Enrichment of causal knowledge between nouns by extracting causal context

Abstract

Causal knowledge is pervasive in various aspects of life. It allows us to plan actions to achieve goals. Many researchers have studied the tasks of automatic identification and extraction of causal relations between nouns from natural language text. However, the noun-noun causal relations may not reveal specific circumstances in which they are plausible. We notice that in causal relations between nouns, contextual information of causes can change the possibility of bringing about effects. Two nouns, earthquake and tsunami, for example, can be extracted as showing a causal relation, but it holds true only under a certain condition such as massive undersea earthquake that increases the chance of creating a tsunami. We refer the words like massive and undersea the causal context. Assuming that a causal relation between nouns needs a causal context for enhanced understanding, we focus on extracting causal contexts for the enrichment of a given causal relation. Our research is based on the premise that it would be possible and useful to extract the causal context words from sentences encoding causality between a cause noun and an effect noun. Based on the premise, we propose an automatic method for extracting context words of a cause noun. After extracting the context words, we build a classifier for determining whether the extracted words are actually the causal context. For learning our classifier, we propose two new features: a linguistic feature and a statistical feature. The experimental result shows that these proposed features improve the performance of the classifier.