In a joint effort between the Massachusetts Institute of Technology (MIT) and other U.S. institutions of higher education, scientists have shown off an AI system with the ability to pick up on the nuances of human language with relative ease. In addition to being able to learn lower-level language patterns automatically, this model is also capable of automatically learning higher-level language patterns that can apply to numerous languages.
The researchers trained and evaluated the model using problems from 58 distinct language-specific linguistics textbooks. Each task contained a list of words with appropriate word-form modifications. For 60 per cent of the problems, the model was able to provide a correct set of rules to explain these word-form modifications.
This technique could be applied to the study of linguistic hypotheses and the investigation of minor parallels between the ways in which distinct languages modify words. The system discovers models that can be easily comprehended by people, and it learns these models from modest amounts of data, such as a few dozen words.
Additionally, the system makes use of numerous tiny datasets rather than a single large one. This is closer to how scientists propose hypotheses, which is to look at numerous related datasets and develop models to explain phenomena across those datasets.
The researchers chose to investigate the relationship between phonology and morphology in their endeavour to create an AI system that could automatically train a model from numerous related datasets (the study of word structure).
The researchers utilised a machine-learning method called Bayesian Programme Learning to create a model that could learn grammar or a set of rules for putting words together. Using this method, the model creates a computer programme to address a challenge.
The grammar that the model believes is most likely to explain the words and their meanings in a linguistics problem is known as the programme. They created the model using Sketch, a well-known software synthesiser created by Solar-Lezama at MIT.
Meanwhile, Parkinson’s disease is renowned for being challenging to diagnose because it primarily depends on the emergence of motor symptoms such as tremors, stiffness, and slowness, although these symptoms frequently occur several years after the disease’s onset. An AI model created by MIT researchers may identify Parkinson’s disease solely by observing a person’s breathing patterns.
The tool may determine whether someone has Parkinson’s disease based on their nocturnal breathing patterns, which are like the breathing patterns that occur while sleeping. A neural network is a collection of connected algorithms that mimics the way a human brain functions.
Cerebrospinal fluid and neuroimaging have been investigated as potential screening tools for Parkinson’s disease over the years, but these techniques are invasive, expensive, and require access to specialized medical facilities, preventing them from routine testing that would otherwise enable early diagnosis or ongoing disease progression monitoring.
The MIT researchers showed that a Parkinson’s assessment using AI may be carried out each night at home while the patient is sleeping and without having to touch them.
To do this, the scientists developed a device that resembles a Wi-Fi router for a home but instead of providing internet access, it emits radio signals, examines how they are reflected off the surrounding area, and then, without any physical contact, extracts the subject’s breathing patterns. There is no effort required from the patient or caregiver since the breathing signal is then sent to the neural network to passively assess Parkinson’s.
The study, which was conducted in partnership with other universities, clinics, and the Massachusetts General Hospital, has significant implications for the development of Parkinson’s medications and clinical care.