An investigation into figurative language in the ‘LOLITA' NLP system

The classical and folk theory view on metaphor and figurative language assumes that metaphor is a rare occurrence, restricted to the realms of poetry and rhetoric. Recent results have, however, unarguably shown that figurative language of various complexity exhibits great systematicity and is pervasive in everyday language and texts. If the ubiquity of figurative language cannot be disputed, however, any natural language processing (NLP) system aiming at processing text beyond a restricted scope has to be able to deal with figurative language. This is particularly true if the processing is to be based on deep techniques, where a deep analysis of the input is performed. The LOLITA NLP system employs deep techniques and, therefore, must be capable of dealing with figurative input. The task of natural language (NL) generation is affected by the naturalness of figurative language, too. For if metaphors are frequent and natural, NL generation not capable of handling figurative language will seem restricted and its output unnatural. This thesis describes the work undertaken to examine the options for extending the LOLITA system in the direction of figurative language processing and the results of this project. The work critically examines previous approaches and their contribution to the field, before outlining a solution which follows the principles of natural language engineering.