Non-Local Contexts Help Resolve Ambiguity

Abstract

This paper addresses nonlocal context effects in the interpretation of ambiguous utterances in natural language. We examine equivocation as a form of discourse ambiguity and demonstrate that nonlocal contexts can resolve ambiguity by providing a method for exploring the effects of global context. Of particular relevance is that the locus of ambiguity within the texts analyzed is within and across quotations included in larger texts that are representative of summaries of speeches as reported in newspapers. This research has relevance to sentiment analysis through the ramifications that sentiment relevant to financial markets cannot necessarily be detected from quoted texts alone, even when the text quoted in the article is that of the Federal Reserve Board chair. We think it safe to say that most research on sentiment analysis does not distinguish between direct text and text present indirectly via quotation. The texts we use as experimental items in our study involve a mixture of quoted and nonquoted statements of Alan Greenspan, texts which are relevant to domain-specific decision making. The results we report suggest that sentiment analysis research is mistaken if it does not parse for qutoational contexts of sentiment bearing words. Our results show that nonlocal contexts strongly influence decision making behavior in response to ambiguous texts.