Microblog

8/22/2024

I just finished Mark Rosenfelder's excellent article entitled "When do people learn languages?", in which he spends an entire section debunking a widespread myth: that children learn languages easily. I've long disagreed with that claim, so I was happy to come across his article.

Here's a summary of his arguments:

• "Children begin learning languages at birth... and haven't really mastered it subtleties before the age of ten years"
• Language learning isn't effortless for children: "children don't learn a language if they can get away with not learning it"
• "A child is likely to end up as a fluent speaker of a language only if there are significant people in her life who speak it"
• "It's a myth that children learn to speak mainly from their parents. They don't: they learn mostly from their peers"

He mentions that many people believe children to learn language better than adults, then refutes this idea.

"One may fall back on the position that language may be hard for children to learn, but at least they do it better than adults. This, however, turns out to be surprisingly difficult to prove. Singleton examined hundreds of studies, and found them resoundingly ambiguous. Quite a few studies, in fact, find that adult learners progress faster than children .... Even in phonetics, sometimes the last stronghold of the kids-learn-free position, there are studies finding that adults are better at recognizing and producing foreign sounds."

And he mentions a few reasons that children learn languages so well:

• "They can devote almost their full time to it. Adults consider half an hour's study a day to be onerous."
• "Their motivation is intense .... children can get very little of what they want without learning language(s)."
• "Their peers are nastier. Embarrassment is a prime motivating factor for human beings"

8/22/2024

See this thread on X about peculiarities of Danish. I'm pretty skeptical of claims that "all languages are equally complex", partially because of research like this. Interesting findings:

• "Danish has an unusual speech opacity (consonant reduction)"
• "Danish children do present delays in language acquisition: At 15 months, Danish children possess a median vocabulary of 90 words, compared to 140 for Norwegian kids and 150 for Swedish ones"
• "Up to 8 years of age, Danish children have more difficulties [compared to other Scandinavian children] with inflectional morphology, e.g. declining regular and irregular verbs in simple past"
• "Adult native speakers of Danish do NOT seem to have issues with Danish"
• "Danish native speakers are equally affected by near and distant sentential context in their disambiguation of words and phonemes, while Norwegians just won't wait for distant contexts and will make their decisions earlier."
• "The speech signal does not get clearer as Danish native speakers grow up, so we hypothesized that they learn to put increased focus on additional sources of information (e.g. context)"

They ran a nifty experiment to demonstrate this. Two cool findings:

• "Danish native speakers are equally affected by near and distant sentential context in their disambiguation of words and phonemes, while Norwegians just won't wait for distant contexts and will make their decisions earlier."
• "Forcing them to wait makes them more similar to Danish native speakers"

I love learning about stuff like this. If you do too, try reading about the difference in rates of dyslexia across languages!

One more thing: check out the Duostories for Toki Pona and Toki Pona Glyphs (sitelen pona!)

8/14/2024

First, interesting things that I've read recently:

• In 5 Excerpts From JD Vance’s Emails to a Transgender Classmate, Vance comes off as a fairly moderate and thoughtful person — comforting if he ends up becoming VP.
• I found One Israeli Hostage’s Unusual Experience in Gaza fascinating — particularly Atzili's description of her captors. She described them as educated, English-speaking, highly religious members of Hamas who seemed protective and decent, and who hadn't known about the looting or taking of female captives.
• The Wikipedia article about Kurrent, an old style of German cursive.
• A Tweet about training GPT-2 to multiply better by starting out with CoT and then removing tokens.
• No One Expects Young Men To Do Anything and They Are Responding By Doing Nothing. I enjoy Henderson's writing on "luxury beliefs", and he points to something that I think is super important. Strong social norms can seem repressive but often make life better, especially for those who are predisposed by their genes or environments to bad choices.
• The Life of Michael Ventris, who deciphered Linear B. (I didn't know Linear B was used to write Greek until after reading!)
• The undercover agent who wasn’t chronicles the story of James Epps, who was accused of being a "false flag" government operative and instigating the Jan. 6 attack on the Capitol. Very instructive and fascinating to see how quickly conspiracy theories can grow and become weaponized even against their adherents.
• America Has Too Many Laws

  If you were to sit down and read through all of our criminal laws and regulations—or at least flip through them—you would find plenty of surprises. You would learn, for example, that it’s a federal crime to damage a government-owned lamp in Washington, D.C.; consult with a known pirate; or advertise wine by suggesting its intoxicating qualities.

• A reunion, a vigil and reasons to celebrate discusses the life of Pacific Islanders in the U.S. and especially Utah. Interesting quote from the article:

  Feltch-Malohifo’ou believes some behavior problems could be addressed if officials understood Pacific Island culture better. “I’ve told judges that if you put a mother on the stand, a church leader, an elder and they would have to do community service for the offender, you would never have anybody else go back to jail. Because they are not going to let their mother paint graffiti off a wall. I’ve told judges they’ve got to use the cultural things that matter,” she said.

• The Case for Choosing Death, Not Immortality

7/16/2024

This is pretty simple, but quite helpful if you're running jobs on a compute node that doesn't have internet access.

On the login node or another machine with internet access, run the following Python code:

import datasets
 
x = datasets.load_dataset("my_dataset")
 
x.save_to_disk("./my_dataset_local")

Then, if needed, copy the files to the machine running your job. Now, from that offline machine, loading the dataset is simple!

y = datasets.load_from_disk("./hellaswag_local")
y # DatasetDict({...})

7/16/2024

• You can turn on Markdown detection in Google Docs! Just go to Tools > Preferences > Automatically detect Markdown
• Brave Browser for iOS supports swiping left and right between between tabs — you just have to swipe below the URL bar.
• If you need to run a .ts file from the command-line, use Bun instead of ts-node! It's much simpler and doesn't have all the weird issues with the package.json "type" field.