You meet a word in your textbook and know it at once. An hour later you need it in a sentence and it has gone. You follow every recording in the course, and two people chatting at a bus stop are a blur.
Both gaps have one cause. Recognising a language and producing it are different abilities, and the checks that are quickest to build, a choice of four, a pair to match, a card to flip, measure the first. This guide shows how to test yourself when learning a language on your own so that the second is measured too.
EdProcto's page for self-learners lists its question shapes for the job, though a notebook and a voice recorder cover most of it. The general method, for any subject, is in our guide to testing yourself with no teacher.
Why can you recognise a word and still not produce it?
Because knowledge of a word comes in strengths, and picking it out of a list is the weakest. Laufer and Goldstein (2004) built a test that asks for a word in four ways and tried it on 435 learners of English. The four formed a ladder of difficulty that held for common and rare words alike.
| The question asks you to | Example (Spanish) | In the study |
|---|---|---|
| Pick the meaning from options | perro: dog, cat or bird? | Easiest |
| Pick the foreign word from options | dog: perro, gato or pájaro? | Second |
| Give the meaning yourself | perro means ____ | Third |
| Give the foreign word yourself | dog is ____ | Hardest |
The examples are ours; the order is theirs. If every check you take is a tap on one of several options, your score reports the two lowest rungs, and a conversation asks for the top one.
What did the research on vocabulary find about testing yourself?
It found that pulling a word out of memory keeps it better than seeing or saying it again.
Karpicke and Roediger (2008) had college students learn 40 Swahili words with their English meanings over rounds of study and testing. Once a pair had been recalled a single time, it either stayed in the later tests or left them. A week on, students who went on being tested recalled about 80% of the pairs. Those whose recalled pairs left the tests recalled 36% if they still reread every pair each round, and 33% if not. Every group had predicted about half.
Kang et al. (2013) set the "repeat after me" drill against its reverse. Learners saw pictures of objects and heard their foreign names. In one drill they heard a name and repeated it aloud; in the other they tried to say the name before hearing it. Tested at once in one experiment and after two days in the other, the second drill gave better understanding of the words, better production and pronunciation that was no worse. The paper's abstract does not name the language.
Which direction should the test run?
Both, and if there is time for only one, from your language into the new one. Steinel et al. (2007) had 129 Dutch university students learn 20 English idioms from English into their first language or the reverse, then tested both directions at once and three weeks later. Asked to produce the English, those who had practised producing it had a sizeable advantage; on recognition the other group did better. Webb (2009) found a similar split for single words among learners of English in Japan, and concludes that if only one method is used, the productive one may be the more effective.
How far apart should the tests be?
Days apart, and having a gap matters more than its pattern. Nakata (2015) had 128 Japanese college students learn 20 English words on gaps that stayed equal or widened, at four amounts of spacing from none to long. Widening had a limited, statistically significant advantage; the amount of spacing had a large effect.
Bahrick et al. (1993) followed four people for nine years as they relearned 300 pairs of English and foreign words in 13 or 26 sessions spaced 14, 28 or 56 days apart. Tested up to five years later, 13 sessions at 56 days had held about as well as 26 at 14. Four learners is a tiny sample: a direction, not a timetable.
How do you test yourself in each skill?
Give each skill a test in which you produce something, then check the result against a source other than your own opinion.
| Skill | The self-test | What it cannot tell you alone | Shape in EdProcto |
|---|---|---|---|
| Vocabulary | Type the word, both directions | Whether it arrives in time in speech | Translate, one word |
| Grammar | Fill gaps in your own sentences | Whether you use the form unprompted | Fill in the blank, cloze passage |
| Listening | One hearing at normal speed, then type it | How you manage an unfamiliar voice | Listen once, dictation |
| Reading | An unseen passage with gaps or questions | Whether your passage was too easy | Cloze passage |
| Speaking | Record an answer, compare with a model | Your accent, or a live exchange | Spoken answer, pronunciation, read aloud |
Vocabulary: type it, in both directions
Copy the unit's word list into two columns and cover one. Go from your language into the new one first, then back. Type or write every answer: "I would have got that" is a judgement made after seeing the answer, which is recognition again.
Grammar: gaps in your own sentences
Take sentences from the unit, or sentences you wrote that came back corrected, and blank the one form the unit teaches: the verb ending, the article, the case. Your own corrected sentences are our preference, since each holds a mistake you really make.
A gap proves less than it seems to: it points at the slot, and in speech nothing points. So add two or three whole sentences to translate. Our guide to cloze and fill in the blank questions covers which word to blank and which answers to accept.
Listening: one hearing at normal speed
Use ten to twenty seconds of audio that has a transcript and was not slowed for learners. Play it once, type what was said word for word, then count the words you caught against the transcript.
Catching a known word in running speech is a skill of its own. Matthews and Cheng (2015) gave 167 Chinese learners of English at tertiary level a partial dictation of common words and a listening comprehension test. Recognition by ear of words from the third thousand most frequent predicted 52% of the variation in comprehension scores. That is a correlation in one group, not proof that dictation raises comprehension, but a score earned on slowed audio describes slowed audio. Our guide to listening questions weighs one play against two.
Reading: an unseen passage
Choose a paragraph you have not studied, from the next unit or a text of that level. Restore eight blanked words, or read it once, cover it and answer questions: true, false or not said.
Speaking: record, then compare with a model
Pick a question from the unit, record yourself answering for 30 to 60 seconds without notes, and play the recording beside a model: the textbook's dialogue, or an answer a tutor corrected. Listen once for content (what was missing, what was wrong) and once to count the pauses.
Judging your own sounds is less reliable. Dlaska and Krekeler (2008) asked 46 advanced learners of German to compare their articulation of particular speech sounds with a native speaker's. Their verdicts matched the raters' in 85% of cases, yet they caught only half of the sounds the raters judged inaccurate.
EdProcto's spoken answer, pronunciation and read aloud shapes record you in the browser. A reader marks them by hand, so alone you listen back and judge against the model. Speak and compare shows a transcription of what you said to confirm or correct, and marks it against the model by rule.
Where do the questions come from when there is no teacher?
From three things you already have: the textbook unit, a transcript and your corrected writing.
- The unit. Its word list gives the vocabulary test, its example sentences the grammar gaps, its dialogue the model for speaking.
- A transcript of the course audio, a video or a podcast: the key for listening, the text for reading.
- Your corrected writing. Five sentences a week, checked by a tutor or against the book's key: each correction is a new gap.
Questions you write yourself only ask about what you noticed. For questions you did not write, paste any of the three into the quiz maker on EdProcto's home page, and Procto AI writes them from that text.
What does a week of self-testing look like?
Six sessions of about ten minutes, one skill each, with Monday's words back on Thursday. The plan is our arrangement, not a schedule from a study.
| Day | Session | Count | Minutes |
|---|---|---|---|
| Mon | New words, typed both ways | 15 words, 30 answers | 10 |
| Tue | Grammar gaps, then Monday's misses | 12 gaps | 8 |
| Wed | Clips heard once and typed | 3 clips of 15 seconds | 10 |
| Thu | Monday's words again, then older ones due | 30 answers and more | 12 |
| Fri | One reading passage | 120 words, 8 gaps | 8 |
| Sat | Recorded spoken answers | 2 of 45 seconds | 10 |
That is 58 minutes. The minutes are estimates, so work out your own: time one session and divide by its answers. At 20 seconds a typed word, 30 answers take 600 seconds, the ten minutes in the first row. In EdProcto, a quiz taken in study mode falls due again in a few days, so Thursday's session arrives without a diary.
How do you know a word is known?
Call a word known when you have typed it correctly, from your own language into the new one, on three separate days. That is our convention, not a finding, with two results behind it.
- Once is too few. In Karpicke and Roediger (2008), students whose pairs left the tests after one recall remembered about a third of them a week later.
- Repeats in one sitting are poor value. Nakata (2017) had 98 Japanese learners study 16 English words with one, three, five or seven retrievals each in one session. Five and seven beat one and three on later tests, but with time on task held equal, one retrieval gave the largest gain. We read that as: more words today, and each again another day.
Keep a paper tally: a tick a day for each word, a miss wipes the ticks, three ticks make it known.
What can a self-test not tell you?
It cannot judge your accent, stand in for a conversation or award you a level.
For accent it takes another listener, as the German study above shows. A conversation has a second person, who sets the topic and the speed, and only talking to people tests that.
The Common European Framework's self-assessment grids describe levels A1 to C2 in "I can" statements (Europass, n.d.); the B1 line for listening speaks of clear standard speech, and of broadcasts delivered relatively slowly and clearly. A grid is a description you apply to yourself. A level such as B1 is awarded by an exam body on its own exam, and no plan of self-tests can promise fluency by a date.
How does EdProcto help a language learner?
EdProcto is the best tool for this job when the test must come from your own unit, because its shapes make you produce and its study mode brings the test back: the two things the studies above support.
Paste your material into the quiz maker on the home page and choose a shape. Answers with one right form (one word, fill in the blank, cloze, dictation) are checked by rule against the accepted answers. A translated sentence is an open answer, read against a model answer. For listening you add the address of your own clip, and listen once plays it a single time. In study mode the answer and an explanation follow each question, your misses come round again, and the quiz is due again in a few days.
The language kits are blueprints your material fills, among them Vocabulary, both ways; Grammar gaps; Dictation; and Pronunciation.
The Free plan includes 20 questions a month written by Procto AI. For regular self-study the Learner plan lifts the cap on self-tests, has Procto AI write 100 questions a month, and takes a whole book chapter by chapter with a map of what you know. See the pricing page.
Quick questions
How do I test myself in a language without a teacher?
Make yourself produce: type words in both directions, fill gaps, type what a clip said after one hearing, record a spoken answer, and check each against the book, the transcript or a model. EdProcto's page for self-learners lists the shapes for it.
Why do I understand my course audio but not people talking?
Probably because catching known words in running speech is a separate skill from knowing them on the page. Matthews and Cheng (2015) found it predicted about half the variation in listening comprehension. Test it with a clip at normal speed, heard once and typed.
Can I check my own pronunciation?
Only partly. In Dlaska and Krekeler (2008), advanced learners caught half of the sounds that raters judged inaccurate. Record yourself, compare with a model, and play the recordings to a tutor or a fluent speaker.
How many times must I get a word right before it is learned?
Our convention is three separate days, typed from your own language into the new one. The research gives no number; Karpicke and Roediger (2008) show only that once is too few.
References
- Bahrick, H. P., Bahrick, L. E., Bahrick, A. S., & Bahrick, P. E. (1993). Maintenance of foreign language vocabulary and the spacing effect. Psychological Science, 4(5), 316–321. https://doi.org/10.1111/j.1467-9280.1993.tb00571.x
- Dlaska, A., & Krekeler, C. (2008). Self-assessment of pronunciation. System, 36(4), 506–516. https://doi.org/10.1016/j.system.2008.03.003
- Europass. (n.d.). Common European Framework of Reference for language skills. European Union. https://europass.europa.eu/en/common-european-framework-reference-language-skills
- Kang, S. H. K., Gollan, T. H., & Pashler, H. (2013). Don't just repeat after me: Retrieval practice is better than imitation for foreign vocabulary learning. Psychonomic Bulletin & Review, 20(6), 1259–1265. https://doi.org/10.3758/s13423-013-0450-z
- Karpicke, J. D., & Roediger, H. L. (2008). The critical importance of retrieval for learning. Science, 319(5865), 966–968. https://doi.org/10.1126/science.1152408
- Laufer, B., & Goldstein, Z. (2004). Testing vocabulary knowledge: Size, strength, and computer adaptiveness. Language Learning, 54(3), 399–436. https://doi.org/10.1111/j.0023-8333.2004.00260.x
- Matthews, J., & Cheng, J. (2015). Recognition of high frequency words from speech as a predictor of L2 listening comprehension. System, 52, 1–13. https://doi.org/10.1016/j.system.2015.04.015
- Nakata, T. (2015). Effects of expanding and equal spacing on second language vocabulary learning: Does gradually increasing spacing increase vocabulary learning? Studies in Second Language Acquisition, 37(4), 677–711. https://doi.org/10.1017/S0272263114000825
- Nakata, T. (2017). Does repeated practice make perfect? The effects of within-session repeated retrieval on second language vocabulary learning. Studies in Second Language Acquisition, 39(4), 653–679. https://doi.org/10.1017/S0272263116000280
- Steinel, M. P., Hulstijn, J. H., & Steinel, W. (2007). Second language idiom learning in a paired-associate paradigm: Effects of direction of learning, direction of testing, idiom imageability, and idiom transparency. Studies in Second Language Acquisition, 29(3), 449–484. https://doi.org/10.1017/S0272263107070271
- Webb, S. (2009). The effects of receptive and productive learning of word pairs on vocabulary knowledge. RELC Journal, 40(3), 360–376. https://doi.org/10.1177/0033688209343854