MLU: Calculation, Norms, and Clinical Use — A Practical Guide for SLPs

Mean Length of Utterance is the average number of meaningful units in a child’s spontaneous utterances. Roger Brown introduced the metric in A First Language (1973) as a way to summarise the grammar of his three longitudinal participants — Adam, Eve, and Sarah — and the field adopted it almost overnight because it does something no test score can: it produces a single, defensible number from the child’s real spoken language. Six decades later it is still the first metric on every clinical Language Sample Analysis report.

MLU works because grammar gets longer before it gets more complex. A two-year-old saying "doggie sleep" produces an MLU of 2. A five-year-old saying "the doggie is sleeping in his bed" produces an MLU of about 7. The growth curve from 1.0 to roughly 6.0 morphemes is a remarkably consistent index of expressive-language maturation in monolingual General American English children, which is why ASHA, Owens (2014), and the SUGAR research team all keep recommending it as the headline metric for school-based reports.

There are two flavours: MLU in morphemes (MLU-m) and MLU in words (MLU-w). MLU-m counts every meaningful unit, including bound morphemes such as past-tense -ed and plural -s. MLU-w counts whitespace-separated words. They report related but distinct things, and most clinicians end up reporting both.

The single biggest source of error in clinical MLU reporting is hand-counting morphemes by ear. Brown’s rules are precise, and most of them are also counterintuitive. The list below is the consensus reading of Brown (1973, pp. 54-55) as restated by Owens (2014) and the SALT 2020 transcription manual. If your scoring software (or your tally sheet) does not implement them, your MLU will drift in the same direction every time.

MLU-morphemes is more sensitive to early grammatical development because every newly-acquired bound morpheme adds to the count. It is the right metric for any child below MLU 4.0 — roughly Brown Stages I through III, or developmental ages 18 months to 4 years. Below this ceiling, MLU-m correlates with language age at r > 0.85 in the Brown corpus and at r ≈ 0.7 in the larger SUGAR sample.

MLU-words is easier to compute, requires no morpheme-counting rules, and is the metric most school-age normative datasets actually report. Above MLU 4.0, MLU-m and MLU-w correlate at r > 0.95 — they essentially measure the same thing once the child has acquired the high-frequency bound morphemes. For school-age children (kindergarten through grade 5) the field generally reports MLU-w against the SUGAR or Heilmann reference set and only falls back to MLU-m when the analysis question is specifically about morphological development.

For bilingual children, MLU-w is usually the safer choice because morpheme-counting rules differ across languages and there is no consensus cross-linguistic equivalent for "morpheme" as Brown defined it. Spanish-English bilingual evaluations almost always report MLU-w in each language separately and avoid MLU-m altogether.

Brown (1973) divided early language development into five stages defined by MLU bands: Stage I (MLU 1.0-2.0), Stage II (2.0-2.5), Stage III (2.5-3.0), Stage IV (3.0-3.75), and Stage V (3.75-4.5). The stages map approximately to chronological ages 18-26 months, 27-30 months, 31-34 months, 35-40 months, and 41-46 months for typically-developing monolingual children. The stages remain useful as a rapid clinical gloss but they were derived from three children and they overestimate growth past Stage V — most school SLPs need a different reference for anyone older than 4.

The current state of the art for school-age MLU is the SUGAR (Sampling Utterances and Grammatical Analysis Revised) project — Pavelko and Owens (2017, 2019) — which collected 50-utterance conversational samples from 250 typically-developing US children aged 3 through 7 and published age-banded means and standard deviations for MLU-w, NDW, and percent grammatical utterances. The SUGAR norms are the most defensible reference for monolingual General American English speakers in the 3-7 age band today.

For older school-age children, Heilmann, Nockerts, and Miller (2010) extended the reference range up through age 11 using the SALT databases. Above age 11, MLU as a metric loses sensitivity because typical adolescents already produce most utterances above the 7-morpheme ceiling, and clinicians switch to subordination index, T-units, or IPSyn for syntactic-complexity tracking.

MLU is unstable on small samples. The consensus floor is 50 utterances — below that, your single-utterance noise dominates the mean and the score swings by 0.5-1.0 morphemes between sessions. Pavelko and Owens (2017) collect exactly 50 utterances per child for the SUGAR norms, Heilmann et al. (2010) collect 100, and Eisenberg and Guo (2013) recommend 50-100 for clinical decisions. Anything shorter than 50 is a screening tool, not an MLU report.

The sample also has to be elicited under conditions the norms expect. SUGAR norms come from a free-play conversation between an adult and the child for ages 3-5, and a topic-prompted conversation for ages 6-7. If you elicit a story-retell instead, your MLU will be inflated by 0.5-1.0 morphemes because narrative tasks pull longer utterances. Match the elicitation method to the reference set.

Almost every flagged MLU report fails at one of the same five points. Read this list before your next clinic day; it will save you a re-test and probably save a child from a misclassified eligibility decision.

A defensible IEP goal written from an MLU score names the metric, the structures driving the gap, the target criterion, and the conditions under which the data will be collected. The MLU score on its own is a baseline number; the goal turns it into a measurable plan.

The workflow: identify the SUGAR or Brown reference for the child’s age, calculate the gap in morphemes between the child’s mean and the reference mean, examine the missing structures (which Brown’s morphemes are not appearing in obligatory contexts?), and write a goal that targets those structures with a 6-month criterion. Re-run the same 50-utterance MLU at the next IEP review — the metric is the goal’s own progress monitor.

Three options exist for actually scoring MLU. The browser-based MLU Calculator on this site implements Brown’s rules in code, accepts plain-text utterances with no special transcription syntax, and emits MLU-m, MLU-w, total morphemes, total words, and the matching Brown stage in one pass. It is free, runs entirely client-side (the child’s data never leaves the browser), and is the fastest option for a single sample.

SALT (Systematic Analysis of Language Transcripts) is the long-standing professional alternative. It costs roughly $295 per single-user licence, requires its own transcription syntax, and ships with a 700-child reference database that is broader than SUGAR. For a clinician running 50+ language samples per year on a research-grade workflow, SALT is still worth the investment. For a clinician running 10 samples per year as part of school-based evaluations, the free MLU Calculator covers the same metric on the same Brown rules.

ConductSpeech is the AI alternative for clinicians whose bottleneck is transcription, not scoring. It accepts the raw audio recording, runs the transcription, computes the full MLU/NDW/PGU battery, and drafts a present-levels paragraph. It is the only option in this list that automates the slow step — typing the transcript — so it pays for itself on caseloads above 20-30 students.