Why we correct you this way.

01 / Feedback

First, a chance to find your own repair.

You produce the Spanish yourself. When a meaningful problem is detected, the tutor offers direction before supplying a finished sentence. The aim is to help you connect and use the language you know.

A Bayesian meta-analysis of 52 controlled written-feedback studies found moderate benefits that persisted over time. Direct corrections, indirect corrections, and explanations of language rules had similar average effects. Several approaches can help; the findings do not identify one universal winner. [1]

A separate meta-analysis of 15 classroom studies involving 827 learners found durable benefits from oral corrective feedback. Prompts encouraging learners to generate a repair showed larger effects than recasts, which reformulate the answer for them, particularly on freely constructed responses. [2]

02 / Guidance

Withhold the answer.
Give a useful direction.

“Something is wrong” is not enough. Our first hint is designed to identify a relationship or meaning to reconsider. Every required repair should receive an actionable cue from the start. Later hints should offer more help with those problems, rather than introduce surprise requirements.

Guidance should respond to what you have already understood. If you fixed the number but still need help forming the verb, the next hint should address the form. Correct work should be protected, not dismantled.

A 2017 computer-assisted study investigated increasingly explicit feedback for a Chinese grammatical construction. It provides a relevant example of graduated help, while also showing the need for human intervention when the automated system failed. [3]

03 / Choice

Revealing the answer is part of learning.

Sometimes a cue helps you retrieve something you know. Sometimes you need to learn a form you have not acquired yet. Direct correction is legitimate teaching, and the written-feedback research supports its value alongside other approaches. [1]

You can reveal the answer at any point. The original response plus three retries creates a bounded opportunity to use help, rather than an unlimited guessing loop.

No research establishes that four is better than three or five. We should keep evaluating where repairs happen, when learners ask for the answer, and what they can produce later.

Why feedback comes after each sentence

This is focused sentence practice, so feedback arrives while your wording and intended meaning are still fresh. Research on feedback timing uses varied definitions and finds results that depend on the task, target, and assessment. It does not establish that immediate correction is always better. A fluency-focused conversation may need a different rhythm. [6]

You remain in control

Try again, reveal the answer, practice it from memory, or skip reconstruction. A review of 97 studies on computer-mediated language feedback highlights the importance of task and learner context. Our precise interface choices are practical safeguards for agency and frustration, not separately proven interventions. [7]

04 / Remembering

Seeing it is a start.
Using it later matters more.

After a model is shown, the primary next step is to hide it and express the thought once from memory. Another valid wording is welcome. “Skip for now” remains available without a penalty.

Second-language vocabulary experiments found that retrieving forms, with corrective feedback, improved later retention compared with restudy. This supports the principle behind reconstruction. But vocabulary experiments do not directly establish the benefit of reconstructing complete grammatical sentences in our app. [4]

Problem structures return through spaced review. A 2022 meta-analysis of 48 experiments, 98 effect sizes, and 3,411 learners found a medium-to-large overall benefit from spacing. Longer spacing was especially helpful on delayed assessments. [5]

The important next question is whether you can produce the underlying structure in a new sentence, later, without seeing the old answer. Immediate repair, supported recall, and later independent use are different kinds of evidence.

05 / Trust

The tutor has responsibilities, too.

Open-ended language is not a single-answer exercise. Different vocabulary, word order, or regional forms can express the same thought correctly. Feedback should preserve your meaning and distinguish genuine errors from optional preferences.

Automated tutors can reject valid language or miss a necessary repair. Research on computer-mediated feedback, including graduated-feedback systems, documents limits in automated diagnosis and context. [3] [7]

• Accept valid alternatives and regional forms.
• Separate required repairs from optional style.
• Preserve the thought you were trying to express.
• Restore an attempt when the tutor confirms that earlier guidance was incomplete or misleading.

Restoring an attempt compensates for the tutor’s failure. It does not make the feedback infallible or prove that its accuracy has improved.

06 / Perspective

Supported principles.
Honest limits.

These are our editorial judgments about how the research relates to this design, not formal evidence grades for the app.

What we can reasonably say

Spanish, Personally uses an evidence-informed process that combines sentence production, graduated corrective feedback, access to explicit answers, supported reconstruction, and spaced review.

We do not claim that this exact flow is scientifically proven, that self-correction always beats direct correction, that using every retry is better than revealing, or that immediate recall demonstrates mastery. AI feedback can be wrong, and no single approach works equally well for everyone.