Standardized tests have a place in the SLP's toolkit, but they have fundamental limitations: they sample behavior in artificial, highly structured conditions, they may not reflect how a child communicates in everyday contexts, and they can underidentify language disorders in children who are good test-takers but struggle in connected discourse. Language sample analysis (LSA) addresses all of these gaps. A well-collected, well-analyzed language sample gives you a window into how a child actually uses language — in connected speech, across contexts, with real communicative intent.
This guide walks through the full LSA process, from elicitation to report writing, with practical tips for managing it efficiently in a school-based setting.
Why Language Samples? Ecological Validity
The concept of ecological validity in assessment refers to whether a measure reflects real-world performance. Standardized tests notoriously lack ecological validity — a child who scores within normal limits on a picture-naming task may still produce telegraphic sentences with reduced syntactic complexity in conversation. Language samples capture authentic communicative behavior. They reveal how a child organizes discourse, codes complex meaning, and manages the real-time demands of communication — none of which standardized tests measure well.
ASHA guidelines and the research literature consistently recommend language sampling as a core component of language evaluation, particularly for children from culturally and linguistically diverse backgrounds where standardized test norms may not apply.
How to Elicit a Language Sample
The elicitation context affects the type of language you'll get. Use at least two contexts to capture different registers:
- Conversation: Free conversation about the child's interests, weekend, family, pets. Produces casual register, clause-level structures, and discourse management skills. Sit at the child's level, follow their lead, and avoid yes/no questions. Ask "Tell me about..." and then stay quiet.
- Picture description: Present a complex picture scene (e.g., a busy playground, a kitchen scene) and ask "Tell me everything you see happening." Produces noun phrase elaboration, prepositional phrases, and present progressive structures. Good for short sampling sessions.
- Narrative: Ask the child to retell a story they know well, describe a recent event ("Tell me about the last birthday party you went to"), or generate a story from wordless pictures. Produces complex syntax, temporal and causal connectives, and story grammar structures. Narrative samples are particularly rich for school-age children.
Aim for a minimum of 50 utterances for school-age children (some protocols require 100). Record the sample — trying to transcribe in real-time during therapy is not sufficient for valid analysis. Most SLPs use a voice recorder or phone; make sure you have a quiet room.
Transcription: C-Units vs. T-Units
Before you can analyze a language sample, you need to segment it into units. Two conventions exist:
- C-units (Communication Units): Each main clause plus its attached subordinate clauses, OR a non-clausal response that stands alone as a communicative act ("Yeah," "The red one"). Used for conversational and narrative samples. Preferred for children under age 8 and for conversational samples at any age.
- T-units (Terminable Units): Each main clause plus its attached subordinate clauses. Identical to C-units except that non-clausal responses are excluded. Used more commonly in written language analysis and older children's narrative samples.
For most school-based work with K-8 students, C-units are the practical choice. When transcribing, use standard conventions: mark mazes (revisions, repetitions, fillers) with parentheses or brackets but exclude them from analysis, and mark partial utterances when possible.
What to Measure
Mean Length of Utterance (MLU)
MLU in morphemes is the foundational LSA measure for preschool and early school-age children. Count every free morpheme (each word) and every bound morpheme (-ed, -ing, -s, possessive 's, etc.) across all utterances, then divide by the number of utterances. Brown's (1973) MLU stages provide developmental benchmarks. MLU is most sensitive between ages 2;0–4;6; it loses diagnostic sensitivity by mid-elementary because children are using complex sentences rather than longer simple ones.
Number of Different Words (NDW) and Total Number of Words (TNW)
NDW counts the number of unique word types in the sample (vocabulary diversity). TNW counts every word token, including repetitions. NDW is particularly sensitive to vocabulary development and distinguishes children with language disorders from typical peers even when MLU looks normal. Normative data by age and sample size are available from Leadholm and Miller (1992) and others.
Developmental Sentence Score (DSS)
The DSS (Lee, 1974) is a systematic scoring of 8 grammatical structures: indefinite pronouns, personal pronouns, main verbs, secondary verbs, negatives, conjunctions, interrogative reversals, and wh-questions. Each structure is scored by developmental level (1-8 points), and a sentence point is added for completely correct sentences. DSS provides a detailed profile of which grammatical structures are at, below, or above age level. It is time-consuming to score but provides unmatched specificity about morphosyntactic targets.
Subordination Index and Clausal Density
For school-age children, MLU may plateau while syntactic complexity continues to grow via embedded clauses. The subordination index (total clauses divided by total utterances) captures this. A child producing mostly simple sentences will have a subordination index near 1.0; a child using complex, embedded sentences may approach 1.5-2.0. Clausal density and the Index of Productive Syntax (IPSyn) are also commonly reported.
Interpreting Norms and Writing Up Results
Compare your measures to published normative data. Key references include Leadholm and Miller (1992) for MLU, NDW, and TNW by age; SALT norms (Miller & Iglesias) stratified by age and elicitation context; and DSS norms by age (Lee, 1974). Report measures with confidence intervals where possible — language samples have inherent variability, and a single number without context can mislead.
In your evaluation report, integrate LSA findings narratively rather than presenting raw scores in isolation. For example: "Analysis of a 100-utterance narrative sample revealed an NDW of 87, which falls below the 25th percentile for [Student]'s age, consistent with the vocabulary limitations observed on standardized testing. MLU of 7.2 morphemes per utterance is within normal limits; however, clausal density of 1.08 suggests that [Student] is producing primarily simple sentences with limited subordination, which may affect expository writing demands in 4th grade."
Software Options
Manual calculation of LSA measures is feasible for a quick clinical summary but becomes impractical for full evaluations. Consider:
- SALT (Systematic Analysis of Language Transcripts): The gold standard. Web-based ($149/year) or desktop. Built-in normative databases, automatic calculation of most measures, transcript coding conventions. Most SLPs who do formal LSA use SALT.
- CLAN (Computerized Language ANalysis): Free software from the CHILDES project. Steeper learning curve than SALT but powerful for research-level analysis.
- Clinical Discourse Analysis: For qualitative measures of conversational discourse (topic maintenance, conversational repair, presupposition), manual clinical discourse analysis may still be necessary — software tools don't fully capture pragmatic adequacy.
Efficiency Tips for Busy School SLPs
Language sampling has a reputation for being time-consuming — and it can be. But several practices reduce burden without sacrificing validity:
- Record and transcribe later: Collect the sample during a session, then transcribe at your desk. Don't try to transcribe in real-time during the interaction.
- Elicit during existing sessions: Rather than scheduling a separate "language sample session," integrate elicitation into ongoing therapy. A narrative retelling during a regular therapy session can double as a data collection task.
- Targeted samples for monitoring: For progress monitoring (not full evaluation), a 25-utterance narrative sample scored for NDW and MLU is sufficient to track change over time.
- Use SALT's Reference Database: Rather than manually consulting multiple normative references, SALT's integrated database provides immediate comparison to age-matched peers, saving significant report-writing time.
Language sample analysis is one of the most clinically valuable skills in an SLP's repertoire. The investment in learning it — and systematizing its use — pays dividends in more accurate identification, more targeted intervention, and more persuasive evaluation reports.