The neurons in our brain communicate in a language that is encoded within patterns of joint electrical activity. We could try to catalog a “dictionary” of the individual words in those patterns, but learning from such a dictionary would be impractical: There are a huge number of potential words in the language. A new study
appearing in eLife
suggests how producing a “thesaurus” for the brain’s language could enable us to understand words that have not been seen before; its findings reveal that the vocabulary used by nerve cells when they transmit information is rich in “synonyms.” According to Prof. Elad Schneidman
of the Weizmann Institute of Science’s Neurobiology Department, who led the study, the organization of the neural code he and his group uncovered may have evolved to enable our brains to decode information and interpret new input. The setup they propose is highly similar to certain telecommunications systems designed to deal with getting signals through noise.
Schneidman, who conducted the study together with former research student Dr. Elad Ganmor and Prof. Ronen Segev of Ben-Gurion University of the Negev, explains that understanding the most basic mechanisms of sensing or thought have proved challenging computationally. Neurons work together in large groups – hundreds or thousands may fire together within different brain modules – and this means that the number of possible patterns created by the “zeroes and ones” of active and silent cells is somewhere in the high gazillions. On top of that, the cells are unreliable and prone to static in their responses; so although each neuron is mostly silent, with hundreds of billions of them, the brain’s communications networks are highly noisy. Schneidman and others have made progress in deciphering some of the principles governing the vocabulary of these firing patterns, but the underlying semantic structure – the rules and syntax governing the brain’s communications – have remained obscure.
Part of the problem has been the difficulty in zooming in to the activities of a large handful of nerve cells at once. To address this problem, the researchers used recordings of electrical activity in patches of the retina of a salamander, which were conducted in Segev’s lab. These patches, containing hundreds of cells, were kept alive and functional in a lab dish for many hours. In this setting, the firing patterns of the cells were recorded while film clips 10 seconds long were shown in repeating loops, hundreds of times over. This enabled the researchers to precisely observe the response: The firing pattern of a group of neurons at each frame in the clip was logged as a single word.
Same image, different words
But as the clips were displayed again and again, the nerve cell patches kept responding with different words for the same image. A closer look showed that these were sometimes just “alternate spellings” – the difference of a letter or two – but often they were different words altogether. “The accepted idea was that a small difference in spelling would not change the meaning. But our findings show, surprisingly, that this could be like writing ‘night’ instead of ‘light.’ The neurons often seem to use something like ‘lamp’ and ‘torch’ along with ‘light’ to describe the same thing,” says Schneidman.
By analyzing the various responses to each image in the clip, the group was able to define a cluster of synonyms for every stimulus. The design of codes in telecommunications engineering offers an explanation for this “synonym cluster” organization: If the space of all possible neural words could be plotted as a sort of graph, this arrangement would place each word close enough to be included in its semantic cluster but far enough away from other clusters to avoid confusion. This concept is used to transfer information in noisy man-made systems, precisely because it avoids the accidental “night-light” mix-up.
To test this idea, the group showed a new film clip to the neurons and recorded the responses. After putting these responses into the synonym cluster model, they were able to “translate” the new firing patterns back into a description of the visual imagery in the clip.
This new insight into the language of the brain may be an important step in the endeavor to crack its rules of syntax. The number of possible patterns that can arise from the brain’s neurons is so vast that no pattern need ever repeat itself over a lifetime, and this, says Schneidman, means that the brain is constantly using new words. “We are coming to understand that noise is a feature of our brain’s language scheme – not a fault in the system – so a ‘thesaurus’ may indeed be the thing that enables us to learn and generalize to other types of brain processes. So,” says Schneidman, “we need to figure out how our brains turn that noisy code into perception, consciousness, and thought.”
Prof. Elad Schneidman’s research is supported by Martin Kushner Schnur, Mexico; and Mr. and Mrs. Lawrence Feis, Winnetka, IL.