How to Calculate MLU Without Losing Your Mind
I still remember the first time a parent asked me, mid-IEP, "wait, so how did you actually get that number?" I was talking about MLU. I knew the answer. I also knew that the spreadsheet I'd used the night before had a hand-typed total in row 53 that I was about 80% sure of. That's a bad place to be.
This is the guide I wish someone had handed me in grad school — Brown's rules in plain English, the edge cases that trip up even experienced clinicians, and a way to keep the math defensible when a parent (or an attorney) asks how you got there.
In a hurry? Skip the math and use the free MLU calculator. Paste a language sample, get morpheme counts and Brown's Stage in seconds. No signup.
First, what MLU actually is
Mean Length of Utterance is the average number of morphemes per utterance in a language sample. It's a developmental index, not a diagnostic threshold — and it only means something when the sample behind it is long enough, spontaneous enough, and recent enough to actually represent how the student talks on a normal day.
Brown (1973) recommended 50 to 100 complete, intelligible utterances (A First Language: The Early Stages). I aim for 50 as my working floor. If I couldn't get there — behavior, attention, the kid just wasn't having it — I say so in the report. That sentence has saved me more than once.
The counting rules, in plain English
Every morpheme in a complete, intelligible utterance is counted once. The whole game is knowing what counts as one versus two. The rules below follow Brown's original conventions and the widely adopted SALT counting guidelines (SALT transcription conventions).
- Free morphemes — every standalone word: dog, run, blue.
- Inflectional morphemes — plural -s, possessive -'s, past tense -ed, third person singular -s, present progressive -ing. Each one adds a morpheme.
- Contractions — count the auxiliary or copula separately. He's running = he + is + run + -ing = 4 morphemes.
- Catenatives — gonna, wanna, gotta, hafta count as one. (I know. I argue with myself about this every time.)
What does NOT count
- • Disfluencies and false starts. I-I-I want = 2 morphemes, not 4.
- • Fillers — um, uh, like (when it's a filler, not "I like it").
- • Direct imitations of whatever you just said.
- • Routinized phrases — thank you, oh my god — count as one.
- • Compound words and proper names — birthday, Mickey Mouse — one each.
- • Irregular past tense and irregular plurals — went, feet — one each. They don't get a bonus morpheme for being irregular.
The formula (it's not scary)
MLU = total morphemes ÷ total utterances
52 complete utterances totaling 218 morphemes? MLU = 218 / 52 = 4.19. Report it to two decimal places. That's the whole math.
A worked example
| Utterance | Morphemes |
|---|---|
| The dogs are barking. | 5 |
| He's gonna fall. | 4 |
| I went to grandma's house. | 6 |
| She wanted more juice. | 5 |
| Look! | 1 |
Five utterances, 21 morphemes, MLU = 4.20. Notice: went is one morpheme (irregular past), grandma's is two (free + possessive), and gonna is one (catenative).
Why I will never let an AI calculate MLU for me
I've tested it. Repeatedly. Paste a transcript into a chatbot, ask for MLU, and you'll get a confident number back. It will also be wrong in non-obvious ways — overcounting contractions, missing irregulars, occasionally inventing an utterance that wasn't there. The output reads beautifully, which is exactly what makes it dangerous.
MLU flows directly into eligibility decisions and goal-setting. The number has to be deterministic and reviewable. I segment the utterances. The tool counts morphemes against published rules. I review and adjust any count I disagree with. The average is computed mechanically. The AI's only job is helping me describe what the number means — never how it was calculated.
Sources
- • Brown, R. (1973). A First Language: The Early Stages. Harvard University Press. Publisher page
- • Miller, J. F., & Iglesias, A. SALT (Systematic Analysis of Language Transcripts) — Transcription & morpheme conventions (PDF)
- • ASHA Practice Portal — Language In Brief: morphology & MLU
- • Owens, R. E. (2020). Language Development: An Introduction (10th ed.). Pearson — Brown's morpheme stages reference.
Skip the spreadsheet
SLP Draft has a built-in MLU calculator that applies Brown's rules to each utterance, lets you adjust the morpheme count on any line, and drops the final figure straight into your language sample section — with every count visible for review.
Deterministic counts. Every morpheme reviewable.