Live Corrections

Every language learning app has the same dirty secret: somewhere in their database is a translation that is subtly wrong, stiff, outdated, or just weird. Nobody says "I am extremely well, thank you" in casual Spanish. Nobody actually orders food the way textbooks teach you to. The word the app uses for "car" is the formal one your grandmother prefers, not the word your friends use.

Most apps never fix these, because most apps don't have a fast way for players to report them and an even faster way for someone to act on the report. Word Exchange Plaza does, and it pays you for using it.

What Live Corrections Actually Is

Live Corrections is a crowdsourced language corrections system built directly into the game. Any time you see or hear a word, phrase, or translation that looks wrong, you can tap a correction button, tell us what's wrong, and propose a better version. The submission gets reviewed — often by another player, sometimes by us — and if it holds up, the fix ships to everyone.

In exchange, you earn points. Real points. They count toward your weekly leaderboard position and, depending on the correction, can be worth more than finishing a drill round. The principle is simple: improving the course is worth at least as much as consuming it.

Why This Matters More Than It Sounds

Language is a living thing. Textbooks are dead. Every language learning app is trying to bridge that gap, and every one of them fails in the same place: the moment a phrase shipped in the database stops matching the way real people actually speak, there is nobody to notice except the learners. And learners, in most apps, have no way to say anything about it.

Live Corrections is our answer to that. The learners are the best possible quality filter — they're the ones who'll actually encounter every phrase, often more than once, and they include native speakers, heritage speakers, and intermediate learners who can all spot different kinds of problems. We built a pipeline to channel all of that directly into the database.

What the Crowdsourcing Research Says

The most-cited piece of evidence on whether crowdsourced quality can match expert curation is the December 2005 Nature study comparing Wikipedia and Encyclopædia Britannica articles in the natural sciences. Subject-matter experts reviewed 42 article pairs and found that Wikipedia averaged about four errors per article and Britannica about three — a measurable but small difference, and one that has narrowed in the two decades since^[1]. The takeaway is not that crowds beat experts. The takeaway is that a sufficiently engaged crowd, with a low-friction reporting mechanism and a basic review layer, produces work that is close to expert quality and improves much faster than expert-only work as more people show up.

Language data has the additional advantage that the relevant "experts" are not rare. Native speakers of Hindi, Arabic, French, Spanish, and Norwegian number in the hundreds of millions, and a sizable fraction of any given course's player base will include native or heritage speakers of the target language. Live Corrections is a way of acknowledging that the most reliable judgment of "is this how a person actually says this" comes from people who say it every day, not from a contractor on the other side of the world.

This is also why we are honest about Live Corrections being a complement to, not a replacement for, professional content review. The team still does first-pass review on every item before it ships. The crowd catches the things the team missed, the things the team got wrong, and the things that change over time as a language evolves.

How the Correction Flow Works

Here's the loop from the inside:

1. You're playing a normal round of hands-free practice and something sounds off.
2. You tap the correction icon on that item.
3. A short form asks: what's wrong? (Translation, audio, script, formatting, other.) What do you propose instead?
4. You submit. The item is flagged in the queue.
5. Reviewers — other players who've earned review privileges, plus the team — look at the submission.
6. If accepted, the fix goes live for everyone, your name goes on the contributor list for that item, and you receive points.

The flow is intentionally fast. You should not have to leave the game to file a correction. The whole thing lives inside the round so the cost of reporting something is basically zero.

What You Can Correct

Most submissions fall into a handful of buckets:

• Wrong translation. The English gloss doesn't match the target-language phrase.
• Stiff or unnatural phrasing. The phrase is technically correct but nobody actually talks like this.
• Audio problems. The speaker mispronounced something, or the clip cuts off, or the audio doesn't match the text.
• Script or formatting issues. Wrong diacritics, wrong direction, wrong case.
• Register mismatch. The phrase is marked casual but is actually formal, or vice versa.
• Missing context. The phrase is correct in one situation and wrong in the one the game is using it in.

Earning Points by Improving the Course

Here's the part that turns this from a nice-to-have into a real feature: accepted corrections earn points. Points feed directly into the weekly leaderboard, which means a learner who submits three good corrections in a week can move up the rankings as much as a learner who grinds drills. Contribution is on equal footing with consumption.

Point values depend on the severity and quality of the correction. A typo fix is a small reward. A full rewrite of a stiff phrase that several other players flag as better is a larger one. A native speaker who identifies and fixes a systematic problem across many items earns a lot. The goal is to make the point economy feel fair to the people doing the hard work.

Community as Co-Author

Live Corrections is the practical expression of our build-in-public philosophy. The plaza is not a finished product being polished in a closet. It is a living database being improved by the people who use it every day. Native speakers keep it honest. Intermediate learners keep it realistic. Beginners keep it humane — they're the ones who notice when an explanation assumes too much.

It also makes the app harder to mess up. No one person — not even us — can catch every problem across five languages and thousands of items. Thousands of learners, each catching one or two problems, can. The math works in our favor, and yours.

When Two Native Speakers Disagree

One thing the alpha has surfaced — and a thing we did not expect to be as common as it has been — is the case where two native speakers genuinely disagree about whether a phrase is correct. This happens for predictable reasons. Spanish has a different default register in Mexico City than in Buenos Aires. Hindi vocabulary in Delhi is not always Hindi vocabulary in Mumbai. French formality cues in Paris and Montreal split in non-obvious ways. There is no single "correct Spanish" the same way there is no single "correct English" — there is just the speech community you are aiming at.

The plaza handles this by treating dialect-level disagreements as tags rather than as wrong-versus-right judgments. When two native speakers from different regions both submit corrections that are valid in their region, both get accepted, and the item gets a region tag so future learners can see it. Over time the goal is to let learners select a target dialect at the deck level (Castilian Spanish vs. Mexican Spanish vs. Rioplatense, for instance) and have the system serve the right variant.

This is not solved yet. It is on the roadmap. But the principle — that the database should reflect how language actually is, not how a single textbook author thinks it should be — is the core of why Live Corrections exists in the first place.

A Note on Reviewing

Not every correction is accepted. Some corrections are wrong. Some are matters of dialect preference where both options are valid. Some come from well-meaning learners who haven't seen enough of the target language to know whether their proposed fix is actually better. The review layer exists to handle all of that. The review layer is also a place where experienced players can earn points — reviewing good corrections efficiently is itself a contribution to the plaza.

We're working on tuning the review system as we go, and — as with everything else — the fastest way to improve it is to use it, find what's broken, and tell us. That's the loop.

One concrete example from the alpha: an early Hindi item shipped with the formal pronoun for "you" in a context that would normally use the casual one. A native-speaker player flagged it within the first week, proposed the casual version, and the fix shipped to every Hindi learner in the plaza two days later. That same player has since submitted dozens of similar corrections and sits comfortably in the top of the per-language leaderboard — not because they grind drills, but because they sharpen the course every time they play. That is what Live Corrections is for.

Step into the plaza, play a few rounds, and the next time something sounds off, submit a correction. It's the fastest way to shape the app into the language learning tool you actually want.