Due to the exponential growth of user-generated Web content and ever-increasing access of emerging countries to the Web, the demands for quality localised Information Access tools has grown stronger and stronger. Providing a quality Information Access nowadays implies, however, involving Adaptive Personalisation, Semantic Web, and Artificial Intelligence techniques to filter non-relevant, offensive, inappropriate, and harmful content that traditional Information Retrieval techniques are not able to filter. To allow such systems to operate with accuracy, Information Extraction and Knowledge Representation technologies are required; while a lot of effort has been put into developing such tools for English content, relatively little effort has been put into localising them. Localisation, as a matter of fact, implies a great deal of effort and overcoming several non-trivial challenges. First and foremost, localised Information Extraction tools must cope with different languages, which is a challenge few research works have tackled due to the lack of linguistic resources and best practices that affect several non-English idioms. Adopting the right language, however, is not enough and localised Knowledge Representation should also be culture sensitive, i.e. aware of the many cultural factors that influence people's perception and behavior, which is a topic that has been mostly neglected up to now by the Artificial Intelligence research community. In this thesis we present a comprehensive discussion of localisation of Information Extraction and Knowledge Representation techniques, introducing multilingual Keyphrase Extraction and culture-sensitive Semantic Relatedness as case studies of multilingual and multicultural knowledge-intensive applications. The several experiments performed show that the proposed techniques, framework, and systems are effective, efficient, and provide a powerful tool that can be proficiently integrated into different applications to address localization and multiculturality issues.
