When I first built TubeVocab, I thought the core loop was obvious: watch a YouTube video, click an unknown word in the subtitles, save it as a flashcard, and review it later with spaced repetition.
That worked well for recognition. Learners could see a word again and remember what it meant. But recognition is only half of listening. The harder part is hearing a messy sentence at native speed and reconstructing what was actually said.
That gap is why I added dictation practice after the flashcard system was already working.
Recognition was too easy
Interactive subtitles are helpful because they lower friction. If you hear a phrase like "it turned out to be harder than expected," you can pause, click "turned out," and save the phrase before the context disappears.
The problem is that subtitles also hide weakness. A learner may think they understood the sentence because they read it while listening. Remove the text, and the same sentence becomes a blur of reductions, linking sounds, and weak forms.
In product terms, the first version of TubeVocab solved capture and review. It did not fully test auditory decoding.
Dictation exposes the exact failure point
Dictation looks old-school, but it gives unusually clear feedback. If the learner types:
I gonna go there
when the speaker said:
I'm going to go there
the mistake is not just spelling. It reveals that the learner missed contraction, grammar, and connected speech in one short phrase.
That is more useful than a generic "wrong" answer. It tells the system which part of the listening pipeline failed:
- The learner did not hear the reduced "I'm."
- The learner recognized the meaning but not the grammar.
- The learner may need the full sentence, not just a single-word flashcard.
Once I saw enough of these mistakes, dictation stopped feeling like an add-on. It became the missing bridge between watching and active recall.
The useful unit is a sentence, not a word
Single-word flashcards are still useful for vocabulary, especially for low-frequency words. But listening practice should usually operate at sentence or phrase level.
Words change shape in real speech. "Want to" becomes "wanna." "Did you" becomes "didja." Function words get swallowed. Stress moves toward the content words. A dictionary entry cannot teach that alone.
The better loop is:
- Watch a real YouTube clip.
- Save useful words or phrases from the subtitle line.
- Review the word meaning later.
- Revisit the original sentence as dictation.
- Compare the typed answer with the actual subtitle.
That keeps the vocabulary tied to the original listening context instead of turning it into a detached word list.
I avoided making it a punishment mode
The biggest design mistake with dictation is making learners type long paragraphs. It becomes tiring fast, especially on mobile.
I kept the unit short: one subtitle line or one compact sentence. The goal is not transcription endurance. The goal is to notice where sound and text diverge.
I also try to avoid marking every tiny punctuation difference as failure. If the learner hears the words correctly, punctuation should not dominate the result. The system should be strict about missing words and substitutions, but forgiving about cosmetic differences.
That makes dictation usable as a daily practice loop instead of a school exam.
The product lesson
The surprising lesson was that flashcards made dictation more valuable, not less.
Flashcards handle delayed memory. Dictation handles real-time listening. You need both if the goal is usable English rather than a bigger saved-word count.
For an ESL tool built around YouTube, that matters. Real videos are full of interruptions, jokes, unfinished sentences, and background noise. A learner who can recognize a saved word in a clean card still needs practice hearing it inside that chaos.
That is the direction I want TubeVocab to keep moving: not just "save words from videos," but turn real video into a complete vocabulary and listening practice loop.
Top comments (0)