Within the framework of a cooperation for several years, IBM and the University of Stuttgart want to advance AI research on the interaction of language and knowledge. The AI Horizons Network is a worldwide network of researchers and PhD students, which was established by IBM to jointly advance the application of artificial intelligence, machine learning, machine speech processing and related techniques in a number of research projects and experiments. To date, over 80 scientific papers from the network have been published worldwide.
Chatbots now provide a user-friendly interface to information sources and various services. Whether questions about insurance benefits, the processing status of a loan, or package tracking, most of these interactions are question and answer scenarios that can be implemented within a given context using machine language processing (NLP). One of the long-term goals of the AI research community is to further develop NLP in such a way that, in the future, AI will be able to interpret human communications and interactions so well that systems will be able to formulate responses to queries independently of context. This development stage is called Natural Language Understanding (NLU).
Thanks to NLU, AI systems could in future provide answers based on the analysis of context information. For example, a language assistant in a mobile device could use GPS information to infer not only the relevance of certain place names, but also that the environment inspires the user to change the subject matter ("Do we have partner companies in this area in the field of logistics?"). In order to achieve this, scientists still have to overcome a number of challenges - including the development of an interface between unstructured data such as oral utterances, but also tweets and social media comments on the one hand and structured data, such as those found in tables or databases, on the other hand.
The development of such an interface between language and structured knowledge is the goal of the three-year cooperative project entitled "Language-Knowledge Interaction" between the Institute for Machine Language Processing at the University of Stuttgart and IBM Research Europe.
"Perhaps the most striking feature of human language is that its expressions are like chameleons, in that they can adapt their meaning according to context," explains Jonas Kuhn, Professor of Computational Linguistics at the University of Stuttgart and head of the project on the university side. "As a result, people can be surprisingly concise in communicating complex issues - even in rapidly changing contexts. Natural language understanding systems must find clues in language and text that point to the correct interpretation contexts. Deep learning methods today are able to recognize relevant patterns in training data. But it is still a great challenge to go beyond learning the right solutions for a specific task. Many application scenarios require systems that can generalize and provide a rationale for the decisions made".
"As part of the AI Horizons Network, IBM researchers are working with world-class faculties and brilliant students on a series of ambitious research projects and experiments designed to accelerate the application of AI, thereby adding value to society," says Dr. Anika Schumann, Artificial Intelligence Manager at IBM Research and responsible for the project on the company side. "The results of this collaboration have the potential to influence and change the application of AI in areas as diverse as healthcare, materials science and finance.
The network's projects are designed to apply AI to a variety of applications in areas such as health, environment, logistics, and education. The network is concerned with the entire so-called AI stack, from the analysis of the unstructured and structured data required for training the systems to the construction of novel computing infrastructures needed to optimize the new data-intensive workloads in a digital world.
The project "Knowledge-Language Interaction" will focus specifically on the development of an automated interface between unstructured and structured data using machine learning. Manually created translation rules are precise, but limited to a fixed grid of differences. Thanks to recent advances in deep learning, it is now possible to automatically induce complex, multidimensional representations for text and speech from natural usage patterns in data.
The collaboration will make the University of Stuttgart the first institution in Europe to join the IBM AI Horizons Network. Universities around the world, such as the Massachusetts Institute of Technology (MIT), IIT Bombay, Université de Montreal and University of Massachusetts at Amherst, are working with IBM within the network.