Language learning tools should prioritise the intertwined nature of languages and cultures, valuing learning from their respective communities. This project stems from the idea that YouTube videos can be valuable resources for such language learning. However, there's a gap in understanding which videos best meet the needs of learners at various proficiency levels. My goal was to start bridging this gap by interviewing language experts to identify video features that correlate with language proficiency and using machine learning to classify these videos.

I started with a problem area and looking at what product could fix it and came up with an app idea. I looked at what components would be needed to build that product, in my case an algorithm that would tell you who the video is suited for, eg an intermediate learner. From here, I knew I had to break this down into a simpler task, so decided to build a classifier. Also, not knowing much about language teaching, I felt expert interviews were necessary. My data is a catalogue of rated language videos used by language learning apps such as Busuu.

These videos are not made by Creators but but are curated by a team of language experts, directors and actors. However, they are made to mimic creator content. The main contributor was seeing the process as a whole, through a design thinking perspective where I was prioritising the user, and the hypothetical people who would own the videos. This allowed me to focus on doing research on and building something useful for language teachers and learners, not just building it for the sake of innovation or because it works.

Without synthesising the insights from experts, I wouldn’t know about the database, how experts feel about AI/ML used in language learning and most importantly I would not have been able to extract the relevant data for the model.

A website  - https://zuzanas-marvelous-site-2ceca4.webflow.io/