Kow Kuroda

Large-Scale Verb Entailment Acquisition from the Web

Textual entailment recognition plays a fundamental role in tasks that require indepth natural language understanding. In order to use entailment recognition technologies for real-world applications, a large-scale entailment knowledge base is indispensable. This paper proposes a conditional probability based directional similarity measure to acquire verb entailment pairs on a large scale. We targeted 52,562 verb types that were derived from 108 Japanese Web documents, without regard for whether they were used in daily life or only in specific fields. In an evaluation of the top 20,000 verb entailment pairs acquired by previous methods and ours, we found that our similarity measure outperformed the previous ones. Our method also worked well for the top 100,000 results.

Large Scale Relation Acquisition Using Class Dependent Patterns

This paper proposes a minimally supervised method for acquiring high-level semantic relations such as causality and prevention from the Web. Our method learns linguistic patterns that express causality such as “x gave rise to y”, and uses them to extract causal noun pairs like (global warming, malaria epidemic) from sentences like “global warming gave rise to a new malaria epidemic”. The novelty of our method lies in the use of semantic word classes acquired by large scale clustering for learning class dependent patterns. We demonstrate the effectiveness of this class based approach on three large-scale relation mining tasks from 50 million Japanese Web pages. In two of these tasks we obtained more than 30,000 relation instances with over 80% precision, outperforming a state-of-the-art system by a large margin.

Generating information-rich taxonomy from Wikipedia

Even though hyponymy relation acquisition has been extensively studied, “how informative such acquired hyponymy relations are” has not been sufficiently discussed. We found that the hypernyms in automatically acquired hyponymy relations were often too vague or ambiguous to specify the meaning of their hyponyms. For instance, hypernym work is vague and ambiguous in hyponymy relations work/Avatar and work/The Catcher in the Rye. In this paper, we propose a simple method of generating intermediate concepts of hyponymy relations that can make such (vague) hypernyms more specific. Our method generates such an information-rich hyponymy relation as work / work by film director / work by James Cameron / Avatar from the less informative relation work/Avatar. Furthermore, the generated relation work by film director/Avatar can be paraphrased into a new relation movie/Avatar. Experiments showed that our method successfully acquired 2,719,441 enriched hyponymy relations with one intermediate concept with 0.853 precision and another 6,347,472 hyponymy relations with 0.786 precision.

Large scale similarity-based relation expansion

Recent advances in automatic knowledge acquisition methods make it possible to construct massive knowledge bases of semantic relations, containing information potentially unknown to their users. However for certain data mining tasks like finding potential causes of a disease or side-effects of a drug, where missing a small piece of information can have grave consequences, the coverage of automatically acquired knowledge bases is often insufficient. This paper explores the use of automatic hypothesis generation for expanding a knowledge base of semantic relations, using distributional word similarities obtained from a large Web corpus. If successful, such a method can drastically improve the coverage of automatically acquired semantic relations, at the expense of a slight reduction in accuracy. We show that large scale similarity-based relation expansion works quite well for this purpose. Using a 100 million Japanese Web page corpus as input, we could generate a substantial amount of new semantic relations that were not found in the input corpus but whose validity was confirmed in a much larger Web corpus, i.e., by using a commercial Web search engine.

A web service for automatic word class acquisition

In this paper we present a Web service for building NLP resources to construct semantic word classes in Japanese. The system takes a few seed words belonging to the target class as input and uses automatic class expansion to suggest semantically similar training samples for the user to label. The system automatically generates random negative training samples as well, and then trains a supervised classifier on this labeled data to generate the target word class from 107 candidate words extracted from a corpus of of 108 Web documents. This system eliminates the need for expert machine learning knowledge in creating semantic word classes, and we experimentally show that it significantly reduces the human effort required to build them.