Repositori URV

Identifier: TDX:911

Handle: https://hdl.handle.net/20.500.11797/TDX911

Authors:
Borge Holthoefer, Javier

Abstract:
Following a classical conception of Artificial Intelligence (one that aims a definition of cognitive mechanisms and their implementation in computers), this thesis explores the problem of knowledge organization. In particular, it draws attention to the linguistic and semantic memory, trying to find out how semantic relations emerge between words. To achieve these objectives, we rely on three main sources: use of empirical data from psycholinguistics and neuropsychology; the use of complex systems (statistical physics) methodology to build and simulate dynamic models; and finally the utilization of technologies at our disposal both for obtaining new data (Internet) as well as sufficient storage capacity and processing speed for massive data manipulation. From this point of view, rooted in Cognitive Science, many applications may arise, some of them strongly linked to current problems in the field of Computer Science, such as unsupervised information extraction, enrichment of databases and language electronic resources (Wikipedia, WordNet, etc.). and improve consultation systems (query-based systems). In Chapter 2 the methodologies that have helped build the rest of the work are established. Chapter 3 is devoted to clarify (i) the kind of data that have been used in the large-scale study of language and cognitive phenomena around it, and (ii) review some of the major contributions to the date about language and cognition. In Chapter 4 the Random Inheritance Model is introduced, which represents an attempt to understand how does semantic similarity between words and semantic categories emerge. Results are compared with empirical data obtained from responses with human subjects. In Chapter 5 we present a model of semantic degradation which emulates neurodegenerative