Hey, remember that time a study that aggregated speech norms broke the internet? Haha, wow, that was fun, and also traumatic. I’m glad there aren’t too many papers that challenge long-standing ‘norms’ and fundamentally shift the information speech pathologists are communicating to parents.
 

*Record scratch*

OK, so if you’re in pediatrics (and you’re reading this, so you probably are) you might, at some point, have used some sort of speech intelligibility patter with parents and teachers like: ‘Kids should be 50% intelligible to strangers at two years, 75% at three years, and 100% intelligible (albeit not error-free) by four years.’ I know I have. It’s easy to remember, and intelligibility is a nice, functional metric by which to measure speech. But did you know these numbers are based on parent report? That is… not ideal, for a variety of reasons, not the least of which is the number of parents who tell us their child said “milk please, mother,” clear as day at two months of age. You might have also found that these numbers didn’t quite fit the typically developing kids in your own life.

Hustad et al. agreed, and then they went and did something about it. Using a sample of over 500 typically developing children who spoke American English, they elicited a standardized sample of single words and sentences. They then got a large number of adults to orthographically transcribe recordings of these samples, checked how on point they were and used mathematical modeling to work out, from that data, how intelligibility changes over childhood. This is a data-heavy paper, so we’re just going to play the highlights for you. 

Let’s start with what to expect from children at different ages, based on the models in this paper:

The authors looked at eight and nine-year-olds as well, but intelligibility growth slowed dramatically by this age—we’re talking at most, around 3 percentage points per year.
 

But this is really different from what we were using before, hey? 

Yup, it's off by a year or more. The authors kindly provide their own variation of the old 50%/75%/100% metric, when they tell us children should be 50% intelligible by 4 years, 75% intelligible by 5 years, and 90% intelligible a little past 7 years.

When you think about the methods, the difference isn’t so surprising. Parents are much better at understanding their own children than unfamiliar adults, so it makes sense that they’d overstate intelligibility estimates. And the adults in this study had zero context for what these kids were saying, which is not typically the case when we converse with children, so these results likely understate functional intelligibility. It would be nice to see some research on intelligibility with context, but in the meantime, we can probably assume it sits somewhere between these new values and the old parent-reported values. On a similar note, I’ll admit to some pretty considerable surprise that even between 9;6 and 10 years, we only expect 92.8% intelligibility at a decontextualized single-word level. I sort of wanted some adult data to compare to, because this seems really low, given that the stimuli were controlled and the kids weren’t like, naming the endless different types of Pokémon.
 

Another interesting finding from this paper was around the question, ‘when can we expect intelligibility to improve the fastest?’ The answer is between around 30 to 41 months (i.e. from around 2;6 to around 3;5), with less intelligible children tending to experience this ‘accelerated period’ later than more intelligible children. This matters for treatment planning (maybe we could try to ‘ride the wave’ and support children with speech sound disorders with some more intensive input during this period), as well as for counseling parents.
 

An important caveat is that the kids in this study were overwhelmingly white children of college-educated parents in one fairly small geographic area, and that the listeners, based on the recruitment methods and mean age, seemed to be largely college students in the same geographic area. The authors don’t discuss any dialectical differences, but this is definitely a factor we clinicians need to consider. Replication with a more diverse range of speakers and listeners would be really useful to help us generalize these data to our caseloads.

Nevertheless, what we can take from this is that our old 50/75/100% thing has to go. It looks like intelligibility development is more complex, more variable, and slower than we thought. So, let’s rethink our messaging, rewrite our parent handouts, and brainstorm how to use this new knowledge about periods of faster growth to our advantage. I think we understand one another.

This review is part of our “Start Here” content guide for Language Assessment in EI.