200 Finnish subtitle words. Mä is not one of them

Here is the clean headline number: 98 of the 200 most frequent alphabetic word forms in Finnish OpenSubtitles 2024 appear as exact strings in the dated Duolingo snapshot we checked. 102 do not.

That sounds like Duolingo missed half the language. It did not. Oli, olen and ollut are three forms of olla, and an exact-string audit counts them separately. The more revealing miss is not number 102. It is a two-letter word that never reaches the top 200 at all: mä.

The top 200, with the audit attached

The ranking below comes from the OPUS OpenSubtitles 2024 Finnish monolingual frequency file. It covers 570,353,193 tokens from 139,983 subtitle documents.¹ We lowercased the supplied word forms, combined capitals with lowercase, removed punctuation and kept alphabetic tokens. These are surface forms, not dictionary headwords.

“In snapshot” means the same lowercase character string appears somewhere in the 780-entry third-party course snapshot. It does not mean a particular learner encountered it, and “not exact” does not mean Duolingo never teaches the underlying idea.

Ranks 101-200

The 102 “misses” are three different things wearing one badge

The exact comparison is real. Its most tempting interpretation is not.

1. Inflections Duolingo may teach as a family

The snapshot includes minä, but the corpus also ranks minun, minua, minulle, minut and minusta. Marking all five “not exact” proves only that an entry list is a poor proxy for everything a course lets a learner construct.

The same happens to verbs. Tulla may be learned as an infinitive while a subtitle delivers tule, tulee or tuli. Finnish makes this problem larger than English because case, person, tense and participles live on the word.

2. Ordinary words outside a short beginner course

Forms such as takaisin (back), heti (immediately) and kunnossa (all right) are not grammatical mirages. They are useful dialogue vocabulary absent as exact entries from this particular snapshot.

The careful phrasing matters. The My Little Word Land list says it was compiled from skill vocabulary and was complete in January 2022.² It is not an official, current export from Duolingo. Course paths change, and the absence of a string from this file cannot support the word “never.”

3. Spoken forms the ranking misses before Duolingo gets a chance

The subtitle file puts minä at 17 and sinä at 22. The lowercased forms mä and sä do not make the top 200. Neither do oon, mun, sun or niinku.

That is not how a room sounds. Research on Finnish teachers recorded in the early 1990s found that the same teachers used standard minä and sinä almost invariably with foreign learners, but almost never with Finnish interlocutors.³ Subtitles are written for the eye. Even when the scene is casual, the line may be normalised on its way to the screen.

Eight families are more useful than 200 isolated cards

The pairs are not translations. They are recognition bridges. A learner who knows minä but has never met mä can know the pronoun and still lose the sentence.

A frequency list is a map of its source

OPUS describes OpenSubtitles as subtitles collected from the OpenSubtitles website, aligned and cleaned for language research.⁴ That scale is its strength. Film and television dialogue supplies questions, commands, pronouns and small connective words in enormous quantities.

Its weaknesses are just as concrete. Much of the Finnish material consists of translated subtitles rather than transcripts of original Finnish speech. Subtitle conventions compress speech. Punctuation and casing are noisy. Different releases of the same title can repeat material. Above all, a subtitle is not a phonetic transcript.

For comparison, the Donate Speech project collected about 3,600 hours from more than 20,000 speakers across Finland and asked professional transcribers to preserve repetitions, hesitations, partial words and non-verbal sounds.⁵ That is much closer to the material a “spoken Finnish” headline promises. Access and licensing make it unsuitable for a casually downloadable public ranking here, and its donor prompts also make it different from two people interrupting each other over coffee.

The older Samples of Spoken Finnish corpus contains 696,376 transcribed words from 50 municipalities, mainly dialect interviews recorded for books published from 1978 to 2000.⁶ It is genuine speech with extraordinary regional reach. It is also a different generation and interaction type. There is no corpus that simply equals “what Finns say.”

The method, including the boring bits that change the winner

We downloaded the Finnish frequency file linked by the OPUS API for OpenSubtitles v2024. The source reports 570,353,193 Finnish tokens. Processing was:

1. Read every supplied frequency and word-form pair.
2. Convert forms to lowercase and add counts for forms differing only by capitalisation.
3. Keep tokens made of letters, allowing an internal apostrophe or hyphen only when surrounded by letters.
4. Sort by the recomputed count and take the first 200.
5. Tokenise the Finnish field of the 780-entry January 2022 course snapshot in the same lowercase form.
6. Mark an item “in snapshot” only on exact string equality.

We did not lemmatise. We did not correct subtitles. We did not silently replace standard forms with colloquial ones. Counts are left out of the display because corpus size makes them look more universal than they are; the order is the useful part.

The reproducibility anchor is the corpus release, not Finnish IRL. OPUS publishes the monolingual file and describes the wider release; the OpenSubtitles paper documents collection and alignment.¹⁴

What to learn first if your goal is listening

Take ranks 1-50 and circle the families, not the rows. Put on, olen, ole, oli, olet, olla and ovat under olla. Put minä, minua and minun together. Then add the spoken partners.

Listen for the small items before the glamorous nouns. Ei, et and en tell you who is negating. Mitä, miksi, miten and missä tell you what answer the next turn requires. No, niin, ihan, kai and vaan tell you how the speaker is steering the exchange.

Then learn the six absent forms that expose the whole limitation: mä, sä, oon, mun, sun, niinku. None appears in this top 200. All six belong in your ears.