Ideas for Langible

Anki exports

Closed-source platforms are a risk for users. At the same time, open-sourcing Langible at this stage would make it difficult to monetize, making it hard to recoup and reinvest development costs to build a better platform. However, users should not be locked in to Langible, and so we should allow users to export their data to Anki, which is open-source and has a large community.

Custom, user-created courses

Users should be able to curate and share their own custom courses using exercises with both sentences and custom flashcards. This would, for example, allow teachers to create courses for their students, and help learners focus on a specific topic.

Custom, user-created flashcards

Allow users to create their own flashcards, just like Anki. Later on, also allow users to share their flashcards with the community.

Integrated dictionary

We integrate Wiktionary into the Langible database, so that users can click on a word in an exercise sentence and get a definition. We use part-of-speech analysis to show only the relevant definitions for the word as it is being used in that sentence.

Leaderboards

Leaderboards are a great way to keep users engaged. We already have a point system which rewards users for completing exercises, but we should also have a leaderboard to show who has the most points. These leaderboards should be globa, and have daily, weekly, and monthly rankings.

"Minimal pairs"

This feature would show the user two similar words, such as "ti" and "tú" in Spanish, and ask them which is grammatically correct in a given sentence. This would help the user to learn to distinguish between similar words.

It would not be limited to just minimal pairs in the strict sense. Rather, any two words which are similar in some way could be used. For example, this mode could be used to help the user learn to distinguish between "hablar" and "decir", or "ser" and "estar" in Spanish.

To start with, a simple algorithm could be used to identify words which differ by only one letter. Alternatively, we could use ChatGPT to generate a list of words which are often confused by learners.

Multi-choice cloze exercises

While we already have the flashcard mode for cloze exercises, a multiple-choice variant could also be an option. This is the default mode in Clozemaster, which is a successful product.

Per-word proficiency

We keep track of the user's track record for each individual word. Every time the user encounters a word in a sentence, we keep track of whether they successfully completed that sentence exercise.

This gives us a rough idea of how well the user knows each word. This information can be used to, for example, pick which word should be the cloze in a cloze deletion exercise.

To improve the granularity of the tracking, we should separate words by their part of speech, for example, "to run" (verb) and "run" (noun) should be tracked separately. This would allow us to give the user more precise feedback on their proficiency.

A mini-forum for each sentence

When learning a language, it is often useful to discuss the sentences that you are learning. Right now, Langible is closer to Clozemaster or Anki, where the focus is on the sentences themselves.

Building a community around sentences would let users ask questions and discuss sentences with other learners.

Duolingo used to have a discussion forum for each sentence, which was appreciated by the users. It was removed because it was difficult to moderate. In Langible, we can use the same idea, and include an upvote-downvote system to help moderate the comments.

Swiping placement test

To gauge their current level of knowledge, the user is shown words in order of frequency, and presses "known" or "unknown" for each word. To speed things up, we can use a kind of "binary search" to skip forward until we find a word that the user doesn't know. When the user presses "unknown" three times in a row, we can assume that we have found their level of proficiency.

Label images in your target langauge

Given an image, use AI to find the objects in the image, and generate labels in the target language.

Create flashcards from images

The user uploads an image, and Langible uses AI to identify what is in the image. Langible then suggests a sentence for the user to learn, based on the content of the image, and the user's current level of proficiency.

This lets us add a new type of exercise to Langible, the image-based flashcard, which is fun and engaging, and probably helps with retention as the user would have both a visual aid in learning, and a personal experience in creating the flashcard.

Convert text to flashcards

Given a text, Langible finds which words the user has not yet learned, and queues them for learning.

User-directed learning

The user can choose which sentences they want to learn by entering a prompt in either their native or their target language. Langible then uses AI to find the sentences that match the prompt best, so that the user can choose to add those to the queue that they want to learn.

User-supplied translations

Allow users to contribute new translations to the database. This would allow the database to grow faster, and would also allow the community to correct mistakes in the translations.

We'd need to moderate the sentences, however, the biggest issue is that we would want to contribute these translations back into the Tatoeba corpus. It might be best to simply encourage users to contribute to Tatoeba instead.

Writing aid

A target-language writing aid feature. The user writes a short text in their target language, submit it to the community for feedback, and/or get feedback from an AI that checks for grammar and spelling mistakes.

• We could provide an optional writing prompt to help the user get started.
• We could expand this to also allow the user to record themselves speaking, and get peer feedback on their pronunciation.

Word order drill

Duolingo has a word order exercise, where the user is given a sentence in their target language and has to put the words in the correct order. This is a good exercise, particularly in languages with strict word order, and so we should add it to Langible.

Duolingo also has the reverse exercise, where the user is given a sentence in their native language and has to translate it into their target language, but that is probably a waste of time and we should not add it.