The Vozelu Method

When words come back

Vozelu uses previous answers to decide when to review each Spanish word. It asks the learner for the word, shows the answer after the attempt, and adjusts the next review date.

Trying to remember an answer before looking it up is called retrieval practice. Leaving time between attempts is called spacing. Research supports both practices, although it leaves open the exact intervals an application should use.

Reviewing after a correct answer

A correct answer shows that the learner could recall the word at that moment. Further attempts help maintain that ability. Karpicke and Roediger (2008) tested this with 40 Swahili-English word pairs. A week later, students recalled about 80 percent of words they had continued retrieving, compared with 33 to 36 percent when testing stopped after the first correct answer. Continued study alone did not produce the same benefit.

Rowland (2014) and Dunlosky and colleagues (2013) found support for testing and spaced practice across studies. Failed attempts can also help when followed by the answer, as Kornell, Hays and Bjork (2009) found and Metcalfe (2017) reviewed. This supports providing feedback after an error and another opportunity to practise.

Leaving time between attempts

An answer may be easy to remember immediately after seeing it. A later attempt tests whether it remains available after other material has intervened. Pyc and Rawson (2009) compared roughly one-minute and six-minute gaps between attempts to recall word pairs. One-week recall was 30 percent after the shorter gap and 54 percent after the longer one.

Useful gaps also depend on how long the material needs to be remembered. Cepeda and colleagues (2006, 2008) examined this relationship across studies and test delays. In a Spanish classroom study, Lindsey and colleagues (2014) found that review timed for each student produced better results on a later exam than either concentrating practice into one period or giving everyone the same spaced schedule.

The amount of spacing matters more consistently than a particular pattern of expanding gaps. Karpicke and Bauernschmidt (2011) and Nakata (2015) found limited support for treating expansion itself as the main advantage. These findings support spaced review, but do not establish Vozelu's exact intervals.

Choosing a review date

Vozelu calculates a separate review date for each word. It uses a set of mathematical formulas that estimate how the chance of remembering the word falls over time. Each answer changes that estimate, which changes when the word should appear again. This calculation is the job of the scheduler, the part of the app that plans reviews.

For example, recalling “silla” after several days gives the app evidence that the word can be remembered across that gap. Forgetting it gives different evidence and can bring the next review closer. These are estimates based on answers, not direct measurements of memory.

The formulas Vozelu uses come from an existing scheduling algorithm called Free Spaced Repetition Scheduler, or FSRS. FSRS keeps estimates of how difficult each word is for the learner and how long they are likely to remember it. Vozelu uses those estimates to set review dates.

The estimate applies to the word as practised in the app. A learner may recall its meaning from an English prompt but miss the same word in speech. The surrounding grammar can also affect an answer. A single estimate cannot distinguish all these causes.

Vozelu currently aims to ask again when it predicts a 95 percent chance of recall. An early local measurement recorded about 65 percent correct first attempts against a prediction of 95 percent. The prediction was too optimistic for those attempts. The measurement does not explain why. Possible causes include the demand of typing from memory, the exercises, answer marking, word selection and the model's fit. The higher target remains a provisional setting.

Selecting and recording practice

The first meeting introduces the word. Later exercises ask for it. A correct answer, a difficult success and a wrong answer affect the next review date differently. A retry after feedback provides practice but is recorded separately from the first attempt, since the learner has just seen the answer.

Reading or hearing a word records an encounter. It leaves the review schedule unchanged because the app has no answer from which to assess recall. This recording rule allows for learning through exposure while keeping it distinct from demonstrated recall.

Sessions select words whose review dates have arrived and allow room for new words. Recent difficulty reduces the number of new words introduced. Returning learners can receive words they are more likely to remember first. After a missed day, words remain due and the next answer is assessed using the time that has actually elapsed.

The app records attempts and predictions so their accuracy can be checked. Further evaluation needs to compare predicted and observed recall, time spent practising, and later retention. The review target, first follow-up delay, new-word allowance and rules for changing exercises remain choices to test with Vozelu's learners.

What learners practise

Vozelu asks the learner to produce a Spanish word from memory, most often in a sentence with one word missing. Later exercises add audio and whole-sentence translation. The learner submits an answer before seeing the correction.

For example, an early exercise can show “chair” and ask for “silla”. A later exercise can ask for the same word in “Mi ___ es roja.” The first practises the connection between meaning and word. The second adds the need to understand the sentence.

Recall and feedback

Rowland (2014) combined results from many studies and found larger testing benefits when learners supplied answers from memory than when they recognised them among options. The comparison depends on feedback. Kang, McDermott and Roediger (2007) found that short-answer practice outperformed multiple choice when learners received the correct answer afterwards. Without feedback, the order reversed in their experiment.

Multiple choice also produced learning benefits in Roediger and Marsh's (2005) study, alongside a risk of remembering incorrect options. Butler and Roediger (2008) found that feedback reduced that risk.

Vozelu distinguishes choosing an answer from supplying it. Normal practice asks for typed answers. If typing is disabled, some exercises offer choices. A correct choice counts as a less secure success when setting the review date. The grammar course asks learners to type the form being practised.

Starting with individual words

A sentence can make initial word learning harder by requiring the learner to understand surrounding language at the same time. Webb, Yanagisawa and Uchihara (2020) found vocabulary gains from word lists and flashcards. Their combined estimates favoured those activities over gap-filling on several measures, although results varied greatly between studies and some comparisons were inconclusive. Barcroft (2002, 2004) also found that additional processing demands could interfere with learning new word forms.

Vozelu uses a sentence gap as the standard first attempt. A word card without a sentence is used only for a word the app has no evidence for, when its meaning has few close neighbours, and for at most a quarter of a session. A learner's very first session follows a fixed script instead: four words are met, typed in a sentence, then heard. This combines practice on individual words with practice in context. The sequence has not been shown to teach meanings faster than continued word-card practice.

Changing the exercise

Practice moves from sentence gaps to audio and later to whole-sentence translation. This sequence is called the exercise ladder. A word reaches the audio exercise when the app estimates it will be remembered for longer, or earlier, after two correct typed answers in a row. The transition points are product settings rather than established measures of language ability.

An audio exercise asks for a missing word from a recorded sentence. For words the app expects the learner to remember well, it may initially hide the text. This practises identifying a word in a short recording. Understanding a conversation requires additional skills.

Whole-sentence exercises provide feedback across the answer. The result for the particular word being practised updates its review schedule. Errors on unfamiliar surrounding words remain visible without becoming separate vocabulary failures.

The sequence needs evaluation by exercise type. Remembering a written word for longer does not establish readiness for a new listening or grammatical task.

Checking answers

Vozelu checks the answer after submission. It leaves the text unchanged while the learner types, without colouring letters or revealing them one at a time. Asking to see the answer provides help and is recorded separately from unaided recall.

The checker must accept valid alternatives and distinguish typing mistakes from language errors. For example, a one-letter difference can be either an accidental keystroke or a different verb form. Rules for handling such answers need tests of the actual Spanish, separate from the research supporting recall practice.

A wrong answer receives feedback and returns for another attempt. Since the learner has seen the correction, that retry is kept separate from the first attempt when assessing recall.

Learning from sound and text

Shadowing asks the learner to repeat a recording aloud. Vozelu does not assess the speech in shadowing. In the dialogues, a Spanish speech recognition model checks which of the expected lines the learner said. At the normal setting every word of the line must be recognised, and a lenient setting judges only the whole line. The check does not score pronunciation and does not yet run on iPhones. The learner can also type the lines. Dictation asks the learner to write the whole recorded sentence. Hamada (2016) reported shadowing gains that differed by learner level. Kiany and Shiramiry (2002) found improved listening scores with added dictation, but did not test how much remained after a further delay.

Reading and listening can teach vocabulary without explicit word practice. Pellicer-Sánchez and Schmitt (2010) measured gains from reading a novel. Uchihara, Webb and Yanagisawa (2019) found a relationship between repeated encounters and learning across studies. Vozelu records these encounters separately because an encounter alone cannot show whether an individual can recall the word.

Reading and vocabulary practice

Vozelu offers Spanish texts with definitions available by tapping a word. A learner can also practise vocabulary from a chosen text before or after reading it.

Texts with familiar language can support faster reading and comprehension. Texts with unfamiliar words can introduce vocabulary, especially when help is available. Recommendations need to account for both purposes.

Estimating vocabulary coverage

Vocabulary coverage is the share of words in a text that a reader knows, counting each occurrence. If a reader knows 95 of the words in a 100-word passage, its coverage is 95 percent. Grammar, subject knowledge and the reader's ability to draw inferences also affect comprehension.

Research commonly discusses coverage near 95 or 98 percent. Laufer (1989) found a relationship between coverage and meeting a comprehension criterion. Hu and Nation (2000) found differences across coverage conditions. Their often-quoted 98 percent figure was estimated from the relationship in their results, rather than tested as a separate condition.

Schmitt, Jiang and Grabe (2011) found a gradual relationship without a sharp boundary. Kremmel and colleagues (2023) and Webb, Pellicer-Sánchez and Wang (2025) provide further reasons to avoid treating a particular percentage as a universal threshold. In beginning Spanish readers, Herman and Leeser (2022) found that the relationship depended on the type of comprehension measured.

Vocabulary counts also use different units. English studies often count word families, grouping a base word with related words such as “help”, “helpful” and “helpless”. Vozelu counts dictionary headwords, called lemmas. For example, “hablo” and “hablamos” belong to “hablar”. A vocabulary target expressed in word families cannot be transferred directly to Vozelu's count.

Limits of the current calculation

Vozelu's calculation counts words marked familiar through practice or an initial level estimate. It also credits names, function words such as pronouns and prepositions, a list of everyday object words, and words available for definition but excluded from vocabulary practice.

These additional credits can overstate knowledge. Being able to look up a word does not show that the learner knows it. Nor does an object's familiarity establish knowledge of its Spanish name. The displayed share therefore combines evidence about the learner with assumptions made by the app.

The credits were introduced to avoid making too much of the library appear inaccessible. A more informative estimate would separate demonstrated knowledge, assumed knowledge and words for which there is no evidence. That change remains to be implemented.

The labels “readable”, “close” and “locked” are recommendations. Every text can still be opened. The app can therefore give access without counting untested words as known. Answers to comprehension questions can help assess the recommendations, although they cannot validate each assumption in the vocabulary count.

What reading contributes

Nakanishi (2015) and Jeon and Day (2016) reviewed studies in which learners read substantial amounts of material. They found improvements in reading performance, including reading speed. Reading also develops vocabulary, with different aspects of word knowledge changing at different rates.

Waring and Takaki (2003) found that recognition and translation gains after a simplified reader declined over three months. After reading a novel, learners in Pellicer-Sánchez and Schmitt's (2010) study were better at recognising supplied meanings than recalling meanings without help. These findings distinguish outcomes that a single count of “words learned” would combine.

Direct word practice gives selected vocabulary further attention. Reading provides uses of those words within sentences and connected passages. The evidence supports both activities without establishing one order for every learner.

Connecting a text to practice

Tapping a word opens a definition selected for its use in the sentence. Finishing a text records encounters with some of its vocabulary. These actions leave review dates unchanged because neither supplies an answer from which to assess recall.

A practice session focused on a text uses its vocabulary. This lets the learner prepare for a passage or revisit words that interrupted reading.

The library contains journalism, travel writing, public-domain literature, episode transcripts and material for earlier learners. Texts written with controlled vocabulary and grammar are called graded readers. They can help beginners read connected passages before they can follow unsimplified writing.

The recommendations need evaluation against learners' experience. Relevant questions include whether they can follow the text, whether definitions help them continue, and whether practising its vocabulary improves a later reading.

Balancing the kinds of practice

Vozelu offers vocabulary practice, reading, listening, grammar and spoken dialogues. The learner chooses which to open, and every activity stays open however much the learner has already practised that day. The app does not divide the day into a fixed share of minutes for each activity.

The kinds of practice the app provides

Nation (2007) describes four kinds of language-learning activity. They are reading and listening for meaning, speaking and writing to communicate, studying vocabulary and grammar directly, and becoming faster with familiar language. He calls these the four strands and recommends roughly equal time for each. That recommendation is a course-design principle rather than an experimentally established optimum.

Vozelu's five activities cover these kinds of work unevenly. Vocabulary and grammar exercises both involve direct study. Reading and extended listening practise understanding messages. In the dialogues, the learner says their own lines of a short scripted scene aloud and hears the other characters' recorded replies. Each dialogue comes back one, three, seven, 21 and 60 days after the earlier visits. Typing or saying a prescribed line differs from speaking or writing an original message, which the app does not currently teach.

Learners can reread or listen again, but Vozelu also lacks a dedicated programme for becoming faster with familiar material. Research on skill learning indicates that improvement depends partly on the task practised, as discussed by DeKeyser (2007). Word recall can support communication while leaving a need for communication practice.

Learner choice

Self-determination theory examines how control over an activity and a sense of competence affect motivation. Deci and Ryan (2000) discuss the theory, and Noels and colleagues (2000) examine related associations in language learners. Patall, Cooper and Robinson (2008) found across studies that having a choice increased interest in the activity on average.

These findings support giving learners control over practice. A learner who listens to Spanish for an hour elsewhere needs a different balance from someone whose only practice is in Vozelu, and the app cannot see a book read on paper, a podcast played elsewhere or a conversation. Learners therefore set the balance between activities themselves.

Focused and mixed practice

A learner can focus a session on a text, topic, collection or grammar concept. The focus limits the words available to the session. A word retains its practice history when it appears under another topic.

Mixing exercise types instead of completing each type in a block is called interleaving. Brunmair and Richter (2019) found that its effects varied with the material. Nakata and Suzuki (2019) found better performance on a later test from mixed grammar practice, despite more errors during training.

Grouping vocabulary by topic, alternating grammatical structures and repeating one word are different arrangements. Evidence about one does not establish the value of the others. Vozelu needs comparisons using the exercises and sequences its learners encounter.

Listening instruction

Vozelu provides short audio exercises, text playback and longer episodes. These offer listening material. They do not yet provide a course in listening strategies, such as predicting a passage's subject and checking an initial interpretation against what follows.

Vandergrift and Tafaghodtari (2010) found benefits from instruction that helped learners predict, monitor and evaluate their understanding. Dalman and Plonsky (2025) report positive results across studies of listening-strategy instruction. Such instruction is a possible addition to Vozelu's existing audio.

Evaluating the balance

Time recorded in the app describes use but cannot establish that a balance between activities caused learning. Evaluation should examine whether learners continue practising and improve at the activities they intend to develop.

Practice without points or penalties

Vozelu records practice without awarding points or ranking learners against each other. It shows a streak of practised days, and rest days and streak freezes do not break it. Errors receive feedback and another attempt. This chapter explains those choices and the evidence that bears on them.

Rewards and competition

Rewards have different effects depending on the task and the measure of success. Deci, Koestner and Ryan (1999) found that some external rewards, such as money or prizes, reduced later voluntary engagement. Verbal praise had a different effect, and results depended on whether the task was interesting to begin with.

Gamification means adding game features such as points and leaderboards to another activity. Reviews by Sailer and Homner (2020) and Bai, Hew and Huang (2020) found positive learning effects. Hanus and Fox (2015) reported declining motivation in their study. Mekler and colleagues (2017) found increased activity without a corresponding increase in interest in the activity itself.

Vozelu displays words practised and answers given. It leaves out a separate points currency so that these measures retain their stated meaning. Without a leaderboard, another learner's pace does not determine the displayed result. These are design choices, not findings that rewards or competition prevent learning.

Missed days

A streak counts consecutive days of activity. Silverman and Barasch (2023) found that showing an intact or broken streak affected subsequent decisions. Aulagnon and colleagues' 2025 working paper reports increased platform use from highlighting streaks. Few pupils took its final test, which limits conclusions about learning.

Lally and colleagues (2010) found that one missed opportunity did not materially disrupt habit formation in their study. This does not establish the effect of a long absence or predict how a streak affects return visits to Vozelu.

In Vozelu, a day counts towards the streak once the learner finishes a session of at least eight graded answers. The learner can choose weekly rest days, which never count against the streak. Every five practised days earn a streak freeze, with at most two held at once, and a missed day that is not a rest day uses one automatically. A missed day without a freeze pauses the streak, and practising on two of the next three days restores it. Otherwise the next practised day starts a new streak, and the app does not report the loss. These rules tolerate isolated missed days, which is consistent with the finding of Lally and colleagues (2010), but their effect on return and on learning has not been tested.

After two or more missed days that are not rest days, or while the streak is paused, the next session is a shorter Quick session of eight reviews without new words, beginning with words the learner has answered correctly before. Vozelu keeps the learner's practice history, and the next answers update the review schedules using the elapsed time. Return rates after absences still need to be measured.

Errors and continued practice

An error in Vozelu leads to a correction and another attempt. It does not consume a life or end the session. This keeps practice available when an answer shows that the learner needs it.

Studies of rewards that participants can lose show behavioural effects without establishing a general learning advantage, as reported by Balart, Ezquerra and Hernandez-Arenaz (2022) and Levitt and colleagues (2016). Reports of anxiety about app mechanics also exist, but self-selected forum comments cannot establish its frequency among all learners.

The decision to keep practice available rests on the purpose of the exercise. An incorrect answer provides material for further work.

Review counts and goals

Vozelu assembles sessions from words whose review dates have arrived. It shows the planned session rather than an accumulating total that the learner must clear. The learner chooses the session size and can set a daily goal of one or more sessions. Practice remains available after that goal is met.

The reviewed research does not isolate the effect of displaying an unbounded review count. Hiding that count is therefore a design judgment. Evaluation needs to check whether this makes returning easier and whether learners still have enough information to plan their work.

Control over answers and pacing

In screen-based practice, the learner submits an answer before receiving feedback and chooses when to continue. The app leaves letters unchanged during typing. It records requests to see an answer separately from answers supplied from memory.

Chan and colleagues (2024) found that answer deadlines reduced accuracy during practice without changing the later testing benefit in the same way. This limits what can be inferred from faster or less accurate answers during a timed exercise. It does not establish that deadlines are always harmful.

Decisions to evaluate

Session length, visible review information, activity order and rules for changing exercises should be evaluated against their purposes. Learners need to understand their results, plan their practice and continue after errors. A design choice can be revised when it fails to support those needs.