How to take a language sample and compute MLU

Mean length of utterance, or MLU, is the average number of morphemes a child puts into one utterance. It is one of the simplest numbers you can pull from a language sample, and one of the easiest to get wrong. Two SLPs can score the same transcript and get different answers if they handle contractions, fillers and repetitions differently.

This page walks through the sample, the counting rules and the math, then scores a short fictional transcript line by line. Goal writing for this area lives on the expressive language IEP goals page, and the speech therapy IEP goals pillar covers the rest of the field.

What you need before you start

• A recording device. A phone or tablet is fine. You will not catch every morpheme live.
• 20 to 30 minutes with the student in play or conversation, with toys, books or pictures they like.
• A transcription plan. A spreadsheet with one row per utterance works. Language analysis software works too.
• Your counting rules on paper. Decide them before you count, not halfway through.

Language samples show how a student uses language in real talk, which a standardized test cannot fully capture. A complete speech-language assessment combines both.^[1]

Step-by-step: from recording to MLU

Step 1: Collect a natural sample

Play or talk the way you would with any child. Ask fewer questions than feels natural. Questions such as "What's that?" pull one-word answers and drag MLU down. Comment instead ("Uh oh, the tower fell") and wait.

Aim for at least 50 complete, intelligible utterances. Many clinicians collect more, up to 100, so they can drop the first few minutes while the student warms up. Note the setting, the partner and the materials, because each one changes the sample.

Step 2: Transcribe and split into utterances

Write down exactly what the student said, errors included. Then split the talk into utterances. An utterance ends at a clear pause, a drop in pitch at the end of a thought, or a breath before a new idea. "Doggie running. Him fast" is 2 utterances, not 1.

For older students who string clauses together with "and then," many SLPs split by communication unit (one main clause plus anything attached to it) instead. Pick one method and say which you used.

Step 3: Mark what to leave out

Some utterances and words do not count. Mark them before you start counting.

• Leave out whole utterances that are unintelligible in any part, direct copies of what the adult just said, or rote strings such as counting, the alphabet or song lyrics.
• Leave out words that are fillers (um, uh, er), false starts and revisions (in "the, the cat," count "the" once), and stuttered repetitions (count the word once).
• Keep single-word utterances such as "no," "yeah" and "hi." They are real turns.

Step 4: Count morphemes in each utterance

A morpheme is the smallest unit that carries meaning. "Cat" is 1 morpheme; "cats" is 2.

Negative contractions are the one place where conventions split. One approach counts "don't" and "can't" as 1 morpheme for a young child who never says "do" or "can" on their own. Another always counts the "n't" separately. Either is defensible. Write your choice on the scoring sheet.

Step 5: Divide and round

Add up the morphemes in every counted utterance. Divide by the number of counted utterances. Round to 2 decimal places.

MLU = total morphemes / total utterances

You can compute MLU in words the same way, counting each word as 1. Some teams report both. The words version is simpler for bilingual students, where morpheme rules differ across languages.

Step 6: Report it with the context

In present levels, report the MLU with the date, the setting, the partner, the number of utterances and the counting rules. A bare "MLU 2.8" tells the next reader very little. Pair it with what the sample showed: which grammar endings appear, which are missing and how many different words the student used.

Worked example: scoring a short transcript

Liam is a fictional preschool student. His SLP records 25 minutes of car play and transcribes 58 utterances. Here are 12 of them, scored with the rules above.

Across the full transcript, the SLP drops 8 utterances (3 unintelligible, 3 imitations, 2 rote) and keeps 50. Those 50 contain 141 morphemes and 128 words.

• MLU in morphemes: 141 / 50 = 2.82
• MLU in words: 128 / 50 = 2.56

The present levels line reads: "In a 25-minute car play sample with the SLP (50 counted utterances, contractions counted as 2 morphemes), Liam's MLU was 2.82 morphemes. He used -ing in 6 of 9 required contexts and plural -s in 4 of 7. He did not yet use is or are as a helping verb." That sentence tells the team where to aim. A goal could target is plus -ing in play, measured in the next sample. See the present levels speech therapy examples page for more wording.

Comparing MLU to norms

Compare an MLU only to reference data that used the same counting rules and a similar sampling context. A conversation sample and a story retell give different numbers for the same child. If you use software with a comparison database, collect your sample the way that database did.

MLU is also only one piece. Late or limited word combinations can be one sign of a language disorder, but diagnosis rests on the full picture: history, testing, observation and samples.^[2] Do not qualify or dismiss a student on MLU alone. IDEA requires a variety of assessment tools and no single measure as the sole basis for eligibility.^[3]

Common mistakes

Weak: Counting from a sample built on questions such as "What color?" and "What's this?" Better: Use comments and pauses. Questions invite one-word answers and give a falsely low MLU.

Weak: Counting "the, the, the cat" as 4 morphemes. Better: Drop repetitions and false starts. Count "the cat" as 2.

Weak: Counting "went" as 2 morphemes because it is past tense. Better: Irregular past forms are 1 morpheme. Only regular -ed adds a second one.

Weak: Changing the contraction rule between the fall and spring samples. Better: Write the rules on the scoring sheet and use the same sheet every time.

Weak: Reporting "MLU 2.8" with no setting or count. Better: Report the setting, partner, number of utterances and rules with the number.

Weak: Using 20 utterances because the student was quiet. Better: Collect another session. Short samples swing widely with one long or short utterance.

Using MLU in goals and progress monitoring

MLU can be a progress measure for early language goals, but it changes slowly and needs a full sample each time. Many SLPs take a sample at baseline, at midyear and at the annual review, and track a specific grammar form between samples with quicker probes. The speech therapy data collection guide covers the session-by-session methods.

For goal wording, the expressive language goals page has templates for utterance length and elaboration, and the syntax and morphology goals page has goals for specific endings counted in a sample. Students working on first word combinations fit the two-word phrases and early sentence goals page. Check any goal you draft in the goal checker.

Frequently asked questions

Use it in IEP Atlas

• MLU in the IEP glossary
• Language sample in the IEP glossary
• Expressive language goals (speech language)
• Syntax and morphology goals
• Expressive language in the IEP glossary
• IEP goal checker

Related guides

• Speech therapy IEP goals: examples by area
• Expressive language IEP goals
• Speech therapy data collection in session
• Two-word phrases and early sentence goals
• Supporting gestalt language processors
• Speech therapy goals for echolalia and gestalt language processors

Sources

1. Spoken Language Disorders, ASHA
2. Late Language Emergence, ASHA
3. 34 CFR §300.304: Evaluation procedures, U.S. Department of Education (IDEA website)

Written and checked by the IEP Atlas editorial team under our editorial policy. This guide describes general practice and federal rules. It is not legal advice; your state may add requirements, and the IEP team makes the decisions for each student.