Syllable Counting And The Secret To The 'Speed' Of Languages

Recent studies from a French laboratory of linguistics reveal surprising aspects of human language that make it even more mysterious than it sounds.

So much to say, so little time...
So much to say, so little time...
Yann Verdo

LYON — We all know somebody who speaks with machine-gun speed, and we know others who speaks slowly, dragging a conversation on and on. And on ... Polyglot speakers will likely notice that the difference in the flow of speech not only varies among individuals, but also among languages as well. One need not read Murakami or Cervantes to know that Japanese and Spanish are spoken rather quickly.

These observations have led a team of linguists from the France-based Laboratory for Language Dynamics (University Lumière Lyon II) to ask an interesting question: Are languages that are spoken faster than others more efficient in transmitting information? Last month they published their findings in the journal Science Advances, and the results are quite extraordinary.

For their experiment, the linguists of Lyon applied the information theory of Claude Shannon to their recordings, asking 170 speakers of 17 different languages to read out loud a series of texts. First observation: If it seems intuitive that some languages seem faster than others to the ear this is perfectly justified. Linguists measure the readers' speech rate, by the number of syllables pronounced per second, and found that rate varies significantly between languages. Japanese speech pronounces 8.03 syllables per second while Vietnamese 5.25 and Thai 4.70. Note how variations have nothing to do with geographical distribution, since these Asian languages fall on different ends of the spectrum.

The information density of Japanese is eight times lower than that of Vietnamese.

But speech rate alone says nothing about the supposed effectiveness of information transmission of different languages. Linguists studied another parameter to determine this, one which may be a little more difficult to grasp: the syllabic density of information. The main author of the study, linguist François Pellegrino explains: "If a syllable can be easily distinguished from those which precede it, it is because it brings little information, in Shannon's sense of it; if, on the contrary, it is difficult to distinguish, it brings more information." Consider an example from French, where the commonly used "because" is a two-word construction: "parce que," so that when we hear "parce," we are able to predict with near certainty what the next syllable will be, which means that the information density of the "que" in this case is almost nil.

Not all languages ​​encode the same average amount of information (measured in bits) in each of their syllables. "Average information densities of different languages ​​vary over an interval ranging from 5.03 bits per syllable for Japanese, to 8.02 bits for Vietnamese," says Pellegrino. To say Japanese has an average information density of about 5 bits ber syllable means that there's a 1 in 32 chance of predicting the next syllable. Whereas Vietnamese amounts to a 1 in 256 chance to predict the next possibility. Thus, the information density of Japanese is eight times lower than that of Vietnamese, since it's eight times easier to predict the next syllable in Japanese.

Photo: Nick Fewings

So linguists observed an inverse relation between these two parameters — speech rate (measured syllables per second) and information density (measured in bits per syllable): The higher the speech rate, the lower the information density, and vice versa. This lead linguists to discover a remarkable phenomenon: Both parameters taken together, speech rate and information density, add up to the total information rate of a language. The information rate of all languages across the globe has been calculated at 39 bits per second. No matter how dissimilar languages appear from one another, how fast or slow they are to our ears, all languages on Earth will convey the same amount of information over a given amount of time.

"To be effective in terms of information transmission, languages have the choice between two opposing strategies: Either they favor a high rate of speech at the price of a low density of information, or they do the opposite," Pellegrino explains. French is a "medium" language in this regard, roughly equidistant from the two extremes of speech rate (6.85 syllables per second) and information density (6.68 bits per second).

Our vision of language is evolving constantly.

That information rates of languages are fixed to 39 bits per second probably owes nothing to chance, but is bounded to the cognitive capabilities of the how brains process language. If a language were below the threshold of 39 bits per second, it would not allow speakers to cope with the complexity of the world leading to that language's elimination. If it were over the threshold, cognitive capabilities would overload (we cannot permanently maintain the production or processing of too much information), and the language would die. So the threshold of 39 bits per second corresponds to a niche, both biological and cultural in origin, which defines the zone of viability in which human languages can exist.

Our vision of language is evolving constantly. Previous studies at the Lyon laboratory have shown that new sounds may appear in a language. For example, the addition of nasal vowels (like the French "an", "in", "on" etc.) doubles the total number of vowels a language has. More sounds and more syllables increase the total information density of a language. But does this mean it's possible to dislodge a language from its limit of 39 bits per second? No, responds François Pellegrino, "Our hypothesis is that a change to the structure of language modifies its syllabic density, which also leads its speakers to modify their speech rates inversely, in order to preserve optimal information rates." Not unlike the way Darwinian mechanisms in living species subject them to the laws of evolution: adapt or die.

Keep up with the world. Break out of the bubble.
Sign up to our expressly international daily newsletter!

Air Next: How A Crypto Scam Collapsed On A Single Spelling Mistake

It is today a proven fraud, nailed by the French stock market watchdog: Air Next resorted to a full range of dubious practices to raise money for a blockchain-powered e-commerce app. But the simplest of errors exposed the scam and limited the damage to investors. A cautionary tale for the crypto economy.

Sky is the crypto limit

Laurence Boisseau

PARIS — Air Next promised to use blockchain technology to revolutionize passenger transport. Should we have read something into its name? In fact, the company was talking a lot of hot air from the start. Air Next turned out to be a scam, with a fake website, false identities, fake criminal records, counterfeited bank certificates, aggressive marketing … real crooks. Thirty-five employees recruited over the summer ranked among its victims, not to mention the few investors who put money in the business.

Maud (not her real name) had always dreamed of working in a start-up. In July, she spotted an ad on Linkedin and was interviewed by videoconference — hardly unusual in the era of COVID and teleworking. She was hired very quickly and signed a permanent work contract. She resigned from her old job, happy to get started on a new adventure.

Others like Maud fell for the bait. At least ten senior managers, coming from major airlines, airports, large French and American corporations, a former police officer … all firmly believed in this project. Some quit their jobs to join; some French expats even made their way back to France.

Share capital of one billion 

The story began last February, when Air Next registered with the Paris Commercial Court. The new company stated it was developing an application that would allow the purchase of airline tickets by using cryptocurrency, at unbeatable prices and with an automatic guarantee in case of cancellation or delay, via a "smart contract" system (a computer protocol that facilitates, verifies and oversees the handling of a contract).

The firm declared a share capital of one billion euros, with offices under construction at 50, Avenue des Champs Elysées, and a president, Philippe Vincent ... which was probably a usurped identity.

Last summer, Air Next started recruiting. The company also wanted to raise money to have the assets on hand to allow passenger compensation. It organized a fundraiser using an ICO, or "Initial Coin Offering", via the issuance of digital tokens, transacted in cryptocurrencies through the blockchain.

While nothing obliged him to do so, the company owner went as far as setting up a file with the AMF, France's stock market regulator which oversees this type of transaction. Seeking the market regulator stamp is optional, but when issued, it gives guarantees to those buying tokens.

screenshot of the typo that revealed the Air Next scam

The infamous typo that brought the Air Next scam down

compta online

Raising Initial Coin Offering 

Then, on Sept. 30, the AMF issued an alert, by way of a press release, on the risks of fraud associated with the ICO, as it suspected some documents to be forgeries. A few hours before that, Air Next had just brought forward by several days the date of its tokens pre-sale.

For employees of the new company, it was a brutal wake-up call. They quickly understood that they had been duped, that they'd bet on the proverbial house of cards. On the investor side, the CEO didn't get beyond an initial fundraising of 150,000 euros. He was hoping to raise millions, but despite his failure, he didn't lose confidence. Challenged by one of his employees on Telegram, he admitted that "many documents provided were false", that "an error cost the life of this project."

What was the "error" he was referring to? A typo in the name of the would-be bank backing the startup. A very small one, at the bottom of the page of the false bank certificate, where the name "Edmond de Rothschild" is misspelled "Edemond".

Finding culprits 

Before the AMF's public alert, websites specializing in crypto-assets had already noted certain inconsistencies. The company had declared a share capital of 1 billion euros, which is an enormous amount. Air Next's CEO also boasted about having discovered bitcoin at a time when only a few geeks knew about cryptocurrency.

Employees and investors filed a complaint. Failing to find the general manager, Julien Leclerc — which might also be a fake name — they started looking for other culprits. They believe that if the Paris Commercial Court hadn't registered the company, no one would have been defrauded.

Beyond the handful of victims, this case is a plea for the implementation of more secure procedures, in an increasingly digital world, particularly following the pandemic. The much touted ICO market is itself a victim, and may find it hard to recover.

Keep up with the world. Break out of the bubble.
Sign up to our expressly international daily newsletter!