Microsoft and NVIDIA recently announced a new AI-driven method to interpret speech that could transform how we chat with our electronics. It’s part of a growing movement changing how computers understand speech, also called Natural Language Processing (NLP). “The models powering NLP are becoming bigger and more advanced and getting closer to human comprehension,” AI expert Hamish Ogilvy told Lifewire in an email interview. “One of the big advances is that NLP is going beyond simple keywords. You might be accustomed today to typing or speaking one or two keywords to get search results, but newer natural language processing models use context to provide richer results.”
Chat Bots
NVIDIA and Microsoft have teamed up to create the Megatron-Turing Natural Language Generation model (MTNLG), which the duo claims is the “most powerful monolithic transformer language model trained to date.” The AI model runs on supercomputers. But researchers found that the MTNLG model picked up human biases as it combed through mountains of human speech samples. “While giant language models are advancing the state of the art on language generation, they also suffer from issues such as bias and toxicity,” the researchers wrote in a blog post. “Our observations with MT-NLG are that the model picks up stereotypes and biases from the data on which it is trained.” “Google has had the obvious lead here, but NLP technology is going to be everywhere,” Ogilvy said. “For text- and voice-based searches, users can be more descriptive because NLP understands more than just the text; it understands the context of what you’re looking for to return better results.”
Quantum Chats?
Quantum computing might be one way to advance the field of NLP. On Wednesday, the company Cambridge Quantum announced lambeq, which it claims is the first quantum toolkit for NLP. The company says the tool allows the translation of sentences in natural languages using quantum circuits run on quantum computers. Quantum computing is a type of computation that uses the unusual properties of quantum states, such as superposition, interference, and entanglement, to perform calculations. “The way quantum computers handle NLP is very different from classical machines. In fact, NLP is ‘quantum native,’” Bob Coecke, the chief scientist at Cambridge Quantum, told Lifewire in an email interview. “This is due to a discovery we made some years ago, that the grammar governing sentences and meaning takes a very similar structure to the maths used to program quantum computers.” Coecke said that quantum NLP could lead to better voice assistants and translation tools. Another promising approach to improving speech recognition, called data-centric AI, was launched earlier this year. Data-centric AI focuses on the quality of data used to train a model instead of improving the algorithms. “The data-centric approach has been proven to be more efficient than the traditional model-centric approach, in terms of AI task capability improvement,” Zac Liu, a data scientist at the company Hypergiant, told Lifewire in an email interview. “In short, when data scientists improve the NLP data, it almost guarantees they will have a better NLP model and better NLP capability.” The next step is integrating computer vision models with NLP, such as training an AI model to watch videos and produce a text summary of that video, Liu said. “The application of this advancement could be limitless, from health care, reading radiological films and providing preliminary diagnosis, to designing homes, clothes, jewelry, or similar items,” he added. “The customer could explain the requirements verbally or in written form, and this description can be automatically converted to images or videos for better visualization.”
