Immediate sentence recall happens while a sentence is still in view; delayed sentence recall happens after the sentence has been hidden, when only your memory can supply the words. In a structured comparison of the two, you read a target sentence, cover it, and then try to reproduce it from memory . The value of the contrast is that immediate recall reveals how well you registered the sentence on first contact, while delayed recall reveals how well that registration survived without the original text in front of you. A browser tool that freezes both phases onto a single fixed route lets you see, sentence by sentence, which words drop out, swap, or get added once the page goes blank. Sentence Memory Test is built around that contrast: it shows each target for up to six visible seconds, lets you hide it earlier when ready, and then asks you to type it from memory with word-level feedback.
What Immediate and Delayed Recall Actually Test
Immediate recall measures the strength of the very first encoding: how much of the sentence you can repeat while the original text is still right there. Delayed recall measures the strength of the memory trace after the stimulus is gone and you have had no opportunity to glance back. The two tasks are not redundant. A sentence that survives a six-second visible window intact can still collapse once it is covered, because the cover removes external scaffolding and forces the brain to generate the words itself rather than read them.
In research designs, the same sentence list is usually shown once, then recalled once right away, then recalled again later after a delay. That sequence isolates what is lost between the two time points. Sentence Memory Test compresses that design into a single short round: one sentence at a time, one visible window, one typed recall attempt, with the option to reveal the same target again and try once more before moving on.
Sentence Memory Test Puts Both Phases on a Fixed Route
The game runs exactly eight sentences in a fixed order. Each sentence appears as one complete English target on a single card. You study it while it is visible for up to six seconds, and you may hide it earlier than that when you feel ready. The moment the card is hidden, the delayed-recall phase begins: you type what you remember into a text area and submit it for a word-level comparison. There is no countdown on the typing itself, so your recall reflects what your memory holds, not how fast your fingers move.
The visible-time safeguard is enforced with a monotonic browser deadline rather than a timer callback. If you switch tabs or open a shared overlay while the sentence is in view, the study phase pauses and only the remaining visible time is preserved. When you return and explicitly press Resume, a fresh deadline is created based on the visible portion you had left. That means hidden time never quietly becomes extra study time, and a paused sentence never accidentally consumes its full six-second window after a long distraction.
How to Play Sentence Memory Test
Each round walks through the same transitions in order: a visible study window, a typed recall from memory, and, after a mismatch, a same-sentence reveal.
- Open the round and let the first complete target sentence appear on the card, then read it carefully during the visible study phase.
- Wait until the visible window expires, or press Hide as soon as you feel ready; either action ends the immediate-recall window and starts the delayed phase.
- Type the sentence from memory in the text area and submit it; submission triggers a word-level diff against the original target.
- If the comparison is a perfect match, the route advances to the next sentence; if it shows a mismatch, inspect the alignment of matching, substituted, inserted, and missing words.
- Press Continue to reveal the same sentence again with a fresh visible token, then hide it and try one more time before the route moves on.
- Repeat until all eight fixed targets are completed, then read the summary for completed sentences, total mismatches, cumulative token edits, and points.
How Word-Level Edits Compare Your Typing
After a mismatch, the diff is computed through a disclosed normalization policy applied in a fixed order. The submitted string first receives Unicode NFKC compatibility normalization, then lowercasing, then outer trimming of surrounding whitespace. Repeated internal ASCII whitespace is collapsed into single spaces, and one trailing period, question mark, or exclamation mark is ignored. Interior punctuation is never silently removed, words are never reordered, and no synonym substitution is applied — so "boxes" cannot be accepted as "cartons," and "stacking" cannot be accepted as "piling."
Once both strings are normalized, a dynamic-programming token alignment finds the minimal sequence of matches, substitutions, insertions, and deletions. The order of that minimum matters: one omitted word does not push every later word into the substitution column, and a repeated word has to align consistently across both strings. The result is a transparent line-by-line view of where your typed attempt drifted from the original target, which is the practical bridge between "I thought I got it" and "I see exactly where it fell apart."
| Normalization Step | What It Changes | What It Does Not Change |
|---|---|---|
| NFKC compatibility | Folds compatible Unicode forms | Interior punctuation marks |
| Lowercasing | "Follow" becomes "follow" | Word order or synonyms |
| Outer trimming | Leading and trailing whitespace | Internal spaces between words |
| ASCII whitespace collapse | Double spaces become single spaces | Capitalization inside words |
| One terminal punctuation | A trailing period is ignored | Question marks or exclamation marks inside |
What's in the Sentence Set and How It Was Audited
All eight target sentences are original to the product rather than copied from a dictionary or corpus. Each sentence has a central verb whose argument pattern is anchored to a pinned Princeton WordNet 3.1 sense key and sentence-frame number, then independently checked on an exact Cambridge Dictionary verb page. Cambridge definitions and examples are not stored or reproduced inside the game; the cross-check is a usage check, not a copy. The targets cover everyday actions such as people following a guide, arranging postcards, stacking boxes, mailing a parcel, renting bicycles, faxing a form, chopping carrots, and enjoying a concert.
The published WordNet 3.1 Copyright 2011 notice preserves the primary source terms where they appear. You can verify each verb's usage on the dictionary site directly — for instance, the Cambridge Dictionary entry for arrange and the Cambridge Dictionary entry for chop — without the route claiming those pages as its own source.
How the Score Works and What It Doesn't Claim
The summary screen reports four fixed numbers: completed sentences, total mismatches, cumulative token edits, and points. Points begin at 800. A mismatched submission stays on the same sentence, adds one error to the mismatch total, and costs up to 25 points, but the running total never drops below zero. A perfect match advances the route exactly once, and the next sentence starts from wherever your running point total stands.
It is important to read those numbers as descriptive outputs of a fixed recreational route, not as a memory rating. The eight sentences are not normed against any reference group, the route is short, and familiarity with the same sentences makes a second pass easier. The game does not measure working-memory capacity, does not diagnose a cognitive or language condition, does not estimate intelligence, does not assign a percentile or CEFR level, and does not claim that practice improves the brain. For a parallel comparison of recall against a different study method, the Definition Recall vs Rereading guide walks through how a quiz-based route reports its own numbers.
When the Browser Pauses the Visible Study
Several safeguards make sure that the visible window is honest. A monotonic browser deadline replaces a naive timer, which means a paused or backgrounded tab cannot quietly use up the full six seconds. A backward timestamp, a stale timeout, or an old interaction token is rejected outright. Held keys, duplicate clicks, old-round events, input submitted after completion, and any user controls hidden behind a shared overlay cannot mutate the game state. Only the documented system pause may preserve remaining visible time, and Resume always creates a new deadline from the time left on the original window.
Entry into the round is untimed, so the typing speed itself does not affect the result. The target card, text area, diff list, and controls use written labels and do not rely on color alone, and a live status announces each phase change. The single-column layout wraps cleanly on a 390-pixel viewport. No microphone, synthesized voice, audio recording, runtime dictionary request, account, or uploaded text is involved, and the only thing the local shell remembers is a best point total — the sentence attempts themselves are not saved and no memory profile is created.
| Recall Phase | What Happens | How the Game Handles It |
|---|---|---|
| Immediate (study) | Target sentence is visible on the card | Monotonic deadline of up to six seconds; pauses on hidden tab |
| Transition | Card is hidden by timer expiry or by Hide | Delayed-recall phase begins; remaining time is preserved if paused |
| Delayed (typed recall) | User types from memory and submits | Entry is untimed; only accuracy and edits affect the result |
| Feedback | Word-level diff after a mismatch | Minimal alignment of matches, substitutions, insertions, missing |
| Repeat | Same sentence revealed with fresh visible token | Continue button; mismatches add to error and point totals |