Everyone has been told to watch TV in the language they're learning. Almost nobody is told what to do when, three episodes later, they still can't name a single new word they picked up.

That gap isn't a discipline problem. It's a predictable consequence of how attention, comprehension, and memory actually work — and it's the specific problem BingeLingo is built around. This article walks through the research behind each part of the product: why subtitle modes exist at all, why a quiz appears after the credits, and why a word you met tonight comes back in a month.

Television really is good input — the evidence is not vague

Start with the part that holds up. Watching with subtitles does teach you vocabulary, and the effect is not small.

A meta-analysis covering thirty years of captioning research found large advantages for captioned video over uncaptioned video, both for listening comprehension (g = 0.99) and for vocabulary learning (g = 0.87) (Montero Perez, Van Den Noortgate & Desmet, 2013). And this happens without anyone trying to study: Peters and Webb showed that learners who watched a single full-length TV programme picked up measurable vocabulary at the level of both meaning recall and meaning recognition, purely incidentally (Peters & Webb, 2018).

So the raw material is good. The problem is what sits between you and it.

The comprehension threshold, and why most learners are under it

Comprehension is not a smooth gradient. Below a certain share of known words, a scene stops being input and becomes noise.

Webb and Rodgers analysed 88 television programmes — 264,384 running words — and found that knowing the most frequent 3,000 word families, plus proper nouns and marginal words, yields about 95% coverage of TV dialogue (Webb & Rodgers, 2009). But 95% is the assisted threshold. For unassisted comprehension of spoken text, Nation's estimate is a vocabulary of 6,000 to 7,000 word families (Nation, 2006).

Most people who put on a show in a new language are nowhere near 6,000 word families. That is the entire design constraint. Turn on target-language subtitles too early and you don't get immersion — you get a wall of text you decode instead of a story you follow. Turn on your own language and you get a story with no learning attached.

Both failure modes are real, and they're why BingeLingo ships three tracks instead of one.

Bridge mode: code-switching as a comprehension shield

Bridge keeps your native language as the base line and drops target-language words into it at the points where they occur. It looks like a typo the first time you see it. It is, in fact, a studied technique.

Labutov and Lipson built a system that generates exactly this kind of code-switched text for adult vocabulary learning, optimising which words to switch against a "learnability" metric — the insight being that a reader who never loses the thread keeps absorbing, while a reader who stalls stops (Labutov & Lipson, 2014).

The cognitive-load evidence points the same way. Yuan compared four subtitling sequences with 162 upper-intermediate learners and found that starting with L1 subtitles and then moving to bilingual ones produced the best form and meaning recall and the lowest measured cognitive load (Yuan, 2025). Protecting comprehension first isn't a compromise; it's what makes the vocabulary stick.

There is also direct evidence that you don't need full captions to get the benefit. Montero Perez, Peters and Desmet compared keyword captions against full captions and no captions, and found keyword captions matched full captions on detailed comprehension — though learners themselves rated them as less useful (Montero Perez, Peters & Desmet, 2014). That mismatch between what works and what feels like it's working is a recurring theme below.

The honest caveat: partial-text subtitles are not universally good. Cojean and Martin tested occasional-keyword subtitles on video learning and found that keywords hurt content memorisation, comprehension, and time on task (Cojean & Martin, 2021). That study was about learning subject matter in a language you already speak, which is a different job from acquiring vocabulary — but it is a real result, and it is why Bridge caps how many words it injects per hour rather than annotating everything it can.

Immersion mode: difficulty that's supposed to be uncomfortable

Immersion flips the base line to the language you're learning, with your own language stepping in only where a line would otherwise stop you dead.

This is deliberately harder, and the harder-is-better claim here is specific rather than macho. Bjork's desirable difficulties framework draws the line: when retrieval is effortless, practice does almost nothing for durable memory; when retrieval is effortful but still succeeds, it produces large gains in how well the item is stored (Bjork, 1994). The point is effortful and successful — which is why Immersion still leaves the safety net in.

Eye-tracking gives a sharper reason to stretch. Wang and Pellicer-Sánchez tracked 112 learners across bilingual subtitles, captions, L1 subtitles and no subtitles. Bilingual subtitles beat captions for meaning recognition — but were worse than captions for form recognition, and participants spent more time reading the translations than the target words. Critically, attention to the target words predicted learning gains; attention to the translations did not (Wang & Pellicer-Sánchez, 2022).

Translation is a crutch your eyes reach for automatically. Immersion removes most of it, so your gaze lands where the learning is.

Dual mode: useful, and the most likely to quietly waste your evening

Dual shows both languages as two synced lines. It's the mode people ask for first, and the one the research treats most cautiously.

A recent eye-tracking study of Spanish–English bilinguals found that when dual subtitles appear, gaze shifts strongly to the text — with a marked preference for whichever line sits on top, regardless of which language it is. Comprehension improved for L2 audio but was unaffected for L1 audio, suggesting dual subtitles work mainly as a compensatory aid under load (Romero-Ortells, Perea & Duñabeitia, 2026).

That "position beats language" finding is the warning. Reading habits, not learning intent, decide where your attention goes. And there is a long-standing cognitive-load explanation for why more text isn't automatically better: presenting redundant information across channels forces learners to split attention and reconcile sources, consuming working memory that essential processing needed (Kalyuga, Chandler & Sweller, 1999).

Dual is genuinely good for comparing sentence structure deliberately. It is a poor default for a whole season — which is why it's the one BingeLingo mode that doesn't collect vocabulary for you.

Watching is encoding. It is not retrieval.

Here is the fact that most "learn by watching" advice skips: recognising a word on screen and producing it later are different abilities, and only one of them is trained by watching.

Roediger and Karpicke's test-enhanced learning experiments are the canonical demonstration. Repeated testing beat repeated re-reading on a delayed test a week later — 61% versus 40% — even though the re-reading group was re-exposed to all the material while the testing group only revisited what it could actually recall (Roediger & Karpicke, 2006). The effect reverses at very short delays, which is precisely why re-reading feels more effective while being worse.

The vocabulary-specific version is Laufer and Hulstijn's Involvement Load Hypothesis: retention tracks how much need, search, and evaluation a task demands. In their study, writing with the target words beat filling them in, which beat simply reading them (Laufer & Hulstijn, 2001).

This is what the post-episode recap is for. Every Bridge or Immersion track you watch leaves a twelve-question deck for that title, mixing recognition, recall, listening, and cloze items — the cloze questions rebuilding the actual subtitle line with the word blanked out, so you retrieve it from the scene rather than from a list.

Why the recap expires after a week

The recap deck stays playable for seven days, then closes. That's a memory decision, not a scarcity tactic.

While the episode is fresh, the words still carry their scene — the character who said it, the moment it landed. A month later that context is gone and the same twelve questions are just a vocabulary list, which is a job the vocabulary bank already does better.

The repetition evidence here is genuinely mixed, and worth stating plainly. Lo compared immediate repeated viewing against viewing spaced a week apart, for lower-intermediate learners watching dual-subtitled documentaries. Both beat single viewing — but immediate repetition outperformed spaced repetition on both immediate and delayed tests, which runs against the standard spacing finding. The suggested explanation is proficiency: below a certain level, learners need the massed exposure before spacing can help (Lo, 2024). A recap you do within the week sits close to that massed condition.

The vocabulary bank: expanding intervals, by the numbers

Words that survive the recap move into your vocabulary bank, each one still carrying the subtitle line you met it in. From there, review intervals expand: 1 day, then 3, 7, 14, 30, and 60. Get one wrong and it drops back to the start. Clear five steps and it's marked mastered.

The shape of that ladder comes from the distributed-practice literature. Cepeda and colleagues synthesised 839 assessments across 317 experiments and found that the optimal gap between study episodes is not fixed — it scales with how long you need to remember something. The longer the retention interval you're aiming for, the longer the gap between sessions should be (Cepeda, Pashler, Vul, Wixted & Rohrer, 2006). An expanding schedule approximates that without asking you to declare a deadline.

Keeping the original subtitle line attached to each word matters for the same reason cloze questions do. The line is a retrieval cue tied to a specific evening of your life, which is a far stronger hook than a dictionary entry.

The part no study can fix

Every mechanism above assumes you come back. Most language-learning tools die at that assumption, not at the pedagogy.

Self-determination theory identifies three conditions under which motivation sustains itself: autonomy, competence, and relatedness (Ryan & Deci, 2000). Watching a show you chose, in a mode you picked for the mood you're in, is unusually well supplied on autonomy — you were going to watch something tonight anyway. That is the bet BingeLingo makes: put the learning inside an activity you already want to do, rather than making it compete with that activity for the same half hour.

The streak, the freezes, and the XP all serve that one thing. Not because points teach vocabulary, but because the schedule above only works if someone is there to run it.

What this does not claim

  • Subtitles are not comprehensible input on their own. They lower the threshold; they don't remove it. A show far above your level stays above your level.
  • Incidental gains per episode are real but modest. Peters and Webb measured learning from a single programme, not fluency from it.
  • The evidence is dominated by specific populations — largely Dutch- and Chinese-speaking learners of English, at intermediate levels, under study conditions. Your language pair may behave differently.
  • Some of it conflicts. Keyword subtitles helped in one study and hurt in another; spacing lost to massed practice in one viewing study while winning across hundreds of verbal-recall experiments. The design choices above are informed readings, not settled facts.

References

  1. Bjork, R. A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe & A. Shimamura (Eds.), Metacognition: Knowing about Knowing (pp. 185–205). MIT Press.
  2. Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380.
  3. Cojean, S., & Martin, N. (2021). Reducing the split-attention effect of subtitles during video learning: Might the use of occasional keywords be an effective solution? L'Année Psychologique, 121(4), 417–442.
  4. Kalyuga, S., Chandler, P., & Sweller, J. (1999). Managing split-attention and redundancy in multimedia instruction. Applied Cognitive Psychology, 13(4), 351–371.
  5. Labutov, I., & Lipson, H. (2014). Generating code-switched text for lexical learning. Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, 562–571.
  6. Laufer, B., & Hulstijn, J. (2001). Incidental vocabulary acquisition in a second language: The construct of task-induced involvement. Applied Linguistics, 22(1), 1–26.
  7. Lo, S. (2024). Vocabulary learning through viewing dual-subtitled videos: Immediate repetition versus spaced repetition as an enhancement strategy. ReCALL, 36(2), 152–167.
  8. Montero Perez, M., Peters, E., & Desmet, P. (2014). Is less more? Effectiveness and perceived usefulness of keyword and full captioned video for L2 listening comprehension. ReCALL, 26(1), 21–43.
  9. Montero Perez, M., Van Den Noortgate, W., & Desmet, P. (2013). Captioned video for L2 listening and vocabulary learning: A meta-analysis. System, 41(3), 720–739.
  10. Nation, I. S. P. (2006). How large a vocabulary is needed for reading and listening? Canadian Modern Language Review, 63(1), 59–82.
  11. Peters, E., & Webb, S. (2018). Incidental vocabulary acquisition through viewing L2 television and factors that affect learning. Studies in Second Language Acquisition, 40(3), 551–577.
  12. Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
  13. Romero-Ortells, I., Perea, M., & Duñabeitia, J. A. (2026). Hearing once, reading twice: How dual subtitles shape visual attention in bilingual viewing. Bilingualism: Language and Cognition, 1–11.
  14. Ryan, R. M., & Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55(1), 68–78.
  15. Wang, A., & Pellicer-Sánchez, A. (2022). Incidental vocabulary learning from bilingual subtitled viewing: An eye-tracking study. Language Learning, 72(3), 765–805.
  16. Webb, S., & Rodgers, M. P. H. (2009). Vocabulary demands of television programs. Language Learning, 59(2), 335–366.
  17. Yuan, M. (2025). Effects of the sequential use of L1 and bilingual subtitles on incidental English vocabulary learning: A cognitive load perspective. British Journal of Educational Psychology, 95(2), 565–577.