Bilingual Language Sample Analysis: A 2026 Clinical Reference for School SLPs

A bilingual language sample is not a monolingual language sample run twice in two languages and then averaged. The assessment is structurally different because the underlying linguistic system the clinician is trying to characterise is structurally different: a bilingual child has a single integrated language faculty that distributes itself across two grammars, two phonologies, and two lexicons in ways that depend on age of acquisition, exposure history, language environment, and the specific communicative demands of the moment. A clinician who treats the bilingual sample as two monolingual samples will under-count vocabulary, mis-score grammaticality, and produce eligibility conclusions that the bilingual research literature has shown for thirty years are systematically wrong. The first thing a school SLP has to internalise about bilingual LSA is that the child is not "half an English speaker plus half a Spanish speaker" — the child is one bilingual speaker whose total language competence is the union of the two systems, not the average.

The clinical consequence of this framing is that every methodological decision in the bilingual sample — which language to sample in, what counts as a complete utterance, how to score grammaticality on a code-switched turn, which norms to compare against — has to be made with the bilingual literature in mind, not the monolingual default. The default settings of every commercial LSA tool, every district assessment template, and every legacy normative reference were calibrated against monolingual English-speaking children. A school SLP who uses those defaults on a Spanish-English bilingual child will produce a number that under-states the child’s real language competence and will, with high probability, contribute to over-identification of language impairment in a population that bilingual research has shown is already over-identified. The corrective is not to throw out the calculators or the norms; it is to know which adjustments the bilingual research community has converged on and to apply them consistently.

The honest framing for the rest of this pillar is that bilingual LSA in 2026 is a defensible assessment activity — the methods exist, the norms exist, the calculators on this site work — but the methods and norms are different from the monolingual case in specific ways that have to be applied deliberately. The four sections that follow walk through the four core problems in order: language choice, code-switching scoring, normative reference selection, and eligibility-and-goal writing. Each section names the specific bilingual research finding that drives the recommendation and the specific clinical action a school SLP should take on a typical caseload.

The first methodological decision in any bilingual LSA is which language to sample in, and the answer that the bilingual research literature has converged on for school-age clinical assessment is "both, separately, with two distinct samples elicited under conditions that match the child’s real language environment." The reason is that a child’s linguistic competence in one language is not a reliable proxy for the other; a child can be at the 50th percentile in Spanish narrative complexity and the 5th percentile in English narrative complexity, and the clinically meaningful number is the higher of the two, not the lower one. A monolingual English sample on a Spanish-dominant child will report the lower number, which is the wrong number for an eligibility decision. The right protocol is to elicit one full sample in each language, score each sample with the appropriate language-specific calculators and norms, and then make the eligibility decision against the higher of the two language scores.

The practical elicitation protocol is straightforward but has three details that matter. First, the two samples should be elicited on different days when possible, not back-to-back, because consecutive sampling in two languages produces a pull-to-the-dominant-language effect that reduces the second sample’s validity. Second, each sample should be elicited by an adult who is fluent in the target language for that sample — a Spanish sample elicited by an English-only adult through gestures and pictures will under-represent the child’s Spanish competence, because elicitation depth is bounded by the eliciter’s own language. When a fluent adult is not available, the school SLP should partner with a bilingual interpreter trained in clinical interviewing, or with the child’s family for a home-language sample, rather than producing a degraded English-mediated Spanish sample. Third, each sample should be at least 50 utterances long under the same SUGAR-style protocol used for monolingual samples; shorter samples have the same reliability problems in bilingual LSA that they have in monolingual LSA.

The third detail is the one school SLPs most often get wrong on a tight assessment timeline: producing a single 100-utterance "bilingual" sample where the child code-switches freely is not a substitute for two separate 50-utterance language-specific samples. The mixed sample is useful for a different question — it characterises the child’s code-switching patterns and dominance balance — but it is not a substitute for the two language-specific samples that the eligibility decision actually depends on. The three-sample protocol (English-only sample, Spanish-only sample, and a brief mixed sample for code-switching analysis) is the gold standard for bilingual LSA on a school caseload in 2026, and it is the one this pillar recommends a clinician build into the assessment timeline from the start.

Code-switching — the systematic alternation between two languages within a single utterance or across consecutive utterances by a bilingual speaker — is the methodological hot spot of bilingual LSA. A clinician who treats every code-switched utterance as ungrammatical will produce a PGU score that drastically under-represents the child’s competence. A clinician who treats every code-switched utterance as grammatical without checking the morphosyntactic constraints of either language will produce a PGU score that overstates competence. The bilingual research community has converged on a clear, rule-governed answer to this problem, and the answer is the one this section walks through in detail because it is the single most actionable methodological correction in this entire pillar.

The bilingual literature’s framing is that code-switching is a rule-governed linguistic behaviour, not a deficit. Bilingual speakers code-switch at predictable structural boundaries (between independent clauses, at major syntactic constituents, after function words that license a switch in both languages) and follow well-documented constraints — the equivalence constraint, the free morpheme constraint, the matrix language frame model. A code-switched utterance that respects these constraints is a grammatical utterance in the bilingual’s integrated grammar; a code-switched utterance that violates them is the same kind of error as any other ungrammatical utterance and should be tagged as such. The PGU scoring rule for bilingual LSA is therefore: tag each code-switched utterance as grammatical IF and only if it respects the morphosyntactic constraints of both languages at the switch boundary, otherwise tag it as ungrammatical — the same way a monolingual sample tags any ungrammatical utterance.

The MLU rule is more permissive but follows the same logic. For MLU-w, count every word in a code-switched utterance regardless of language; the child produced the word, the word communicates meaning, the word counts. For MLU-m, count morphemes in each language using the morpheme rules of that language — Spanish morphology is richer than English, so a Spanish utterance with three words may have five or six morphemes where an English equivalent has three. The temptation to "correct" the Spanish morpheme count down to the English equivalent is exactly the kind of monolingual-default error this pillar is trying to prevent, and it produces an MLU-m number that under-represents the child’s grammatical competence by 20 to 30 percent in many bilingual transcripts. The right rule is: count Spanish morphemes by Spanish rules, count English morphemes by English rules, and report MLU-m for each language’s sample separately.

The third scoring decision — NDW and lexical diversity in a bilingual sample — is the place where the "union, not average" framing matters most. A child’s NDW in the English-only sample is the count of distinct English words; a child’s NDW in the Spanish-only sample is the count of distinct Spanish words; the child’s total conceptual vocabulary is the union of the two, deduplicated by concept (not by word form). The conceptual NDW — the count of distinct concepts the child can label in either language — is the metric that most closely matches the monolingual NDW the eligibility decision is calibrated against, because monolingual NDW is implicitly conceptual. Reporting only the English NDW on a Spanish-dominant child is the most common bilingual LSA error in school assessments, and it is the one that drives the largest false-positive eligibility rate in the published literature.

The third problem in bilingual LSA is which normative reference to compare the scored metrics against. The default — monolingual English norms from SUGAR, Brown’s stages, or any other US school-age reference — systematically over-identifies bilingual children with language impairment because the bilingual child’s English numbers reflect English exposure history, not the underlying language faculty. A Spanish-dominant five-year-old who has been in an English-medium school for six months will have an English MLU-w that looks like a typically-developing three-year-old monolingual English speaker, and a clinician who applies monolingual norms to that English MLU-w will produce a "moderate language impairment" eligibility recommendation that the bilingual literature has shown for thirty years is wrong. The corrective is to compare each language’s sample against language-specific norms when they exist and against the conceptual aggregate when they do not.

For Spanish, the most useful published normative reference for school-age LSA in 2026 is the Pavelko and Owens SUGAR work extended to Spanish samples by the bilingual research community, plus the older Restrepo and colleagues work on Spanish-speaking preschoolers. SUGAR Spanish norms exist for MLU in words, NDW, and total number of utterances at the same age bands as the English SUGAR norms, which means a school SLP can run the Spanish sample through the same calculator pipeline (MLU, NDW, total utterances) and compare against age-banded Spanish norms rather than English norms. This is the single most defensible normative-reference change a school SLP can make on a bilingual caseload, and it converts a meaningful fraction of false-positive eligibility recommendations into correct typically-developing classifications. The Spanish norms will not exactly match the English norms because the languages have different morphological richness, different baseline word lengths, and different age trajectories, which is exactly why language-specific norms are necessary in the first place.

For language pairs other than Spanish-English, the published normative reference base is much thinner. There are some published norms for Mandarin, Korean, Arabic, and a handful of other commonly-spoken school-age languages, but the coverage is uneven and the sample sizes are smaller. The 2026 best practice when published norms do not exist for the child’s home language is to (1) elicit the home-language sample anyway, (2) report the metric values descriptively without a normative comparison, (3) use the conceptual NDW as the cross-language proxy for the eligibility decision, and (4) document explicitly in the report that the home-language sample lacked normative comparison and that the eligibility conclusion is based on the higher of the two language samples. This is the conservative protocol the bilingual literature recommends, and it is the one that produces an eligibility conclusion that survives due-process review.

The clinical heart of bilingual LSA — the question every step in the previous three sections is feeding into — is the differentiation between language difference and language disorder. A bilingual child who scores below age-expected levels in English because of limited English exposure is a child with a language difference; a bilingual child who scores below age-expected levels in BOTH languages on age-appropriate elicitation conditions is a child who may have a language disorder. The clinical rule, codified in the bilingual research literature and in ASHA’s practice guidance for bilingual assessment, is that a language disorder must be evident in BOTH of the child’s languages, not just in the second language. A child whose English is delayed and whose Spanish is age-appropriate has a difference, not a disorder, and is not eligible for special education services on the basis of language impairment alone.

The "evident in both languages" rule is the load-bearing finding of forty years of bilingual SLP research, and it is the one that the school SLP’s clinical reasoning has to anchor on when interpreting the metrics from the two language-specific samples. The practical implementation is: if the higher of the two language samples scores within the typical range for that language’s age-banded norms, the child has a language difference and the eligibility recommendation is "not eligible on the basis of language impairment." If the higher of the two language samples scores below the typical range for that language’s age-banded norms, the child may have a language disorder and the next step is a full multidisciplinary evaluation with a bilingual SLP or interpreter on the team. The conceptual NDW can be used as a corroborating cross-language metric in either direction.

A bilingual LSA that follows this protocol produces an eligibility recommendation that is defensible in due-process review for two reasons. First, it correctly applies the "evident in both languages" rule that the courts and ASHA both recognise as the standard. Second, it documents the reasoning chain explicitly: which languages were sampled, which norms were used, what the higher score was, and how the higher score compared to the language-specific normative reference. A district that produces a bilingual eligibility recommendation without this documentation is taking a real legal exposure risk — there is published case law in multiple jurisdictions where school districts have been ordered to redo bilingual evaluations precisely because the original evaluation applied monolingual norms to a bilingual child. The protocol in this section is the protocol that prevents that outcome.

When a bilingual LSA does identify a language disorder — below-age-expected performance in both languages, with both languages sampled under appropriate conditions — the next clinical question is how to write IEP goals that are calibrated against the right baseline and the right normative reference. The bilingual case is structurally different from the monolingual case because the goal has to specify which language the goal is targeting, which language the progress will be measured in, and which language the intervention will be delivered in. Each of these three decisions has clinical and procedural consequences, and a goal that does not specify them is a goal that will not survive a due-process review.

The 2026 best practice for bilingual IEP goals is to write language-specific goals where the underlying impairment is language-specific and to write language-general goals (with explicit language specification in the measurement criterion) where the underlying impairment is general. A goal targeting Spanish past-tense morphology should say "Marcus will produce Spanish past-tense verbs in obligatory contexts with 80% accuracy across three consecutive sessions, as measured by a 50-utterance Spanish language sample scored against SUGAR Spanish norms." A goal targeting general expressive vocabulary should say "Marcus will produce age-appropriate expressive vocabulary in BOTH English and Spanish, measured by conceptual NDW from paired English and Spanish 50-utterance language samples scored against age-banded language-specific norms." The IEP Goal Generator on this site supports both patterns with a language-specification field that defaults to English but accepts other language tags for bilingual goals.

The intervention-language decision is a separate question, and the bilingual research consensus is that intervention should be delivered in the child’s strongest language whenever possible, not in the language the school happens to operate in. A child whose Spanish is stronger than their English benefits more from Spanish-language intervention even when the school is English-medium, because language therapy in the dominant language transfers to the second language more reliably than the reverse. School SLPs do not always have the staffing to deliver intervention in the child’s home language, but they should document the recommendation explicitly in the IEP, and they should use the conceptual NDW as a measurement metric that captures growth in either language. The IEP Goal Generator on this site produces SMART goal sentences that include the language-specification fields as a standard part of the bilingual template.

A bilingual LSA without a careful home-language history is missing the most important contextual variable in the assessment. The child’s exposure history — which language was first, when the second language was introduced, what proportion of the child’s waking hours are spent in each language, what the language environments look like (home, school, extended family, community), and how the family thinks about the child’s language development — is the variable that determines whether a low English score reflects exposure or impairment. The school SLP cannot answer that question from the language sample alone; they need the family interview, ideally conducted in the family’s preferred language and with enough time to capture not just the answers to a checklist but the qualitative description of the child’s communicative style at home.

The structured family interview should cover: age of first exposure to each language, current proportion of exposure to each language across waking hours, primary communication partners in each language, language used for homework and academic tasks, language used for play with siblings, any family concerns about communication, and any history of family members with similar concerns. The last item matters because language disorders run in families and a positive family history shifts the prior probability that an observed delay reflects underlying impairment rather than exposure. The bilingual research literature has shown that family history is one of the strongest single predictors of language disorder in bilingual children, often outperforming any individual normed score.

The interview should also capture the family’s preferred language for school communication and IEP meetings. A family who prefers Spanish-medium communication is entitled to it under federal law, and an IEP meeting conducted entirely in English without an interpreter is not a procedurally valid IEP meeting for a Spanish-preferring family. The school SLP is not solely responsible for arranging the interpreter, but the SLP’s assessment report should document the family’s language preference explicitly so that the IEP team has the information when scheduling the meeting. This is a small administrative detail that becomes a procedural failure if it is missed, and the school SLP’s report is one of the documents that prevents the failure.

Pulling everything together, here is what a realistic 2026 bilingual LSA workflow looks like for a school SLP carrying a mixed monolingual-and-bilingual caseload across an elementary district. The workflow is calibrated for clinical realism on a real assessment timeline, not for research-grade methodology, and every step is named so a clinician can adopt it piece by piece without rebuilding their entire assessment process on day one.

Step one is the home-language interview with the family, conducted in the family’s preferred language and covering exposure history, current language environment, family concerns, and family history. Step two is the bilingual assessment plan: which language samples are needed (typically two language-specific samples plus an optional brief mixed sample), which adult will elicit each one, and on which dates. Step three is the English-only language sample, elicited under SUGAR conditions by the school SLP or a fluent English-speaking adult. Step four is the home-language sample, elicited under SUGAR conditions by a fluent adult or a trained bilingual interpreter — not by the English-only school SLP through gestures. Step five is the optional brief mixed sample for code-switching analysis, elicited in a free conversation context where the child can switch naturally.

Step six is the scoring step. Each language-specific sample is run through the deterministic LSA calculators on this site (MLU, NDW, PGU, DSS or IPSyn as appropriate) using the language-specific scoring rules from the code-switching section above. Step seven is the cross-reference: the English sample against the SUGAR English norms, the Spanish sample against the SUGAR Spanish norms (or against the closest published norms for the home language), and the conceptual NDW against monolingual NDW expectations. Step eight is the clinical reasoning step: identify the higher of the two language-specific scores, compare against the appropriate language-specific norms, and apply the "evident in both languages" rule to differentiate difference from disorder. Step nine is the eligibility recommendation, which is "language difference, not eligible on language-impairment grounds" if the higher score is in the typical range and "may have language disorder, recommend full bilingual multidisciplinary evaluation" if the higher score is below the typical range. Step ten is the IEP goal drafting if the eligibility decision is positive, using the IEP Goal Generator with the language-specification field set appropriately for the goal type.

The total time for a complete bilingual LSA following this protocol is roughly 90 to 120 minutes of clinician time spread across two or three sessions, plus the family interview and the case-history review. This is meaningfully longer than a monolingual LSA, but it is the only protocol the bilingual research literature recognises as defensible for a school eligibility decision. A district that wants to reduce the time burden should staff bilingual SLPs or trained bilingual interpreters; reducing the methodological rigour to fit the timeline is not a defensible response and is the source of the over-identification problem that this entire pillar is trying to correct.

ConductSpeech is built to support the bilingual LSA workflow described in this pillar in the same way it supports the monolingual workflow on the rest of the SLP caseload: HIPAA-compliant transcription, deterministic scoring through the calculators on this site, and structured-in prose-out drafting of the present-levels paragraph and IEP goals. The bilingual extensions are deliberate. The transcription pipeline supports both English and Spanish audio in the same uploaded file, with code-switching boundaries preserved in the transcript. The scoring pipeline lets the clinician score each language-specific sample with the appropriate language’s rules and report them separately. The drafting pipeline supports the language-specification field in the IEP Goal Generator, so a bilingual SMART goal can be drafted directly from the bilingual LSA results without the clinician having to reformat the template by hand.

The positioning matches the honest framing of this pillar exactly. ConductSpeech does not produce eligibility conclusions — the "evident in both languages" rule is a clinical judgement the school SLP makes from the language-specific scores, and the tool surfaces the data without pre-empting the decision. ConductSpeech does not replace the family interview — the home-language history is a face-to-face conversation between the school SLP and the family, and it stays that way. ConductSpeech does not replace the bilingual SLP or the trained bilingual interpreter — the elicitation step requires a fluent adult, and a tool cannot substitute for that. What ConductSpeech does is collapse the transcription, scoring, and first-draft paperwork steps for both language samples into a workflow that takes 30 minutes per sample instead of two to three hours, which is the time saving that makes the methodologically rigorous bilingual LSA protocol affordable on a school caseload that has not historically been staffed for it.

For a school SLP evaluating ConductSpeech on a bilingual caseload, the diagnostic questions are the same as the monolingual case plus three bilingual-specific ones: (1) Does the transcription pipeline support the home languages on your caseload? (2) Does the scoring pipeline use language-specific morphology rules for each language? (3) Does the IEP Goal Generator support the language-specification field for bilingual goals? ConductSpeech answers yes to (1) for English and Spanish today and is adding additional languages quarterly, yes to (2) with separate Spanish and English morphology rule sets, and yes to (3) with the language-specification field built into the bilingual template. The honest framing for the bilingual case is the same as the honest framing for every other case in this article: the clinician owns the judgement call, the calculator owns the math, and the AI saves the clinician the hours that would otherwise be spent on transcription and first-draft paperwork.