An interesting startup from EPFL that is incorporating [[Spaced repetition|spaced repetition]] for language learning in more natural ways, that is altering words with the target language as you engage with website. Found at [timely-learning.com](https://timely-learning.com/) (as of 10/10/2025). One of the primary barriers to traditional forms of spaced repetition is the formulation of flashcards and engaging with them. For the former, time costs become lower as we use cloze deletion, and for the latter engaging with flashcards become easier with [[Anki]] mobile that can be used whilst commuting or in short coffee breaks. Timely takes it another step further, that is incorporating learning and repetition with the natural flow of engaging with text online. There are some other applications that try to follow the ideal of low effort and natural integration, for example [LingoPie](https://lingopie.com/) that integrates into Netflix but lacks control over the content, or Podcastify (that a couple of us made at Lauzhack 2024) that generates podcasts with spaced repetition of words, with feedback through smart watch integration to blend more seamlessly into life. It also takes the main benefit of [[AJATT]] whilst tackling its biggest flaws of lacking personalisation. However, a core issue that arises with a variety of applications is the segmentation of spaced repetition decks that ideally should have some kind of centralised source for a shared [[Student model|student model]] - but with timely having normal flashcard functionality, it could be a possibility to make a complete switch to relying on timely for language acquisition vocab. Below I will denote ideas that may have some relevance to the app: - 23/10/2025 - An interesting research project would be to investigate effects on task performance. That is, does its [[Context switch|context switching]] get in the way to the extent that it impacts our ability to perform tasks or learn other knowledge. If so, it may be useful to have different modes of interaction that could activate at different times of the day. For example, I may not want it to interfere with my work hours, but it could be okay during my leisure time. Could also use modes on different devices, for example, my phone is primarily for leisure and hence I would be okay with context switching. - 10/10/2025 - The gradual takeover of the target language within the text. Hence, going beyond single word replacements for the harder words that are desired, but over time more and more of the words turn into the target language as you know them until almost all gets translated into target language. Hence, all content that you consume will be a blend of your native language and target language, but with the perfect mix associated with your exact skill level in the language + [[Zone of proximal development|zone of proximal development]]. - 10/10/2025 - Integrate into video subtitles! - 10/10/2025 - At OS level so it works outside of the web, such as when I use [[Obsidian]]. - 10/10/2025 - Extensibility so other people can add plugins that tie into the same [[Student model|student model]]. - 10/10/2025 - Storing grammar rules as [[Knowledge component|knowledge components (KCs)]] that can be instantiated through the text when the opportunity arises, and then the [[Spaced repetition algorithms|spaced repetition algorithm]] updates on the underlying grammar rule. Thereby allowing spaced repetition for 'variable-variable' KCs (see [[The knowledge-learning-instruction framework|the KLI framework]]).