How Language Processing Constrains (Computational) Natural Language Processing: A Cognitive Perspective

How Language Processing Constrains (Computational) Natural Language Processing: A Cognitive Perspective Prakash Mondal Cognitive Science Lab, International Institute of Information Technology, Room No. C-8, NBH, Hyderabad 500032, India prakash.mondal@research.iiit.ac.in Abstract. In this paper I aim at sketching out in bare outline a new model/framework of language processing with its implications for natural language processing. Research in theoretical linguistics, computational linguistics and mathematical linguistics has delineated the ways natural language can be represented and modeled. Studies in cognitive (neuro)science, on the other hand, have shown how language processing is implemented in human cognitive machinery. Here an attempt will be made to integrate all these findings into a mathematical model of language processing which will point to some potential constraints on what natural language processing can achieve or do and what it possibly cannot. Keywords: Language processing, mathematical model, theoretical linguistics, computational linguistics, cognitive science. 1 Introduction1 A lot of research in mainstream theoretical linguistics (Chomsky, 1995, 2000; Pollard and Sag, 1994; Bresnan, 1982, 2001; Jackendoff, 2002), mathematical linguistics (Manaster-Ramer, 1987; Kracht, 2003), computational linguistics/natural language processing (Berwick, 1985; Barton et al., 1987; Altman, 1995; Hausser, 2001; Roark and Sporat, 2007) has shown how the representational, structural and logical properties of natural language can be modeled. On the other hand, an emerging body of research in cognitive neuroscience of language (Levelt, 1989; Grodzinsky et al., 2000; Brown, 2001; Pulvermüller, 2002; Hagoort, 2003; Paradis, 2004; Poeppel, 2004) has revealed the ways those properties of language are represented in human brain showing their connection to cognitive mechanisms of language processing. But there seems to be a gulf between these approaches on one hand, and between the integrated results from these studies and natural language processing in artificial intelligence on the other. Apart from that, mathematical linguistics does not seem to have committed itself to considering and reflecting these issues bordering on wider cross-connections. Against this context, it becomes a pragmatic need to see how all these issues can be linked to each other. The present proposal is aimed at bridging the gulf as specified above, which will probably project far reaching implications for natural language processing in artificial intelligence in general. A general architecture of language embedded within a cognitive architecture will first be drawn up to ultimately show how it might constrain natural language processing. In the middle, a mathematical exploration of language and language processing as falling out of the general architecture of language will also be taken up in order to derive significant formal generalizations that the entire architecture offers. The aim is to strike out a synergistic confluence between the (formal) linguistic notions and concepts, and the emerging general architecture of language which will be shown to be bound up with each other. In sum, it will be argued that the nature of rules, constraints, function, processing, language universals and competence as derived from the overall general architecture Copyright 2009 by Prakash Mondal 365 23rd Pacific Asia Conference on Language, Information and Computation, pages 365–374 of language has deep implications …

Content truncated to 500 words