The research

Dilo did not invent its method. It borrowed the loop that the institutions training adults fastest have used for decades, and checked it against the second-language research. This page says what the evidence supports, names each source so you can read it yourself, and marks where Dilo made a choice the evidence does not settle.

Speaking first

What the evidence says: understanding a language does not make you able to speak it. Canadian students who spent years in French immersion reached near-native listening and still had large gaps in speaking; Merrill Swain's output hypothesis explains why. Producing a sentence forces you to notice what you cannot yet say. FSI's own retrospective (Jackson and Kaplan, 1999) says that focused practice of some kind, including drills, appears necessary for almost all language learners to develop confidence and automatic language use.

Dilo's choice: every drill is spoken, and speech recognition checks what came out of your mouth. There is no tapping words into order.

Drilled chunks, used at once

What the evidence says: FSI's Spanish materials follow Earl Stevick's "microwave" format, so called because the step from mimicry to communication is short. Each cycle drills a few phrases to the point of automatic production, then puts them to real use with real names, events and objects, minutes later rather than weeks later. A 2019 meta-analysis of task-based language teaching by Bryfonski and McKay, across 52 studies, found a strong positive effect for learning built around real tasks.

Dilo's choice: one cycle is six to ten chunks for one real task, about ten minutes of spoken drill, then an unscripted roleplay of that task in the same session. The length of the cycle and the number of chunks are Dilo's numbers, not the sources'.

A small vocabulary, in the right order

What the evidence says: Mark Davies' 20-million-word corpus of Spanish shows the top 1,000 words cover about 88 percent of speech and the top 2,000 about 93 percent. Each thousand after that adds one to three percent.

Dilo's choice: 1,500 to 2,000 words across the ladder, verbs and function words first, learned inside phrases you will say.

Spaced review

What the evidence says: spaced review beats massed review in study after study, including Cepeda and colleagues' 2008 study with more than 1,350 participants, which found the best gap between reviews grows with how long you want to remember.

Dilo's choice: earlier chunks come back spoken, and roughly seventy percent of a session is review. The ratio is a design choice.

Correction

What the evidence says: correction helps, and timing matters. Li, Ellis and Shu (2016) found that beginners gained nothing from feedback delivered only after the task, while correction during or right after the attempt worked.

Dilo's choice: a feedback block after each roleplay, with what you said, what a speaker would say, and why. Whether that timing is right for beginners is a question the beta tests. Whether the corrections are right is not proved yet, see below.

An oral check before moving on

What the evidence says: Peace Corps gates swear-in on a live oral interview, typically at Intermediate-Mid, after 10 to 12 weeks of about four hours a day in small groups. A Peace Corps Office of Inspector General evaluation of the Philippines post reports one cohort, learning Tagalog, in which 58 of 63 trainees reached that level. In our reading of the market, no mainstream app blocks progress until the learner passes an unscripted oral check.

Dilo's choice: a three to five minute unscripted check between levels, scored against can-do criteria. It is not in the beta; it arrives after the loop has been measured.

Adults, hours and honesty

What the evidence says: FSI's students are adults, mean age 41, most starting from zero. Jackson and Kaplan write that time on task and intensity are crucial and that FSI's attempts to shorten its programs have not worked. The Defense Language Institute's own evaluation of an internal immersion program found no significant gain in test scores once it accounted for who was selected. Serrano's studies of intensive and extensive courses found the same hours concentrated in weeks outperform the same hours spread over months. Cambridge's guided learning hours put B1 at roughly 350 to 400 hours.

Dilo's choice: Dilo tells you the hours. At one hour a day that is about a year. There is no streak to hide the number.

Not proved yet

The accuracy of Dilo's corrections. No app has measured self-study hours to a B1 conversation. The only controlled result in this category is Duolingo's own trial of its Video Call feature, which is first party and tested English learners, not Spanish beginners. Before launch we blind-rate a sample of Dilo's corrections against a human tutor and publish the agreement rate, whatever it is.

Sources

Get a seat

Waitlist for a closed beta of about 100 seats. A short welcome series, then one email every two weeks on what actually makes adults conversational. We tell you the launch date when we have one.

We keep your name, email and phone number to send you the Dilo notes and a beta invitation, and nothing else. We message your phone only if you tick the box. Unsubscribe from any email. Privacy notice.

You are on the list.

We keep your name, email and phone number for the beta invitation and the notes, nothing else. The invitation comes by email when a seat opens.