Learn a Language From Video Subtitles, Not Flashcard Drills

The strongest evidence-backed idea in second-language acquisition is that people acquire language mainly by understanding messages: input that is meaningful and slightly above their current level. Stephen Krashen's input hypothesis put this most famously, and while the field has argued about the details for decades, the practical conclusion has held up: large amounts of comprehensible, engaging input drive vocabulary growth and listening ability in a way that isolated drills do not.

Video is unusually good input. It is spoken at native speed with native reductions and rhythm. It is contextualised (you can see who is talking to whom and why, which is a huge aid to comprehension). And it is engaging enough that you will actually do a lot of it, which matters more than any single technique. Subtitles are the scaffold that makes video comprehensible before your listening catches up.

The failure mode of subtitle learning is reading the translation and ignoring the audio. Your eyes are faster than your ears, so given English subtitles you will read and stop listening. The fix is to use target-language subtitles (not translations) and to interact with them: pause when a line is unclear, look up the word that blocked you, and save it together with the sentence it appeared in.

Saving the whole sentence matters. A word learned as an isolated translation pair is fragile; a word learned as "the thing that person said in that scene" is attached to a memory, a collocation, and a sound. This is why Nativra's cards carry the sentence and a timestamp back to the video, and why the recommended workflow is a first pass with subtitles and a second pass, later, without.

• Use target-language subtitles, not your native language.
• Pick content where you understand roughly 80–90% already.
• Pause on unknown words; save them with the sentence, not alone.
• Re-watch the same segment without subtitles within a few days.
• Review saved words with spaced repetition rather than re-reading lists.

Input only helps if it is comprehensible. A rule of thumb from extensive-reading research is that you should recognise around 95–98% of the words in a text to read it comfortably; for video, where visuals help, learners often tolerate somewhat lower coverage, but content where you catch one word in three is not input, it is noise. Nativra's CEFR grading tags each subtitle's vocabulary by level so you can see, before watching, roughly what proportion of a video sits above your current level.

Start with content that has clear speech and a predictable structure (interviews, explainers, cooking channels, vlogs with a single speaker) before moving to drama with overlapping dialogue and slang.

Nativra is a study interface for video and subtitles. It fetches YouTube captions or accepts your own subtitle files, synchronises the text to the audio, provides click-to-lookup with CEFR tagging, keeps your saved words with context, and schedules reviews with the SM-2 algorithm or exports them to Anki. It does not host or download videos, and it does not generate lessons for you: the content is whatever you choose to watch.

The core (import, read-along, saving, SRS, export) works without an API key. Contextual explanations and translations beyond the dictionary lookup use your own AI provider key with no markup.