- Researchers from MIT and Google Brain discover how to use deep learning to decipher ancient languages.
- The technique can be used to read languages that died long ago.
- The method builds on the ability of machines to quickly complete monotonous tasks.
There are about 6,500-7,000 languages currently spoken in the world. But that's less than a quarter of all the languages people spoke over the course of human history. That total number is around 31,000 languages, according to some linguistic estimates. Every time a language is lost, so goes that way of thinking, of relating to the world. The relationships, the poetry of life uniquely described through that language are lost too. But what if you could figure out how to read the dead languages? Researchers from MIT and Google Brain created an AI-based system that can accomplish just that.
While languages change, many of the symbols and how the words and characters are distributed stay relatively constant over time. Because of that, you could attempt to decode a long-lost language if you understood its relationship to a known progenitor language. This insight is what allowed the team which included Jiaming Luo and Regina Barzilay from MIT and Yuan Cao from Google's AI lab to use machine learning to decipher the early Greek language Linear B (from 1400 BC) and a cuneiform Ugaritic (early Hebrew) language that's also over 3,000 years old.
Linear B was previously cracked by a human – in 1953, it was deciphered by Michael Ventris. But this was the first time the language was figured out by a machine.
The approach by the researchers focused on 4 key properties related to the context and alignment of the characters to be deciphered – distributional similarity, monotonic character mapping, structural sparsity and significant cognate overlap.
They trained the AI network to look for these traits, achieving the correct translation of 67.3% of Linear B cognates (word of common origin) into their Greek equivalents.
What AI can potentially do better in such tasks, according to MIT Technology Review, is that it can simply take a brute force approach that would be too exhausting for humans. They can attempt to translate symbols of an unknown alphabet by quickly testing it against symbols from one language after another, running them through everything that is already known.
Next for the scientists? Perhaps the translation of Linear A - the Ancient Greek language that no one has succeeded in deciphering so far.
You can check out their paper "Neural Decipherment via Minimum-Cost Flow: from Ugaritic to Linear B" here.
Noam Chomsky on Language’s Great Mysteries
U.S. Navy ships
The U.S. Navy controls patents for some futuristic and outlandish technologies, some of which, dubbed "the UFO patents," came to light recently. Of particular note are inventions by the somewhat mysterious Dr. Salvatore Cezar Pais, whose tech claims to be able to "engineer reality." His slate of highly-ambitious, borderline sci-fi designs meant for use by the U.S. government range from gravitational wave generators and compact fusion reactors to next-gen hybrid aerospace-underwater crafts with revolutionary propulsion systems, and beyond.
Of course, the existence of patents does not mean these technologies have actually been created, but there is evidence that some demonstrations of operability have been successfully carried out. As investigated and reported by The War Zone, a possible reason why some of the patents may have been taken on by the Navy is that the Chinese military may also be developing similar advanced gadgets.
Among Dr. Pais's patents are designs, approved in 2018, for an aerospace-underwater craft of incredible speed and maneuverability. This cone-shaped vehicle can potentially fly just as well anywhere it may be, whether air, water or space, without leaving any heat signatures. It can achieve this by creating a quantum vacuum around itself with a very dense polarized energy field. This vacuum would allow it to repel any molecule the craft comes in contact with, no matter the medium. Manipulating "quantum field fluctuations in the local vacuum energy state," would help reduce the craft's inertia. The polarized vacuum would dramatically decrease any elemental resistance and lead to "extreme speeds," claims the paper.
Not only that, if the vacuum-creating technology can be engineered, we'd also be able to "engineer the fabric of our reality at the most fundamental level," states the patent. This would lead to major advancements in aerospace propulsion and generating power. Not to mention other reality-changing outcomes that come to mind.
Among Pais's other patents are inventions that stem from similar thinking, outlining pieces of technology necessary to make his creations come to fruition. His paper presented in 2019, titled "Room Temperature Superconducting System for Use on a Hybrid Aerospace Undersea Craft," proposes a system that can achieve superconductivity at room temperatures. This would become "a highly disruptive technology, capable of a total paradigm change in Science and Technology," conveys Pais.
High frequency gravitational wave generator.
Credit: Dr. Salvatore Pais
Another invention devised by Pais is an electromagnetic field generator that could generate "an impenetrable defensive shield to sea and land as well as space-based military and civilian assets." This shield could protect from threats like anti-ship ballistic missiles, cruise missiles that evade radar, coronal mass ejections, military satellites, and even asteroids.
Dr. Pais's ideas center around the phenomenon he dubbed "The Pais Effect". He referred to it in his writings as the "controlled motion of electrically charged matter (from solid to plasma) via accelerated spin and/or accelerated vibration under rapid (yet smooth) acceleration-deceleration-acceleration transients." In less jargon-heavy terms, Pais claims to have figured out how to spin electromagnetic fields in order to contain a fusion reaction – an accomplishment that would lead to a tremendous change in power consumption and an abundance of energy.
According to his bio in a recently published paper on a new Plasma Compression Fusion Device, which could transform energy production, Dr. Pais is a mechanical and aerospace engineer working at the Naval Air Warfare Center Aircraft Division (NAWCAD), which is headquartered in Patuxent River, Maryland. Holding a Ph.D. from Case Western Reserve University in Cleveland, Ohio, Pais was a NASA Research Fellow and worked with Northrop Grumman Aerospace Systems. His current Department of Defense work involves his "advanced knowledge of theory, analysis, and modern experimental and computational methods in aerodynamics, along with an understanding of air-vehicle and missile design, especially in the domain of hypersonic power plant and vehicle design." He also has expert knowledge of electrooptics, emerging quantum technologies (laser power generation in particular), high-energy electromagnetic field generation, and the "breakthrough field of room temperature superconductivity, as related to advanced field propulsion."
Suffice it to say, with such a list of research credentials that would make Nikola Tesla proud, Dr. Pais seems well-positioned to carry out groundbreaking work.
A craft using an inertial mass reduction device.
Credit: Salvatore Pais
The patents won't necessarily lead to these technologies ever seeing the light of day. The research has its share of detractors and nonbelievers among other scientists, who think the amount of energy required for the fields described by Pais and his ideas on electromagnetic propulsions are well beyond the scope of current tech and are nearly impossible. Yet investigators at The War Zone found comments from Navy officials that indicate the inventions are being looked at seriously enough, and some tests are taking place.
If you'd like to read through Pais's patents yourself, check them out here.
Laser Augmented Turbojet Propulsion System
Credit: Dr. Salvatore Pais
Robert Erdmann, a senior scientist working for the Rijksmuseum, cannot help but smile when I ask him to explain — in as much detail as possible — how exactly he used artificial intelligence to recreate long-lost portions of Rembrandt van Rijn's most famous painting, The Night Watch (1642). "Most people just want the elevator pitch," he tells me over Zoom.
The Night Watch is a mammoth of a painting, and it used to be even bigger. In 1715, it came into the possession of the bureaucrats in charge of Amsterdam's Town Hall. In order to fit it on their wall, they sliced off all four outer edges of Rembrandt's priceless masterpiece, inadvertently creating the compromised version we know today.
Rembrandt's "The Night Watch," with the missing edges shown in black.
The missing pieces of "The Night Watch" were never recovered, but we know what they looked like thanks to Gerrit Lundens, a contemporary of Rembrandt who copied the painting when it was complete. These missing sections depict the top of the arch, a balustrade at the bottom, and two soldiers of Frans Banninck Cocq's militia company that stood at the far left.
Though the absence of these elements does not make "The Night Watch" any less impressive, their presence greatly alters the painting's look and feel. The balustrade emphasizes the company's movement forward. Together, the four missing pieces shift the principal figures — Cocq and Willem van Ruytenburch — to the right, creating a more compelling composition.
Copy of "The Night Watch" by Gerrit Lundens.
As part of Operation Night Watch, a multimillion-dollar restoration mission, the Rijksmuseum set out to recreate these missing pieces of the painting to show visitors The Night Watch as Rembrandt had originally constructed it. One easy way to do this would be to upload the smaller Lundens copy into Photoshop, blow it up by a factor of five, print it out, and call it a day.
Easy, but far from adequate. As Erdmann puts it: "There's nothing wrong with using an artist like that. However, the final product would still contain traces of that artist's own style." For Erdmann, the only viable solution was to create a series of neural networks — software that mimics the human brain through the use of artificial neurons — to transform the Lundens copy into an "original" Rembrandt.
Humans, unlike computers, aren't able to make perfect copies. Faithful though Lundens' painting is — especially in its visual detail, for example, the number of buttons on a coat, plumes on a feather, or engravings on a halberd — it still contains a myriad of miniscule differences that prevented Erdmann from simply copy-pasting it onto the original.
Perspective was the first and arguably most important item on Erdmann's list. "The geometric correspondence is pretty good at the bottom of the copy," he says. "At the top, that correspondence starts to fall apart; the composition looks stretched out, supposedly because Lundens was unable to reach the top of the painting to get its precise measurements."
Lundens' copy, adjusted for perspective by the AI.
After creating a neural network that could identify corresponding elements in both versions of The Night Watch — from faces and hands to clothing and weapons — Erdmann made a second neural network that could stretch, rotate, foreshorten, compress, and decompress the Lundens copy so that its measurements matched the Rembrandt original as closely as possible.
According to Erdman, this step was "a guide to where we should place the figures on the left, because they need to be consistent with the extrapolation from the original Night Watch." Aside from aligning the two paintings, Erdmann's adjustments also transformed the facial structure of figures like Cocq, bringing them closer to Rembrandt's expert rendering.
Detail of the Lundens copy before perspectival adjustments.
Detail of the Lundens copy after perspectival adjustments.
Just as a painter must tone their canvas before they can work on composition and color, so too did Erdmann have to get the dimensions right before he could move on to the third and final stage of his coding process. Erdmann's next part of the neural network involved — to paraphrase his elevator pitch — sending the artificial intelligence algorithm to art school.
"Not unlike how you might translate a text from Dutch to English, we wanted to see if we could transform Lundens' painterly style and palette into Rembrandt's," he explains, comparing the learning curve to a quiz. To educate it, the AI was given random tiles from the Lundens copy and asked to render the tiles in the style of Rembrandt.
As with any pedagogical situation, Erdmann evaluated the AI's efforts with a corresponding grade. The closer its output matched the contents of the original Night Watch, the higher the grade it received. When grading, Erdmann considered things like color, texture, and representation (i.e., how well does this frowning face resemble a frowning face, or this sword an actual sword?).
"Once you've defined what makes a good copy, you can train the network on thousands and thousands of these tiles," Erdmann goes on. There are 265 gigabytes of memory of thousands of attempts stored, which demonstrates improvement in quality over a very short time. Within less than a day, the error margin between the AI and the real Rembrandt grew so small it became insignificant; the training was complete.
Lundens copy when adjusted for perspective and Rembrandt's style by AI.
Along the way, the AI had developed a thorough understanding of what made Rembrandt Rembrandt. When translating Lundens' copy, it used a less saturated color palette and thicker, sketchier brushstrokes. It even adopted the painter's signature use of chiaroscuro — a technique involving sharp contrasts between light and shadow.
Then it was time for the final exam. Using the knowledge gained from copying Rembrandt, Erdmann ordered the AI to transform the four outer edges of the Lundens copy — removed from the original Night Watch — into Rembrandt's signature style. The result, an unprecedented collaboration between man and machine, is now on display in the Eregalerij of the Rijksmuseum.
Detail of the completed "Night Watch." The two figures on the left were added from the adjusted Lundens copy.
The missing pieces, resuscitated by AI, were printed onto canvas and varnished so that they had a similar gloss to the rest of the painting. The pieces were then attached to metal plates, which were placed in front of the original Night Watch at a distance of less than one centimeter, thus creating an optical illusion for visitors without actually touching Rembrandt's work.
While conservation science is evolving rapidly, the achievements of people like Erdmann are still eclipsed by the artistic genius of the painters whose work they try to preserve, which is a shame because Erdmann's software can be just as inventive as Rembrandt's brushwork. At the very least, Erdmann's problem-solving skills would have made the master proud.
