Lexical Decision Tasks: A Deep Dive

What is a lexical decision task?

In a lexical decision task (LDT) participants see a string of letters and must decide, as quickly and accurately as possible, whether it is a real word in their language (table) or a non-word (dnfke). The response is usually a key press, and the two dependent measures are reaction time and accuracy on the word trials. Non-words are typically pronounceable and orthographically legal (glort rather than xqzpt) so that the decision cannot be made on superficial features alone.

The task is popular because it is simple to explain, takes only a few minutes, and gives a sensitive index of how quickly a word is retrieved from the mental lexicon. It is one of the workhorse methods of psycholinguistics and of research on reading, bilingualism and memory.

Origins and landmark studies

• Rubenstein, Garfield and Millikan (1970) are usually credited with introducing the paradigm. In Homographic entries in the internal lexicon (Journal of Verbal Learning and Verbal Behavior) they used word/non-word decisions to show that words with several meanings (homographs) are recognised faster than unambiguous words.
• Meyer and Schvaneveldt (1971) presented pairs of letter strings and found that a word was judged faster when it followed a semantically related word (bread then butter) than an unrelated one (nurse then butter). This semantic priming effect became one of the main sources of evidence for spreading activation in semantic memory.
• Balota and Chumbley (1984) compared lexical decision with naming and category verification and showed that the word frequency effect (common words are recognised faster than rare words) is considerably larger in lexical decision than in the other tasks. Their argument was that part of the frequency effect in the LDT arises from the decision stage rather than from lexical access itself, a point that still shapes how results from the task are interpreted.

Common variants

• Semantic priming LDT. Each target is preceded by a prime word that is related, unrelated or neutral. The difference in RT between related and unrelated conditions is the priming effect. Varying the prime-target interval (stimulus onset asynchrony) lets researchers separate fast automatic activation from slower strategic expectancy.
• Masked priming LDT. The prime is shown very briefly (often around 50 ms) between a forward mask such as ##### and the target, so that participants are usually unaware of it. It is widely used to study early orthographic and morphological processing.
• Word frequency manipulations. Targets are selected from corpus frequency norms to contrast high- and low-frequency words while matching length and other properties. Frequency effects in the LDT are robust and are a standard benchmark for models of word recognition.
• Bilingual and cross-language LDT. Stimuli from two languages are mixed, or a prime in one language precedes a target in the other, to study how bilinguals store and access their lexicons.

Try the lexical decision task

Press W if the letter string is a word and N if it is not. You have 1.5 seconds for each string.

Behind the scenes: coding the LDT

The embedded demo runs on an earlier version of jsPsych. The code below is the equivalent version written for jsPsych 8 (it also runs on jsPsych 7). It shows a 500 ms fixation cross before each string, presents the strings in random order, and scores each response with compareKeys.

const jsPsych = initJsPsych({
  on_finish: () => {
    jsPsych.data.displayData();
  }
});

const timeline = [];

// Instructions
const instructions = {
  type: jsPsychHtmlKeyboardResponse,
  stimulus: "<p>In this task, you will see a series of letter strings.</p>" +
            "<p>Press <strong>W</strong> if the string is a word, and <strong>N</strong> if it is not a word.</p>" +
            "<p>Press any key to begin.</p>",
  post_trial_gap: 500
};
timeline.push(instructions);

// Stimuli: words and pronounceable non-words
const stimuli = [
  { stimulus: "table", word: true },
  { stimulus: "chair", word: true },
  { stimulus: "house", word: true },
  { stimulus: "water", word: true },
  { stimulus: "dnfke", word: false },
  { stimulus: "glort", word: false },
  { stimulus: "prane", word: false },
  { stimulus: "slint", word: false }
];

// Fixation cross shown before every string
const fixation = {
  type: jsPsychHtmlKeyboardResponse,
  stimulus: "<p style="font-size: 40px">+</p>",
  choices: "NO_KEYS",
  trial_duration: 500,
  data: { task: "fixation" }
};

// The lexical decision itself
const ldt = {
  type: jsPsychHtmlKeyboardResponse,
  stimulus: () => "<p style="font-size: 40px">" + jsPsych.evaluateTimelineVariable("stimulus") + "</p>",
  choices: ["w", "n"],
  trial_duration: 1500,
  data: {
    task: "ldt",
    string: jsPsych.timelineVariable("stimulus"),
    word: jsPsych.timelineVariable("word")
  },
  on_finish: (data) => {
    const expected = data.word ? "w" : "n";
    data.correct = jsPsych.pluginAPI.compareKeys(data.response, expected);
  }
};

timeline.push({
  timeline: [fixation, ldt],
  timeline_variables: stimuli,
  randomize_order: true
});

// Feedback at the end
const debrief = {
  type: jsPsychHtmlKeyboardResponse,
  stimulus: () => {
    const trials = jsPsych.data.get().filter({ task: "ldt" });
    const accuracy = Math.round(trials.filter({ correct: true }).count() / trials.count() * 100);
    const rt = Math.round(trials.filter({ correct: true }).select("rt").mean());
    return "<p>Accuracy: " + accuracy + "%. Mean RT on correct trials: " + rt + " ms.</p>" +
           "<p>Press any key to finish.</p>";
  }
};
timeline.push(debrief);

jsPsych.run(timeline);

A trial with no response within 1500 ms has response and rt set to null and is scored as incorrect. To turn this into a priming experiment, add a prime trial (a short jsPsychHtmlKeyboardResponse with choices: "NO_KEYS") between the fixation and the target, and add a prime and relatedness field to each entry in stimuli. For a frequency study, add a frequency field so that you can split the RTs by condition when analysing.

Hosting it on Cognition

Paste the code into the Cognition editor, check it in the live preview, and share the resulting https://<token>.cognition.run link with your participants. Each trial is uploaded as soon as it finishes; if a participant loses their connection, a service worker keeps the trials locally and resends them when the connection returns, and everything is flushed before your on_finish runs. The CSV or JSON export includes the string, word, response, rt and correct columns used above.

If your design has between-subjects conditions (for example, different prime lists), Cognition balances assignment on the server and exposes the condition number as window.CONDITION. Participant IDs from Prolific or SONA passed in the URL are added to every trial automatically; see our integration guide for completion codes and redirects. Create a free account to host up to 4 tasks with 60 participants each, or read the documentation.