Semantic working memory holds the meaning of words and sentences, while phonological working memory holds the sound shape of language — these two tracks run in parallel inside verbal short-term memory and show distinct patterns of interference in laboratory research. A sentence is a useful example because every full clause carries both a meaning layer (what the sentence is about) and a sound layer (how the words sound when read aloud or rehearsed), so any sentence recall task naturally draws on both. The Sentence Memory Test is a recreational, eight-round browser game built around audited English target sentences: you study each sentence for up to six visible seconds, hide it, type it from recall, and inspect a word-level comparison of matching, substituted, inserted, and missing tokens. The product description is explicit that the route does not measure working-memory capacity, diagnose a cognitive or language condition, or rank ability in any percentile. Treat the exercise as entertainment with a transparent feedback surface for the sentence that was just shown.
Two Tracks Inside Verbal Short-Term Memory
Cognitive psychology distinguishes two kinds of short-term codes that listeners and readers use to keep language active: a meaning-based code that preserves what words and sentences refer to, and a sound-based code that preserves the auditory shape of those words. Researchers describe these as semantic and phonological working memory respectively, and they tend to fail in different ways. Semantic codes are vulnerable when incoming items have similar meanings — a list of category neighbors such as "lemon, lime, orange" tends to be more disruptive to recall than unrelated items. Phonological codes are vulnerable when items sound alike — "man, mad, mat, map" causes more confusion in immediate serial recall than a list of phonologically distinct items.
This division is described across the broader literature on verbal short-term memory, including reviews that treat meaning and sound as partially separable components of language processing. The interest for a sentence recall exercise is that a full English clause carries both kinds of information at once: every verb has a meaning and an argument pattern, every noun refers to something, and every word contributes syllables to the spoken form.
Why a Sentence Sits Across Both Codes
When you read or hear "people following a guide through a museum" you can hold the meaning — there is a group, there is a leader, there is a venue — without holding the sound. You can also rehearse the sound — "PEE-pul FOL-oh-ing uh GIDE" — without committing to a precise interpretation. Most readers do both at once, and the question of which code dominates in any particular recall task depends on the task design, the timing, and the listener's strategy. That is why cognitive scientists often use sentences as stimuli when they want to probe both systems together.
The eight target sentences in the Sentence Memory Test are ordinary English clauses — covering situations such as arranging postcards, stacking boxes, mailing a parcel, renting bicycles, faxing a form, chopping carrots, and enjoying a concert. Each one is short enough to read in the visible window, ordinary enough to engage natural language intuitions, and distinct enough to discourage simple surface memorization. None of this turns the game into a measurement of either memory type; it only means that any recall attempt will engage both meaning and sound in some proportion.
The Sentence Memory Test as a Recreational Route
The Sentence Memory Test presents eight fixed English target sentences written specifically for this product. Each sentence's central verb and argument pattern is anchored to a pinned WordNet 3.1 sense key and template number, and the usage is cross-checked on a corresponding Cambridge Dictionary verb page without copying Cambridge text. The published WordNet 3.1 copyright notice is preserved on the site, and the eight literal targets, normalized tokens, sense keys, and template numbers are frozen so the production route has to match these independent cases exactly rather than construct its own expected answers.
The route is still treated as reference data under the site's English sentence-bank rule, and the description is direct about its limits: the exercise is entertainment, not a clinical or cognitive assessment, and it does not establish school, employment, or treatment readiness. If you want a different sentence-recall angle that also uses this same tool, see the Meaningful vs Random Word Recall walkthrough, which changes the content while keeping the eight-round route.
How to Play the Sentence Memory Test
The eight-round route is short and the controls are explicit. Each sentence is presented, hidden, recalled, and compared; the comparison uses a deterministic word-level alignment so the feedback tells you exactly where your typed text diverged from the sentence that was just shown.
- Open the route and press Start. The first target sentence appears on a card and stays visible for up to six seconds.
- Read the sentence during that window, or press Hide earlier when you are ready to recall.
- Type the sentence from memory into the text area and press Submit. Entry itself is untimed, so typing speed does not affect the score.
- If your submission matches, the route advances to the next target exactly once. If it does not match, the route shows a minimal alignment of matching, substituted, inserted, and missing tokens.
- Press Continue to reveal the same target again with a fresh visible token, then retry. A perfect match after a mismatch still moves on.
- Repeat for all eight fixed targets. The summary reports completed sentences, total mismatches, cumulative token edits, and a points total computed from this route only.
What the Word-Level Edits Show You
The comparison uses a disclosed, narrow normalization policy applied in a fixed order. Input first receives Unicode NFKC compatibility normalization, then lowercasing and outer trimming. Repeated internal ASCII whitespace is collapsed, and one final period, question mark, or exclamation mark is ignored. Interior punctuation is not silently deleted, words are not reordered, and no synonym is substituted. After that, a dynamic-programming token alignment finds the minimal sequence of matches, substitutions, insertions, and deletions.
That alignment matters because one omitted word should not make every later word appear wrong, and a repeated word must be aligned consistently. For example, if you type "people follow a guide" instead of "people following a guide," the missing token shows up as a single deletion at its position rather than a cascade of false mismatches downstream. The feedback tells you where your typed text diverged from the sentence that was just shown; it does not rank the quality of your memory against anyone else.
Scoring, Pauses, and Visibility Safeguards
Points start at 800. A mismatched submission stays on the same sentence, increments the error count, and costs up to 25 points, never dropping the total below zero. A perfect match advances exactly once and does not deduct points. The summary therefore reflects cumulative token edits on the fixed route rather than a response-time rank, and entry timing does not appear in the score.
The six-second reveal uses a monotonic browser deadline rather than assuming a timer callback fires at an exact instant. If the document becomes hidden during study, the round pauses and preserves only the remaining visible time. Opening a shared overlay screen is detected at the document boundary and pauses the same way; in either case, Resume creates a new deadline when the page is visible again. A backward timestamp, a stale timeout, or an old interaction token is rejected, so paused study time is never counted as bonus viewing time.
Comparing Semantic and Phonological Memory
The table below summarizes the qualitative contrast researchers draw between the two memory codes. It is descriptive rather than a measurement: a sentence recall exercise can engage both, but the game described here is not built to isolate either one.
| Aspect | Semantic working memory | Phonological working memory |
|---|---|---|
| What it codes | The meaning of words, sentences, and referents | The sound shape and syllable structure of words |
| Typical interference source | Items with similar meanings or a shared category | Items that sound alike or share rhyming patterns |
| Common laboratory probe | Sentence–picture matching, synonym judgments, semantic categorization | Immediate serial recall of word lists, nonword repetition, digit span |
| Role in everyday language | Comprehension, inference, vocabulary use, discourse tracking | Rehearsing a phone number, repeating a phrase, holding a novel word |
| Behavior when capacity is exceeded | Meaning-level intrusions from semantically related items | Sound-level errors such as order swaps, phoneme swaps, or omissions |
Because the eight targets are ordinary English clauses, every recall attempt draws on both rows of this table to some degree. The route does not, however, separate the two contributions, and any difference in your recall accuracy across the eight sentences should be read as a transparency report, not a measurement of one memory system versus the other.
What This Route Does Not Measure
The product contract is unusually direct on this point. The Sentence Memory Test 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, does not establish school or employment readiness, and does not show that practice improved the brain. Familiarity makes the repeated route easier, and the sentences are not normed against any reference group. If you want a fixed, auditable comparison between what you typed and the sentence that was just shown, the route delivers that. If you want a clinical or educational assessment, this is not the place to look.
The shared shell may remember a best point total locally on your device, but it does not save individual sentence attempts and does not build a memory profile. There is no microphone, synthesized voice, audio recording, runtime dictionary request, account, or uploaded text. The route is a transparent comparison with text that was just shown to you, and the feedback is best read as exactly that.
Related reading: Simple vs Complex Sentence Order in a Word Game.