Today's Edition
KHALED AL KHAWALDEH (ABU DHABI)
Natural Language Processing (NLP), the underlying technology that gives applications like ChatGPT and other generative AI models the ability to understand, comprehend and dictate human languages, is perhaps one of the most interesting and pioneering technologies today.
Professor Preslav Nakov, Department Chair and Professor of NLP at MBZUAI caught up with Aletihad on the sidelines of the International Conference on Computational Linguistics (COLING 2025), one of the largest gatherings of NLP experts anywhere in the world. He said the technology has gone through various revolutions to reach the stage it is at now.
"Natural Language Processing, like AI in general, has gone through an evolutionary development, at the same time it also went through several mini revolutions," he said.
"The first revolution was when people switched from rule-based systems to machine learning."
Dr. Nakov explained that early NLP models relied on predefined rules, crafted by experts to dictate the system's behaviour. Machine learning introduced a paradigm shift, enabling models to learn by observing data instead of following explicit instructions. This marked a monumental shift, paving the way for more adaptable and efficient systems.
"The next revolution was when people switched to Corpus linguistics, which allowed you to really learn the real language," he said.
He elaborated that this meant leveraging vast amounts of real-world language data, which allowed models to gain deeper insights into linguistic patterns and context. This shift was instrumental in creating systems capable of handling the complexity of natural language, establishing a new standard in accuracy and reliability.
"After that, the next revolution was the neural networks, and now, the latest revolution is the kind of deep learning and the very, very latest one which really takes this to the next level with generative pre-training models like ChatGpt," he said.
According to Dr. Nakov, neural networks is a concept which has been revisited and refined multiple times, and has become central to NLP's development. Inspired by the human brain, these systems provided unprecedented capabilities in processing language contextually and holistically. With their resurgence, neural networks cemented their role as the backbone of modern NLP.
According to Dr. Nakov, the latest generative models, such as ChatGPT, differ from their more traditional systems, which often required domain-specific customisation, instead offering versatility-handling tasks like translation, summarisation, and document analysis seamlessly. While not flawless, they represent a significant leap forward in functionality.
Despite their versatility, Dr. Nakov says no single language model dominates the landscape, a feature that he believes will continue into the future. He believes the field will evolve into a diverse ecosystem, where different models excel in specific domains.
"I don't think that there's going to be one system to one language model to rule them all. Yeah, there's one supercomputer and one super AI, right? I mean, this was kind of what many people thought initially when ChatGpt came, but by now, it's clear that this space is getting fragmented," he said.
Institutions like the Mohamed bin Zayed University of Artificial Intelligence are driving this diversity. Dr. Nakov highlighted the university's collaboration on Chase - a leading Arabic language model - which he said demonstrates the value of specialsation. Similarly, efforts to develop models for Hindi, genome data, and other niche applications highlight the growing trend towards domain-specific innovation.
