(See also my follow-up posts later: More notes from raising a late-talking kid and 
Late-talking kid part 3: gestalt language learning.)

I talked late—no words at age 2, ten words at 2¼, lots and lots of words at 2½. Or so I've been told! :-P

My younger kid is also a late talker—maybe 7 words at age like 1½, and still ~7 words now well into age 2. (And not all the same words! He added a couple and dropped a couple.) When my older kid was the age that my younger kid is now, he was chatting away in 10-word sentences. So anyway, as a Responsible Parent, I'm doing some homework about late talking. Here are some of my notes, not in any particular order.

(If you like my speculative neuroscience writing, there's a bit of it in the last section.)

Update later: Yay! Without any interventions, he finally started rapidly increasing his spoken vocabulary just a few months after I wrote this. :-)

Update even later: Ugh, his language progress got much slower after that burst. As I write this, he’s 4, and has a very obvious language delay, and this is a constant source of stress.

Books

I read the 2014 book Late-Talking Children: A Symptom or a Stage? by Stephen M. Camarata. It's excellent, I really trust the guy. I also read the 2002 book The Einstein Syndrome: Bright Children Who Talk Late by Thomas Sowell. Sowell is an economist (!) while Camarata is a professor of speech and language disorders, so predictably Camarata is much more knowledgeable. My understanding is that Sowell's books were valuable at the time for kick-starting public awareness of the existence of bright late-talking kids (more on which below), but now that Camarata's book is out, that's the best resource.

Sowell was not a late talker himself, but Camarata was, and Sowell and Camarata each have a late-talking child.

Camarata has a comprehensive discussion of the full range of conditions that lead to late talking, from autism to apraxia to "nothing in particular, just a passing phase". Speaking of which...

"Einstein Syndrome"

There's a subgroup of late talkers who are smart (when tested non-verbally), often understand spoken language despite not talking, don't have autism or hearing loss or other conditions, and whose speech delay resolves on its own and they wind up settling into the normal range for adult language skills. This turns out to be an interesting bunch—this is what Sowell dubbed "Einstein Syndrome".

I find the term "Einstein Syndrome" super-obnoxious—it sounds like bragging. If we prefer a self-deprecating title, maybe it could have been called "Ed Teller Syndrome", named after one of the worst and most destructive humans in history, afaict. ... or "Lots Of Unremarkable People You've Never Heard Of Syndrome" for that matter. Oh well. Sowell called it "Einstein Syndrome" and the term seems to have stuck.

Anyway, this group turns out to be disproportionately likely to be highly analytical, and to be part of highly analytical families, including unusual numbers of parents who are engineers, physicists, mathematicians, accountants, etc. (Also, professional musicians!) (For my part, I'm a physicist.)

Famous people with Einstein Syndrome include: physicists Albert Einstein, Richard Feynman, and Ed Teller; economists Gary Becker and Richard Rosett; mathematicians Julia Robinson and Srinivasa Ramanujan; concert pianists Clara Schumann and Arthur Rubinstein; and politicians Dick Armey and, um, Benito Mussolini.

Autism, ASD, and misdiagnoses thereof

Camarata gives the impression that these days, non-autistic late-talking kids are at very high risk of being misdiagnosed as autistic by incompetent or careless professionals. This especially applies to school screenings; school screening people may be mainly thinking about classroom placement and funding and following the bureaucratic rules and so on, instead of thinking about doing a correct differential diagnosis of the child. He goes through example after example after example drawn from the children he's worked with at his lab for late-talking children.

He says that giving a non-autistic child an autism-specific early intervention can be quite damaging and traumatic to the child. (Well, he also says that a lot of autism-specific early interventions in use today are just bad period, even for kids with autism! And even beyond autism, that there are a lot of early interventions that are universally counterproductive but still in use. It's a jungle out there!) Likewise, he says that a false autism diagnosis (or false diagnosis of Global Developmental Delay, etc.) on a school record can follow a kid for years and cause endless problems.

Incidentally, Camarata seems a bit annoyed about the whole idea of "autism spectrum"; his perspective seems to be that "classic autism" is very much more serious than the edge case kids who barely qualify as ASD (as judged by a careless person mindlessly following a checklist), and that lumping all these kids together under the banner of "autism" is causing parents, teachers, and others to have mistaken expectations, to do wrong interventions, and to misallocate resources. I have no opinion here, I just thought that was an interesting take. I think he's mostly talking about how the diagnostic criteria get translated into practice in wider society, and not about the fundamental science of autism.

For my part, it's blindingly obvious that my kid does not have classic autism—he's super into elaborate pretend play with his teddy bears (as I was at that age), etc. etc. It's still possible that he has a different condition that needs early treatment, but given the family history of non-problematic speech delay, and other factors I won't get into, I think on balance that it's fine for us to wait another few months and see if he starts talking on his own, before bringing him to a specialist. Maybe this spring or summer. On a gut level, the book definitely got me feeling more worried about my kid being mishandled and mistreated by The System than worried about his healthy development per se. Anyway, I feel ready to venture forth into that world armed with information, having read the book.

(When I was a late-talking kid, my parents got a free town-government-provided child speech pathologist (not sure her exact job title), and she came over once a week, and we played games, and my parents say it was lovely and I had a lot of fun. I figure, in all likelihood, I would have started talking at the same time with or without her. But whatever, no harm done.)

What causes "Einstein syndrome"?

No one knows, but hey, I know a bit of neuroscience, it's fun to speculate. Consider some observations about my kid:

• He has a great understanding of spoken language.
• He can sing recognizable songs (like "da-da-da" in the tune of Twinkle Twinkle Little Star)
• He can imitate non-word sounds, and animal sounds, and he babbles a lot. (Note: non-word sounds and speech are two different parts of the brain! Well, at least for listening; probably for speaking too but I haven't checked the literature, see comment.)
• I think it's quite clear that he wants to talk and is trying to talk, but he really just can't.
• He is excellent at communicating non-verbally, in all sorts of creative ways, including making up a large vocabulary of sign-language-type gestures (with sound effects!) for particular foods and activities.

OK, now we turn for inspiration to Randall O'Reilly et al.'s excellent 2017 paper Deep Predictive Learning: A Comprehensive Model of Three Visual Streams.

To oversimplify, the neocortex is a huge and important part of the brain, and every part of the neocortex is a little learning machine that finds patterns relating the inputs going into that part of the neocortex to each other and to the outputs exiting that part of the neocortex. One of the very interesting aspects of this paper is that they found (in the case of vision processing) that when they hooked all the information streams together at the same time into their final configuration, the learning algorithm just flails around, it doesn't make any progress. Instead, what they had to do was turn on the region-to-region connections in the right order.

First, they turn on one region-to-region connection, then they let the learning algorithm run until it settles down, having learned a vocabulary of patterns that relate those two information streams. Only then do they turn on a second region-to-region connection, to build higher-level patterns out of the lower-level patterns, and so on. In their words: "In the case of vision, the spatial (Where) aspect of prediction can be learned first, independent of the What aspect, and having systematic and accurate high-level spatial predictions … then partitions away that aspect of the prediction error, leaving a residual that is more about object identity (What)." 

According to the paper, developmental neuroscience data indeed supports the idea that different region-to-region brain connections appear at different times in development.

I would also speculate: Maybe these region-to-region connections are not triggered by age, or not entirely by age. Maybe the brain just monitors how frequently the synapses in some region are getting edited, and if the rate gets sufficiently low, it implies that the patterns have been learned, and that's what triggers some new region-to-region connection to come online. (I didn't just make that up entirely, I think I read someone hypothesizing this kind of mechanism in a slightly different context, namely the "sensitive periods" after which learning more-or-less stops in certain brain regions like V1.)

So anyway, I wind up with the idea that there's a region of my kid's brain already building the low-level patterns that will correspond to composing and speaking language, but that this region of my kid's brain is currently disconnected from the mouth/tongue/throat motor control area of his brain. Sooner or later, his brain will decide it's ready, that connection will finally form, and then he'll start talking! Don't know if it's true, but I kinda like that idea. Knock on wood!

Incidentally, Camarata cites this study that says that people who had specific language impairment as kids (delayed speech for no obvious reason) are 55% likely to process language mainly in the right hemisphere, 27% in both hemispheres, and just 18% mainly in the left hemisphere; whereas 90% of the control group processed language mainly in the left hemisphere. He also cites this other study along the same lines. Pretty wild, if true! Score one for cortical uniformity! Not sure how / if that relates to the previous paragraphs :-P UPDATE: Oops, on closer examination, the first study I mentioned says that adults who were speaking normally now, but had had specific language impairment as kids, had 100% left-hemisphere-lateralized speech (i.e. the typical pattern). The numbers I quoted above were for the kids with specific language impairment, i.e. who were not yet talking, or just starting to talk, or something. For the second paper, I'm not sure, but it seems likely that the kids in the experiment were not yet talking (or not much?) at the time of the experiment. So that's not as interesting as a result as I thought at first.