By Bodour Imseeh, Speech and Language Therapist. Published 24 August 2026. Last clinically reviewed 24 August 2026.
Most of us have had a version of this referral. A five year old, Urdu at home, English since nursery, scoring around the second percentile on whatever battery the service uses. The school is worried, the parents are worried, and the question underneath is the one the test cannot answer: is this a child who has not had enough English yet, or a child showing persistent language-learning difficulties across their languages?
We have more to work with than we did ten years ago, though the evidence is thinner and more skewed than most training days suggest. What follows is a working framework, with the strength of the evidence flagged as honestly as I can.
The short answer: a bilingual child should not be diagnosed with DLD on the basis of low English scores alone. The assessment needs to establish whether the child shows persistent language-learning difficulties beyond what their language exposure would predict. That requires evidence from across their languages, a detailed language history, parent report and, where appropriate, dynamic assessment.
CATALISE is unusually direct here. Consensus statement 4 reads: “Some children may have language needs because their first or home language differs from the local language, and they have had insufficient exposure to the language used by the school or community to be fully fluent in it. This should not be regarded as language disorder, unless there is evidence that the child does not have age-appropriate skills in any language.” [1]
That is a diagnostic requirement, not a courtesy. If you have assessed only English, you have not tested the criterion. The first CATALISE paper adds that where a genuine language problem exists it will be evident in the home language, and that where direct assessment is not feasible, family report and dynamic assessment are the recommended routes rather than optional extras [2].
It is worth knowing what our prevalence figures cover. SCALES gives the familiar 7.58% for language disorder of unknown origin at school entry, roughly two children per reception class [3]. It excluded children with EAL, around 10.7% of the screened sample, because assessing 64 home languages was not feasible. We have no UK prevalence estimate for bilingual children, so treat any figure you are offered with suspicion.
Bilingualism does not cause a language disorder, does not worsen one, and is not a reason to withhold a diagnosis. The RCSLT position is unambiguous on all three, and adds that bilingual people are vulnerable to misdiagnosis where diversity is mistaken for disorder [4].
Code-switching is the one I still hear described as a red flag. Spanish-English children with SLI code-switch no more often than typically developing peers, and follow the same grammatical constraints when they do [5]. In a cohort of English-Mandarin five to six year olds, more code-switching predicted better teacher-rated language six months later [6]. It tracks dominance and context, not pathology.
The silent period deserves more caution than it gets. The classic study of children in this phase found they were not silent at all: they were engaged in extensive private speech, rehearsing and recombining what they had heard [7]. Work with Early Years educators found shyness prolonged the phase, and that children did better keeping their home language going alongside non-verbal communication rather than being left quiet [8]. Reduced output is common and not alarming in itself. Using the phrase to justify a year of watchful waiting is a different thing, particularly where a child is quiet in both languages, which is not what the concept describes.
Then there is the advice to switch to English at home. A UK study of Polish-English children found Polish exposure predicted Polish outcomes, with correlations to English outcomes near zero or positive [9]. Home language maintenance alongside majority language proficiency is associated with better child wellbeing and stronger parent-child relationships, though the direction of that relationship is unestablished [10]. The parallel autism literature shows no evidence of harm from bilingual exposure, yet most parents in one review had been advised against it anyway [11].
This is the highest-yield part of the assessment and the one most often reduced to a tick box. Alongside the case history you want a separate language exposure and usage form: age of first exposure to each language, current input and output by person and setting, and the parents’ own views on bilingualism, rather than a label like “Polish at home”.
Three findings make that history interpretable. Amount of exposure matters more than timing of onset: children with roughly equal exposure to two languages matched monolinguals on receptive vocabulary whether exposure began before six months or after twenty [12], and at 30 months, toddlers with 60% or more exposure to English performed like monolingual peers [13]. Convergence also takes longer than services assume. English language learners need more than three years to approach monolingual norms, and get there asynchronously across subdomains [14]; in a longitudinal Chinese-English cohort, most reached monolingual levels on most tests by around 5.5 years of exposure [15].
Asynchrony is the clinically useful part. In a UK sample of Turkish-English children with two to nearly seven years of English, tense morphology and grammatical comprehension held up far better than vocabulary and complex syntax, and environmental factors contributed nothing to tense morphology while explaining much of the vocabulary variance [16]. That is why vocabulary is a poor basis for a diagnostic decision. Some grammatical markers may be less sensitive to differences in exposure than vocabulary, although this depends on the languages involved.
Translating a standardised test preserves linguistic equivalence at best. Functional, cultural and metric equivalence have to hold too, and usually do not: the task may not measure the same construct, items may assume unfamiliar experience, and difficulty ordering and norms do not transfer [17]. A translated test with monolingual norms is not a bilingual assessment.
Conceptual scoring is a real improvement and worth doing: it closed the monolingual-bilingual gap for receptive vocabulary in one study and narrowed it expressively [18]. But in a sample of 247 Spanish-English children, no scoring method, monolingual or conceptual, reached the minimum standard of 80% sensitivity and specificity [19]. It makes a score fairer. It does not make a vocabulary test diagnostic.
Quasi-universal nonword repetition is built from phoneme inventories and syllable structures common across languages. In the clearest demonstration of why that matters, a language-specific Dutch task dropped from 93% to 63% sensitivity with bilingual children, while the quasi-universal version held at 83% sensitivity and 93% specificity in both groups [20]. A meta-analysis of 35 studies found its discrimination approaches rather than comfortably exceeds acceptable thresholds, and recommends pairing it with other measures [21].
Sentence repetition has the strongest single-marker evidence. A meta-analysis of 46 studies across 19 languages found typically developing children outperformed children with DLD by an average of 2.08 standard deviations [22]. In Russian-Hebrew bilinguals, monolingual cut-offs produced inadequate accuracy, while bilingual cut-offs gave 100% sensitivity and 89% specificity for the societal language task alone [23]. The cut-off, not the task, is usually what fails us. The LITMUS authors advise against administering it where second language exposure is under twelve months, or without a detailed language history [24].
MAIN gives you a standardised narrative procedure across more than 100 languages, free after registration, though the protocol states it has not been normed. Macrostructure showed effects of impairment but no effect of bilingualism in one Dutch cohort [25], while an English-Hebrew study found macrostructure did not differentiate and microstructure did [26]. Insensitivity to bilingualism and sensitivity to DLD are not the same property, and macrostructure has better evidence for the first.
Parent report earns its place when direct home language assessment is not possible. The ALDeQ covers early milestones, current first language ability, behaviour and family history [27], and the PaBiQ adds structured exposure indices [28]. A questionnaire drawing on both achieved 85% sensitivity and 84% specificity in bilingual children referred to clinics [29]. Screening rather than diagnosis, but a defensible way of addressing the “any language” requirement.
Dynamic assessment is the most defensible option where no normative data exists, and the most demanding of clinician skill. The consistent cross-study finding is that clinician ratings of modifiability, how readily a child takes up teaching, discriminate between groups, while raw gain scores largely do not [30]. Reported accuracy is good: around 89% sensitivity and specificity for narrative dynamic assessment with English language learners [31], and the UK DAPPLE work found referred bilingual children needed more prompting and retained fewer taught items than matched controls [32].
Two caveats belong in the same breath. Both systematic reviews rate the included studies as methodologically poor, with small samples and few language pairs [30], [33]. And it is not free: DAPPLE runs under an hour [32], while the narrative protocol took about an hour and ten minutes to administer plus another hour of transcription and coding [31].
A meta-analysis of bilingual language assessment found that from 771 citations, only measures from 17 studies were analysable, covering roughly 100 children with language impairment, and confidence intervals for every measure included uninformative values [34]. Diagnostic accuracy evidence is still dominated by Spanish-English work in the US and a handful of European pairs. Descriptive data is a different picture, and more widely available than most of us assume.
A follow-up post, How to Find Language-Specific Evidence and Work Effectively Through an Interpreter, covers the practical steps.
None of that is a reason to fall back on an English standardised score. It is a reason to be explicit about what you tested, in which languages, and what uncertainty remains. A report saying “assessed in English only; home language reported as age-appropriate by both parents; criteria for DLD not met on current evidence” is more useful than one giving a standard score of 68 without comment. This framework will not make the decisions easy. It should make them defensible.
Bodour Imseeh is a Speech and Language Therapist specialising in bilingualism, phonology, speech sound disorders and early literacy. She works across school-based and private caseloads and is fluent in Arabic and English. Read more on the About page.
A phonology screen built for bilingual Arabic–English caseloads — score in one session, no adaptation needed.
View the screenOptional in-article product block — delete on posts that shouldn't carry it.
A practising speech and language therapist since 2021, specialising in bilingualism, phonology and literacy across the UK, US and Middle East.
More about me →
By Bodour Imseeh, Speech and Language Therapist. Published 24 August 2026. Last clinically reviewed 24 August 2026. Most of us have had a version of
حين لا يتكلّم الطفل بالقدر المتوقّع، أو حين يصعب على الآخرين فهم كلامه، من الطبيعي أن يخطر على بال الأهل أنّ اللغتين في البيت هما
انتظار أن يتكلّم الطفل أمر صعب، ويزداد صعوبةً حين يختلف من حولكم في الرأي. أحدهم يقول امنحوه وقتاً. آخر يذكر ابن خالته الذي لم يتكلّم
Waiting for a child to talk is hard, and it’s harder when everyone around you disagrees. One relative says give it time. Another mentions a
Is speaking two languages causing my child’s speech delay? If your child isn’t talking as much as you expected, or their speech is hard for
Good Arabic–English bilingual resources can be hard to find, so I’ve pulled together a list of genuinely free tools and references for assessment, intervention and
