You don't have to speak the home language to assess a child's speech properly

Kate Margetson's 2024 VietSpeech PhD ran four Vietnamese-English children through speech analysis. English-only, all four looked like they had SSD, if that's all you'd heard.

Both languages found SSD in two. Two of the four turned out fine, and that's two children who didn't have the disorder.

I'm not writing your clinic policy from four children. The sample's too small for that, and it's still the clearest picture we've got of what English-only scoring did.

Home-language speech is part of a real diagnosis, not a nicety. You don't have to speak that language to assess the child's speech properly. You do have to get it heard.

English-only labeled all four

English-only analysis is what made all four look disordered. Both-language analysis is what found SSD in two, and that's the only reason the group split.

The two who were fine didn't become fine because someone was nicer about bilingualism. They were fine once Vietnamese sat in the same analysis as English.

That's the stake. Not a training module. Two children who didn't have SSD, and an English-only analysis that said they did.

English-only happens for ordinary reasons. The SLT doesn't speak the home language, and a trained interpreter takes time to book. None of that makes the English score a home-language diagnosis.

Ethics training on this is already saturated. Everyone in the building has been told to assess in the home language. What's still missing, when the English report is the whole file, is that home-language sample next to a dialect-matched adult.

You can't hire a clinician for every home language on the caseload. A language-matched SLT is still the person you'd put in the room if you could. Most of us can't.

So the SLT or SLP who doesn't speak that language still owns the call. English-only is how all four VietSpeech children looked disordered.

Don't translate the English test and keep the numbers. Don't score home-language words on English norms. That's still an English yardstick.

Leave language, fluency, voice, and literacy out of this. Those aren't this job. It's SSD.

The comparison isn't an English score

A qualified interpreter, or a carefully prepared adult who speaks the child's language and dialect, elicits and records the target words. Then the child's recorded on those same words, and the SLT compares that child to that adult, not to English norms.

Dialect has to match, not only the language. A different dialect on the model is a different target. The SLT's listening for a mismatch against that speaker.

When the two don't match, it might be development, dialect, cross-linguistic transfer, ambient phonology, or a true error. English-only scoring skips that sort.

The SLT doesn't have to produce the home-language words. They've got to hear the adult next to the child, then sort the mismatch. That's the assessment in a language the SLT doesn't speak.

What would you have called the two who were fine, if you'd only heard English?

You'd have called them disordered. That's what English-only did in those four analyses.

The later SLT still has to work from what's in the chart. If the adult and the child aren't both there, they're guessing from English, and that's how all four VietSpeech children looked disordered.

There's a named 20-step protocol for this work. It's SACHL, Speech Assessment of Children's Home Language(s), written by Kate Margetson and Sharynne McLeod at the Children's Voices Centre, Charles Sturt University, Bathurst, NSW, and it hasn't been diagnostically validated.

It's still emerging. No Phase 3 clinic trial has been published yet. The protocol will be updated, and S1 from May and July 2024 is the version you've got now.

Don't talk as if your clinic already runs SACHL. There's no published SACHL-named practitioner adoption to copy from.

You don't need the protocol name on the report to do this. The handbook chapter runs to four parts and twenty steps. You still need the adult and the child in the file, compared without English norms.

A 2026 acceptability study asked 360 SLPs and students who'd already sat through SACHL presentations. That paper isn't the protocol paper, and it isn't a child-outcomes trial. Most of them still assess, transcribe, and analyze in English, and they rarely or never do that work in the home language.

Just over 87% said they'd like to use it, and time was the burden they named. That's people who already came to the talk reporting on themselves. You won't get sensitivity or specificity out of it, and it isn't a clinic audit.

That time concern isn't imaginary. Extra time is real.

Who sits as the adult is a real disagreement

There's a real disagreement here, and I'm not going to smooth it. ASHA wants a trained interpreter in the room, and it doesn't want untrained family or a minor doing that job. A child never interprets.

SACHL's step 5 still lets family sit as the assessment partner. Those two instructions don't match. Don't treat them as the same advice.

You'll book a trained interpreter when you can. If a parent or relative models the target words, they're the recorded adult model, not the child's interpreter.

Family members may not elicit a representative sample, and they can be biased. That's already in the literature as a caveat. The dialect still has to match, and it still doesn't mean you skip the adult.

Put both recordings in the chart the next therapist already opens. Label the language and the dialect. Last quarter's English scores don't count as the home-language sample.

If those two recordings are already there, comparable the way SACHL describes, you pass. Don't add a parallel process on top of a file that's already doing the job.

Two of four turned out fine only after both languages were in

Those four children only split two and two once both languages were in the analysis. English-only had all four looking disordered. Two of the four turned out fine, and that's the split you'd miss on English alone.

In Metanoa, the dialect-matched adult and the child's home-language sample can sit in the same chart. AI can structure what the therapist already heard. It doesn't diagnose, and nothing becomes the record until a person has looked.

If you'd like to see that with your own caseload languages, start at metanoa.ai.

Sources

• https://pubs.asha.org/doi/10.1044/2025_AJSLP-25-00141
• https://doi.org/10.1093/9780191964909.003.0001
• https://pubs.asha.org/doi/10.1044/2017_AJSLP-15-0161
• https://www.asha.org/practice-portal/professional-issues/multilingual-service-delivery/