Language Sample Analysis: MLU, TTR & More

A standardized test tells you how a child performs on someone else's items. A language sample tells you what the child actually does with language. It is the least biased, most functional data source available to an SLP, it works across dialects and languages, and it generates goals directly. It is also the piece most often skipped when time is short — which is why exam items keep asking about it.

Collecting a usable sample

Aim for 50 to 100 complete, intelligible utterances, or roughly 15 to 20 minutes of recorded interaction. Fewer than 50 utterances makes MLU unstable. Vary the context: conversation, play, narrative retell, and expository or persuasive tasks for older students each pull different structures. Narratives in particular expose complex syntax and cohesion problems that conversation hides, because conversation lets a child ride on the partner's scaffolding.

Talk less than the child. Use comments rather than questions, avoid yes/no questions, allow silence, and follow the child's lead. A clinician who asks 40 questions in 15 minutes has collected a test, not a sample. Record audio (or video for gesture and AAC use) and transcribe rather than scoring live.

Segmenting utterances

For preschool conversation, segment by C-unit (an independent clause plus its modifiers) or by communication unit boundaries marked by intonation and pause. For school-age narrative and expository language, T-units (main clause plus attached subordinate clauses) are standard. Consistency matters more than the choice: mixed segmentation makes every number meaningless.

MLU: mean length of utterance

MLU in morphemes is total morphemes divided by total utterances, calculated on 50 to 100 utterances. The counting rules matter and are frequently tested:

• Count each free morpheme as one, plus bound grammatical morphemes: plural -s, possessive 's, regular past -ed, third person singular -s, present progressive -ing.
• Do not give extra credit for irregular past ("went" = 1) or irregular plurals ("feet" = 1) — they are assumed to be unanalyzed wholes at this stage.
• Compound words, proper names, and ritualized reduplications ("choo-choo," "night-night") count as one morpheme.
• Exclude imitations, self-repetitions, false starts, fillers ("um," "uh"), and unintelligible utterances.
• Contractions count as two morphemes ("don't" = do + not).

Brown's stages map MLU to expected structures: Stage I (MLU 1.0–2.0, semantic relations), Stage II (2.0–2.5, the first grammatical morphemes appear), Stage III (2.5–3.0, sentence modalities — questions, negatives, imperatives), Stage IV (3.0–3.75, embedding), Stage V (3.75–4.5, conjoining). Brown's 14 morphemes emerge in a broadly consistent order, beginning with present progressive -ing, then the prepositions in and on, then regular plural -s, and ending with the contractible copula and auxiliary.

MLU is a useful index through roughly age 5 or an MLU of about 4.5. After that it flattens and stops discriminating; older students need subordination index, clausal density, and narrative measures instead.

TTR and lexical diversity

Type-token ratio is the number of different words (types) divided by the total number of words (tokens). A TTR near 0.45 to 0.50 is often cited as typical for preschoolers, but the measure has a well-known flaw: TTR falls as sample length grows, because common words repeat. Comparing a 200-word sample to a 500-word sample using raw TTR is invalid.

Length-robust alternatives include the number of different words in a fixed 50-utterance window (NDW), moving-average TTR, VOCD, and the measure of textual lexical diversity (MTLD). If you only report one lexical measure, NDW on a fixed sample size is more defensible than raw TTR.

Other measures worth calculating

• Percentage of grammatical utterances — sensitive to developmental language disorder in school-age children.
• Subordination index — total clauses divided by total C-units; captures syntactic complexity where MLU has plateaued.
• Finite verb morphology composite — accuracy on tense and agreement markers (past -ed, third person -s, copula and auxiliary BE and DO), a well-documented clinical marker of DLD.
• Narrative measures — story grammar elements, episode structure, cohesion, and referencing.
• Communicative functions and pragmatics — requesting, commenting, protesting, topic maintenance, turn-taking, and repair.
• Percent intelligible utterances — a functional intelligibility index that single-word articulation tests cannot give you.

Why language sampling is the answer to bias

Norm-referenced batteries are standardized on samples that may not represent the child in front of you. A speaker of African American English who omits final consonant clusters or the copula is following a rule-governed dialect, not making errors — and structured tests will penalize that. A language sample scored against the child's own dialect, combined with dynamic assessment (test-teach-retest, measuring modifiability), is the recommended practice for culturally and linguistically diverse students. This pairing is one of the most consistently keyed answers on the entire assessment domain.

From sample to goals

Analysis should end in targets. If the sample shows MLU 2.6 with omitted copula and auxiliary BE, target those morphemes in recast-heavy activities. If the sample shows adequate MLU but no subordinate clauses and no story grammar, target complex syntax within narrative retell. If the child produces plenty of language but never repairs a breakdown, target conversational repair. Software such as SALT can automate the counting, but the clinical reasoning is yours.

Exam angle

Expect items asking you to compute or interpret MLU, to identify which measure is appropriate for an older student, to explain why TTR varies with sample length, and to choose language sampling plus dynamic assessment over another standardized test for a bilingual or dialectally diverse child.