MLU and Brown's stages: counting rules, norms and limitations

Mean length of utterance (MLU) is the average number of morphemes per utterance in a sample of a child’s spontaneous speech. Roger Brown introduced it in his 1973 book A First Language as a way to stage early grammatical development, and Miller and Chapman (1981) later showed it rises steadily with age in the preschool years. It remains one of the most used numbers in pediatric language assessment. This guide sets out the counting rules, the stages, the norms and the limits.

Why MLU, and why morphemes

Brown’s insight was that each new piece of grammar a child learns tends to make utterances longer. “Dog” becomes “dogs”, “dog run” becomes “the dog is running”, and so on. Counting morphemes rather than words captures those grammatical additions, because “running” counts as two morphemes while “ran” counts as one.

Brown argued that MLU, not age, is the better index of development in the early years, because his three subjects, Adam, Eve and Sarah, reached the same MLU at very different ages. Miller and Chapman (1981) then studied 123 children aged 17 to 59 months in free play with their mothers and found a correlation of .88 between age and MLU, with MLU rising about 1.2 morphemes per year. That is the basis of the age norms below.

Brown’s rules for counting morphemes

Brown published his counting rules on page 54 of A First Language. The list below follows them, with the examples he gave.

1. Where to start. Begin on the second page of the transcript, unless that page involves a recitation (a song, a counting routine). Count the first 100 utterances that meet the rules below.
2. Fully transcribed utterances only. Exclude any utterance with a blank or unintelligible portion. Portions marked as doubtful transcription are kept.
3. Repetitions. Include exact utterance repetitions. For stuttered words, count the word once in its most complete form. Where a word is repeated for emphasis (“no, no, no”), count each occurrence.
4. Fillers. Do not count “mm” or “oh”. Do count “no”, “yeah” and “hi”.
5. Compounds, names and reduplications count as one. Birthday, choo-choo, quack-quack, night-night, pocketbook and see-saw are each one morpheme, because there is no evidence the child treats the parts separately.
6. Irregular pasts count as one. Got, did, went and saw are one morpheme each, since the child is not relating them to a present-tense form.
7. Diminutives count as one. Doggie and mommie are one morpheme; the child is not using the suffix productively.
8. Auxiliaries and inflections count separately. Is, have, will, can, must and would each count as a morpheme. So do the inflections: possessive ‘s, plural s, third-person singular s, regular past ed and progressive ing. The catenatives gonna, wanna and hafta count as a single morpheme each, because they function as units for young children.
9. Upper bound. The longest utterance (the “upper bound”) is calculated from the whole transcript, not just the 100 utterances.

Common extensions clinicians add

Brown’s list does not cover everything, so most protocols add conventions. State which ones you used, because they change the number.

• Contractions. The contracted copula or auxiliary is a separate morpheme: “he’s going” is four morphemes and “I’m” is two. Negative contractions such as “don’t” and “can’t” are usually two, though some protocols count “don’t”, “won’t” and “let’s” as single units for children who never produce the parts separately.
• Irregular plurals. Mice, feet and children count as one morpheme, following the same logic as irregular pasts. Plurals that never occur in the singular (pants, clothes) are also one.
• Overgeneralized forms. “Goed” and “mouses” count the inflection, because the child is applying the rule. Most protocols count them as two morphemes.
• False starts and revisions. Exclude the abandoned portion and count only the final version.
• Imitations and routines. Many clinicians exclude direct imitations of the adult, rote counting and song lyrics, since they do not reflect the child’s own grammar.

Divide the total morphemes by the number of utterances to get MLU. The free MLU calculator applies Brown’s rules to a typed transcript and reports both MLU in morphemes and MLU in words.

Brown’s five stages

Brown defined five stages by MLU range rather than age. The approximate ages are the ones commonly tabulated from his data and from Miller and Chapman’s regression, and they should be read as rough guides only.

Brown also tracked 14 grammatical morphemes and found that his three children acquired them in nearly the same order, from present progressive ing first to the contractible auxiliary last. Mastery meant 90% correct use in obligatory contexts across consecutive samples. The stages and the morpheme order are separate findings, and a child can be in Stage III by MLU while still missing a Stage II morpheme.

Miller and Chapman’s predicted MLU by age

Miller and Chapman (1981) fitted a straight line to their data. Predicted MLU equals -0.548 plus 0.103 times age in months, and the predicted standard deviation equals -0.0446 plus 0.0205 times age. The table gives the values at three-month intervals.

Two things stand out. The standard deviation grows with age, so the “normal” band at five is almost four times wider than at 18 months. And the sample was 123 middle- to upper-middle-class children in one midwestern city. The authors themselves recommended local norms where possible.

Collecting a 50 to 100 utterance sample

A language sample is only as good as the conditions it was collected in. The steps below reflect common practice in ASHA’s spoken language disorders guidance and the MLU literature.

1. Choose the context. Free play with a familiar adult and open-ended toys (a farm set, a doll house, blocks) produces the most representative talk for children under four. Picture description and story retell can be added for older children, but note that MLU changes with task.
2. Record audio or video. Transcribing live is unreliable. A phone recording is fine in a quiet room.
3. Aim for 15 to 20 minutes of interaction. Follow the child’s lead, comment more than you question, and avoid yes-no questions, which produce one-word answers and drag MLU down.
4. Transcribe verbatim. Mark unintelligible words, false starts and imitations so they can be excluded later.
5. Segment into utterances. Use pauses, intonation and grammatical completeness. Where a child chains clauses with “and”, most protocols split them into separate utterances unless the second clause is subordinate.
6. Select the utterances. Brown used 100. Many clinical protocols accept 50 fully intelligible, spontaneous utterances, which is usually reachable in a 15-minute sample from a child over two and a half. Younger or quieter children may need two.
7. Count and calculate. Apply the rules above, or paste the transcript into the MLU calculator. If the same sample is phonetically transcribed, the PCC calculator can give a speech sound measure from the same recording.

Record the context, the sample size and the counting conventions in the report. An MLU from a 50-utterance picture-description task is not comparable to one from 100 free-play utterances.

Limitations

MLU is quick and sensitive to early grammar, but it has well-documented limits.

• Ceiling. Growth slows after about 36 months. Scarborough, Wyckoff and Davidson (1986) found that the linear relationship reported by Miller and Chapman was only partly replicated in two new samples, with a clear deceleration beyond three years. Above an MLU of about 4 to 5, length reflects style and task more than grammar, and clinicians move to measures such as clausal density, number of different words and specific morpheme accuracy.
• Not a diagnostic test on its own. Eisenberg, Fersko and Lundgren (2001) reviewed the evidence and concluded that MLU alone does not reliably separate children with language impairment from typical peers, partly because sampling conditions change the number so much. Rice, Redmond and Hoffman (2006) found MLU valid and stable as a growth measure in children with specific language impairment, which supports its use for monitoring progress rather than for identification.
• Dialect. Some English dialects, including African American English, allow optional omission of forms such as the copula and third-person s. A child speaking those dialects will score lower on MLU in morphemes without any language difficulty. ASHA’s guidance on spoken language disorders stresses separating difference from disorder.
• Bilingual children. The norms are monolingual. In addition, the morphology of the child’s other language matters: a child whose home language marks less with inflection may transfer that pattern. MLU in words, dynamic assessment and parent report give a fuller picture.
• Sample size and context. Short samples, yes-no questions and unfamiliar adults all lower MLU. A child who is shy for 15 minutes can look a year behind.
• The norming sample. Miller and Chapman’s predicted ranges come from one city and one socioeconomic band. They are a reference, not a cut-off.

What this means in practice

For a two- to four-year-old, an MLU from a good free-play sample is a useful and well-researched index of grammatical development, best read against Brown’s stages and Miller and Chapman’s ranges together. It is one piece of an assessment, alongside standardized testing, parent report and observation. For parents, the related guides on how many words a toddler should say and what to do about a late talker cover the vocabulary side of the same question, and the glossary defines the terms used here.