The WordNet Database: Form, Function, and Use

摘要

Wordnet is a large lexical database of the English language. Like a regular dictionary, it indexes base form words (such as the word run) to meanings (e.g., "move fast by using one's feet" as well as "a score in baseball"). Unlike a regular dictionary, it encodes significant amounts of additional information about the interrelationships of word meanings and lexical forms. Perhaps most helpfully, it marks what words are almost exactly synonymous, and so can be used as a thesaurus in addition to a dictionary. Beyond this, however, Wordnet encodes a number of other relationships, such as the fact that an animal (synonymous with animate being, creature, or fauna) is a type of organism, which is in turn a type living thing. This is called the semantic relationship of type-subtype, and Wordnet encodes semantic and lexical relationships between its entries such as type-subtype, part-whole, substance-whole, member-set, domain-topic, antonymy, derivationally related forms, among others. In addition to this rich repository of language meaning, Wordnet is further notable for its size, containing over 155,000 base wordforms, 117,000 meanings, and 188,000 relationships beyond synonymy, including over 46,000 lexical relationships and 142,000 semantic relationships. Wordnet can be of great use to any application that has to interact with natural language text. In this tutorial, we will first learn about the form of the Wordnet database: the core concepts, what kinds of relationships are encoded in the database, and some caveats about the database contents. We will also examine a small selection of tasks enabled by each type of information encoded in the database. These tasks are provided only as a sample of potential applications, as the range of uses is limited only by one's imagination. Tasks we will learn about include low-level NLP tasks such as lemmatization or root finding (given the inflected form "running" return the root "run", or given the irregular form "is" return the root "be"), all the way up to conceptual processing tasks such as determining that cats and dogs are more similar to one another than to turtles, plants, or cars. In addition to the form and utility of the database, we will learn how to interact with the database programmatically. We will first review ways of loading Wordnet into common databases such as MySQL, Sqlite, PostgresSQL, and the like, such that it can be. After this we will examine how to interface with the database directly within a Java programming language environment, focusing on the library the MIT Java Wordnet Interface (JWI)~2. JWI is small, extremely fast, easy to use, and provides API access to all available Wordnet database information.