, so each has exactly one close neighbour in your vocabulary, and that makes them the hard end of the scale. The page says so beside the number, because a difference measured on this set is not the same quantity as one measured on arbitrary strings.

The second decision is the word set, and it is the one most browser versions skip entirely. Frequency is usually treated as a two-level thing — common and rare — which yields one difference and no way to tell a genuine gradient from a single odd cell. Four ordered bands give a shape, and the shape is falsifiable: if your band 4 is not slower than your band 3, that step did not separate for you, and the page says which steps rose rather than hiding it in an average. Frequency also travels with two things that have to be held still or they do the work instead. Length is the obvious one, controlled here by keeping every band within half a letter of the others. The subtler one is that frequency is not a property of a word, it is a property of a corpus — the count depends entirely on what was counted, and film subtitles, newspapers and books disagree about which words are common. That is why this page publishes an ordering and never a number: an ordering is a claim it can defend, and a per-item frequency value quoted from memory is the exact defect this site exists to avoid.

No percentile appears beside your result and none is coming. The large item-level datasets were collected in laboratories with button boxes; browser keyboard and display latency add an unknown constant that is a substantial fraction of the effects being measured. The standard item-level collection is Balota et al. (2007), The English Lexicon Project, Behavior Research Methods, and this page names it without quoting a value out of it. What survives the move into a browser is the subtraction, since whatever latency your screen and keyboard add is present in both cells of every difference the page reports. The neighbouring pages take the same word knowledge from other angles: the verbal fluency test asks you to produce words against a clock rather than judge them, the vocabulary test estimates how many you know at all, and the language processing test moves up from single words to sentences and reports which level cost you most. If it is speed through continuous text you were after rather than speed on isolated words, the reading speed test measures that with a comprehension check attached. And for the same two-key speeded decision applied to something other than language, the digit symbol substitution test and the trail making test are the classical processing-speed pair. Two more sit on the knowledge this page only borrows: the spelling test online asks you to produce the letters rather than recognize them, which is the ability the forty one-letter changes here are quietly leaning on, and the synonym test asks what a word means instead of how fast you can tell that it is one — the difference between knowing a word and reaching it quickly, which no single latency can separate.

A reaction time here is the interval between the frame that painted the stimulus and the timestamp the browser attached to your key, both read from the same monotonic clock. What neither can see is the display pipeline behind it, so on a 60 Hz screen roughly 16 ms of every figure below is the machine rather than you. That is the timing floor: two numbers closer together than that are the same number, and this page reports no precision it cannot support.

This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish dyslexia or any reading disorder. Only a qualified professional, working with more than a browser, can make that judgment.