Fuzzy lexical acquisition of adjectives

This paper details a method for detecting senses (meaning) of entries for an English lexicon. The examples are based on size adjectives within a semantic-based natural language processing system such as the Ontological Semantic Technology system described in [1] and developed and improved elsewhere [2]-[6]. The method attempts to calculate the approximate meaning of unknown words. For the described case of adjectives, the system observes what properties the modified nouns have. Then, specific instances of the adjective are tracked via a set of determiners from [7] to find out whether the unknown adjective adjusts size to bigger or smaller than average. At the end of any step, a partially completed entry is created for the user to rectify and finalize. The method demonstrates its feasibility of generalization to other properties corresponding to adjectives and, potentially, other parts of speech.

I. Introduction

This paper is a new installment to the research strand on the inherent and ubiquitous fuzziness of natural langauge (e.g., [4]–[5]), which crucially affects natural language understanding. It describes a method for determining the meaning of adjectives in a semantic based approach to natural language processing (NLP). The method presented in this paper is tested on size adjectives from the Brown corpus [8]. It is assumed ([1], [3], [5]), that ontological and lexical acquisition tools are a desired part of any language processing system due to the constant evolution of language. As such, the meaning of the evolved words is inherently fuzzy, as confirmed by the acquisition method. In other words, we want to allow the system to detect any new senses for existing words, existing senses which become defunct, and entirely new words. The focus on adjectives is due to the relative (but fuzzy) consistency of their meanings as (primarily) operators on nouns.