How Limited is the Limit?

International Conference RANLP 2009 - Borovets, Bulgaria, pages 270–274 How Limited is the Limit? Prakash Mondal International Institute of Information Technology, Gachibowli Hyderabad 500032 prakash.mondal@research.iiit.ac.in Abstract In this paper an exploratory map of what intelligent natural language processing systems can achieve will be drawn up, given the advances that have been made in recent years as revealed in the latest developments in practical applications of natural language technology in areas as diverse as natural language generation, natural language understanding, machine translation, dialog system etc. Here a mathematical exploration of the issue in question will lay out the constraints on what they can achieve in their goal of automatizing language processing that humans do. It will be shown that these constraints together constitute a fundamental limit which these systems seem to fail to cross. Keywords Intelligent natural language processing systems; constraints; mathematical exploration; language processing. 1. Introduction In recent years we have encountered a massive change in our conception of what natural language processing systems have achieved [1]. We have also gained a broader understanding of conceptual and empirical challenges that we face today in designing and implementing better systems that can be robust without incurring heavy computational costs [2]. With all this we are perhaps moving more towards the goal of automatization of human language processing in machines. But in spite of what has been gained in terms of theoretical and conceptual understanding of the problems, challenges, prospects in building natural language processing systems, there still seems to be an enormous gap between the level of performance these systems have come up to and how human language processing occurs [3] [4]. Is there any fundamental reason why the gap cannot be bridged? If there is any reason, why cannot we overcome it? And what is it about us that makes us do effortlessly all that these systems are designed to do, but are still far behind fully being capable of doing? It would be argued that a fundamental answer to all these questions perhaps exists. And the fundamental answer underlies a fundamental nature of human language processing mechanism. 2. What has been achieved In recent years we have seen a spurt in the growth of computational models and practical systems in natural language processing. In the domain of natural language generation systems we have seen massive developments in phrase-based and feature-based systems of natural language generation [5], [6]. In the case of natural language understanding we have gone through ELIZA, PARRY, SIR etc. and we have viewed a range of rule-based and statistical natural language understanding systems for tutoring, medical advising etc. [7], [8] [9]. At the same time developments in parsing technology have moved on from rule-based models to data-driven statistical models and now future progress is moving toward hybrid models [10]. In addition, parallel progress has been made in machine translation systems and spoken dialog systems as well [11], [12], [13]. However, even if a lot has been achieved so far, a whole lot more is still to …

Content truncated to 500 words