AbilityBenchFree cloze, 20 gaps, 5 blank types
Sentence completion test scored by what the gap tests
Twenty sentences, one gap each, four options — and the result is five rates rather than one mark, because a gap can be testing a verb sequence, a fixed collocation, a dependent preposition, a connector’s syntax or the register the sentence has already committed to, and missing one of those means nothing like missing another. Free, no account, no clock, every item returned afterwards with the reason its key is the key. If you came here for the projective sentence-completion instrument used in clinical psychology, that is a different test with no options and no answer key, and this page explains the difference rather than pretending to be it.
- 100% free
- No signup
- 20 gaps
- 5 blank types
- Reasoning shown for every item
Two unrelated tests share this name — check which one you came for
This one is the language-testing kind: a sentence with a gap, four options, a right answer. The other is a projective instrument used in clinical psychology, where you finish an open stem in your own words and a trained rater reads what you wrote. That second kind has no options and no key, and it is not on this site.
20 sentences, four options each, no clock. Every gap is doing one of 5 jobs, and the result reports the 5 separately rather than adding them up: which class you miss says far more than how many you missed, because a tense error and a collocation error need entirely different work.
The 5 jobs a blank can be doing
- Verb form and sequence — Which tense the rest of the sentence has already fixed, usually through a time expression or a second clause.
- Which verb the noun takes — The one verb that fixes to this noun, where two or three others mean the same thing and are not used with it.
- The preposition a word demands — The particle a specific verb, noun or adjective governs — depend on, accused of, answerable to.
- The connector's own syntax — Which linking word can stand in this slot, given what follows it — a clause, a noun phrase, or nothing at all.
- Register the frame has already set — Which of four synonymous options belongs in the variety of English the surrounding sentence is written in.
4 items in each class, 20 in total, every one asked exactly once. Four items is a thin denominator and the result says so — the point of the split is to tell you which class to look at, not to grade you within it.
The 20 items are interleaved so that no two of a kind arrive together — with four collocation items in a row you would work out what the class is testing from the item before, and the class rate would stop meaning anything. Nothing tells you whether an answer was right until the end, and every item comes back afterwards with the reasoning.
How to take the cloze
Twenty gaps, interleaved so no two of a kind arrive together, with the class named only after the run.
Read to the end of the sentence before you choose
In most of these items the word that decides the answer is not next to the gap. A since near the front fixes a tense at the back; a semicolon in front of the gap rules out three of four connectors on syntax alone; a frame like I am writing to has already chosen a register before the options appear. Answering from the words either side of the gap is the single commonest way a cloze item goes wrong, and several items here are built to catch exactly that.
Press 1 to 4, or tap, twenty times
Nothing tells you whether an answer was right while the run is going. The items are also interleaved by class rather than grouped, so you cannot infer what a gap is testing from the one before it — with four collocation items in a row the fourth would be free. There is no time limit on any item and no penalty for guessing, though a class of four items where you guessed twice is a rate the results panel will not pretend to trust.
Read the five class rows, not the total
The result opens with the total and immediately says why the total is the least useful thing on the page. Under it are the five classes with what each one tests, and a note attached to any class you did not get right explaining what a miss there actually means — a rule not read, an arbitrary pairing not known, syntax mistaken for meaning, or an audience misjudged. Then every sentence comes back with the key filled in and a sentence on why it is the key.
Technical specifications
| Items and classes | 20 sentences, 4 in each of 5 classes: verb form and sequence, verb-noun collocation, dependent preposition, connector syntax, and register. Every item is asked exactly once, so a run is the whole bank |
|---|---|
| Order | Interleaved: one item from each class in turn, with the class order reshuffled every round from the run seed. Grouping four items of a class together would let the fourth be answered from the pattern of the first three |
| Options and chance | 4 per item, 1 correct, so blind guessing scores 25% and a whole run answered at random averages 5 right. In a 4-item class, 1 right is exactly what chance produces — which is why the panel prints a Wilson interval on the total and calls each class a rate |
| What is not reported | No level, no percentile, no grade, no composite. The five classes measure a rule, two arbitrary word pairings, a syntactic constraint and a judgment about audience; averaging them would produce a number with no referent |
| The other sentence completion test | The projective instrument of the same name — Rotter & Rafferty (1950), Manual: The Rotter Incomplete Sentences Blank, College Form, Psychological Corporation and Loevinger (1998), Technical Foundations for Measuring Ego Development, Lawrence Erlbaum — is open-ended, rated by a trained person against a manual, and copyrighted. Nothing from either is reproduced here and no scored equivalent of them exists |
| What voids an item | An answer arriving before the sentence was painted, and any item the tab was hidden or unfocused during. A voided item is asked again, because 4 items per class cannot absorb the loss of one. There is no fast-answer rule here: the item is self-paced, so a quick answer is a quick answer |
| Feedback | Nothing during the run; everything after it. Each of the 20 items returns with the key inserted into the sentence and one sentence explaining what ruled the other three options out — including which of them are grammatical elsewhere and merely wrong here |
| What leaves the page | Nothing at all. The 20 sentences, the keys and the reasoning notes are inside the page, and the copy button is the only route out — to your own clipboard |
Frequently asked questions
I was looking for the test where you finish the sentence yourself. Is this it?
No — that is a projective instrument and this is a scored cloze. The projective kind gives you an open stem, you complete it in your own words, and a trained rater reads the completions against a published manual; there are no options, there is no key, and the output is an interpretation rather than a mark. The best-known are the Rotter Incomplete Sentences Blank and the Washington University Sentence Completion Test, both copyrighted and both administered by people with the training to read them. This site does not publish projective instruments, and it says so rather than offering a lookalike with a score attached.
Why not just give me one score out of twenty?
Because the twenty items measure five things that are not versions of each other. Missing a verb sequence means a rule in the sentence was not read and can be fixed by being shown it once. Missing a collocation means an arbitrary pairing is not in your memory, and no rule will put it there. Missing a connector usually means the meaning was right and the syntax was wrong. Adding those together produces a number that moves when any of them moves, which is exactly the property that makes it useless for deciding what to do next.
Four items per class is not very many. How much can I read into a row?
Which class to look at, and not much more than that. On four options, one right out of four is what pure guessing produces, so a 1 of 4 row is consistent with knowing nothing about that class and also with a bad run — the difference is not resolvable at this length. What four items do support is a comparison across the five rows within one run: a 4 of 4 next to a 1 of 4 is a real difference in a way that either number on its own is not. The panel prints an interval on the total for the same reason, so the width is visible rather than implied.
Two of the options are things I have definitely seen written. Why is one wrong?
Because several items are built on options that are correct English somewhere else. Come short was ordinary usage a century ago and now reads as a translation; kindly inform me is impeccable in a letter and wrong in a message to the person at the next desk; although marks the same contrast as however and cannot follow a semicolon. In each case the wrong option is not an error of grammar but of fit, and fit is what a cloze measures that a grammar quiz does not. The review says which of the three distractors are wrong for which reason.
Does the register class have a right answer at all?
Yes, because the frame chooses it rather than a style guide doing so. Each register item names the setting inside the sentence — a letter to a regulator, a methods section, a contract clause, a message to a colleague at the next desk — and once the setting is named, three of the four options are in the wrong variety. The class is not testing whether you prefer formal English; it is testing whether you can hear that a sentence has already committed to one, which is why one of the four items runs the other way and marks over-formality as the error.
Is this the same as the sentence completion section of an aptitude test?
The format is the same and the purpose is not. A graduate or hiring aptitude test uses cloze items to rank candidates against each other, so its items are calibrated for spread and its norms are the whole point; the items here are chosen for what they diagnose, and there is no norm behind them at all. If you are preparing for an assessment rather than working out which grammatical relation catches you, the items on this page are still useful practice for the format — but the number at the end is not a predicted score.
Can I see the answers without taking it?
Not before you answer, and all of them immediately afterwards. Feedback during a run would teach the class the next item is about to test, which is why nothing is revealed mid-flight; the twenty reasoning notes are then shown together, and the copy button takes the whole breakdown as plain text. There is no email step, no partial result and no second page — the reasoning for all twenty items is on screen the moment the last gap is filled.
Two tests, one name, and what a gap actually measures
The phrase “sentence completion test” names two unrelated instruments, and a visitor arriving from a psychology course and a visitor preparing for a placement exam both search for it. In clinical and developmental psychology it means a projective technique: open stems the respondent finishes however they like, read afterwards by somebody trained to read them against a published manual. Rotter & Rafferty (1950), Manual: The Rotter Incomplete Sentences Blank, College Form, Psychological Corporation is the best-known, and Loevinger (1998), Technical Foundations for Measuring Ego Development, Lawrence Erlbaum is the one used in developmental research. Those are not quizzes with hidden right answers — the completions are the data, the rater is the instrument, and nothing about them survives being turned into four options with a key. They are also copyrighted, and their validity depends partly on their stems not circulating. In language testing the same phrase means what is on this page: a gap with a defensible correct answer, used to place learners or to rank candidates. This site builds the second and explains the first. The query covers two unrelated things. As a verbal-aptitude item type it is a question format with no norm, for the same reason the other reasoning formats have none. As a psychological instrument it means a projective sentence-completion blank, which is proprietary, is interpreted by a clinician rather than scored, and sits inside the clinical zone this site stays out of.
Within the language-testing version, the interesting question is not how many gaps someone fills but which kind they miss, and the kinds are genuinely different problems. A verb-form gap is rule-governed: something in the sentence — a since, a before anyone noticed, a would not have been — has already fixed the answer, and a miss means that signal was not read. A collocation gap is arbitrary and cannot be reasoned out at all: nothing explains why English makes a promise and does a favor, or why a rule comes into force and falls into disuse, so a miss there is a gap in memory for pairs rather than in understanding. A dependent preposition sits in between, arbitrary but in a space small enough to learn, and the errors cluster on the words that take two — answerable to a body, answerable for a decision. A connector gap looks like vocabulary and is really syntax: however, although and despite all mark a contrast, and one needs a clause, one needs a noun phrase and one needs punctuation in front of it, so a writer can have the meaning exactly right and the sentence still be broken. And a register gap is about audience: all four options are correct English, and the sentence has already decided which variety it is in.
That split is also the reason this page reports no level. A CEFR band or a grade is a single position on a single scale, and the five classes here do not lie on one — somebody can be reliable on tense sequence and hopeless on collocation, which is the ordinary profile of a competent non-native writer and is invisible in any aggregate. The English level test does the level job properly, by walking a ladder of CEFR blocks and stopping at the first one that breaks; this page does the diagnostic job, and the two are worth taking in that order if you want both. For the vocabulary half of the same territory, the synonym test finds the frequency band where recognition starts to fail and the vocabulary test estimates size with a correction for guessing. For the same diagnose-by-class idea applied to writing rather than reading, the spelling test dictates words aloud and sorts the errors by the rule that produced them. If the difficulty is in getting through the text rather than in choosing the word, the reading speed test and the reading comprehension test measure that pair together. And if a hiring assessment is what brought you here, the page on personality assessments for jobs covers what an employer’s battery actually contains, and the language processing test measures how fast meaning assembles rather than how much you know.
No item on this page is timed, no result here is a duration, and there is deliberately no rule against answering quickly. The engine behind the page does carry a 150 ms floor for discarding answers prepared before a stimulus, but it applies that floor only where a response window exists, and this run gives you as long as you like on every gap. What is still thrown away is an answer that arrives before the sentence has been painted at all, and any item the tab was hidden or unfocused during; each of those comes back later in the run, because four items in a class cannot absorb the loss of one.
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 an emotional difficulty, a personality trait or anything at all about your inner life. Only a qualified professional, working with more than a browser, can make that judgment.
The stronger version of that sentence belongs on this page in particular. Because of the name, some visitors arrive expecting an instrument that reads something about them out of what they write. Nothing here does: you choose among four words somebody else wrote, and which one fits a sentence about a kettle carries no information about you whatsoever.
Where the twenty answers are checked
Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.
The sentences, the keys and the twenty reasoning notes all ship inside this page, so the comparison between your choice and the key happens in this tab and no request is made while a run is going. That is also why a reload loses a finished breakdown, and why the copy button carries the run seed: the seed is the only record of the order the sentences came in.