JAWJAW: Java Wrapper for Japanese WordNet

JAWJAW (JAva Wrapper for JApanese Wordnet) is a Java API for Japanese WordNet (wn-ja) database, which offers access to lexical knowledge of a given word, such as hypernym, hyponym, definition, translation (English <--> Japanese).

It's an API that hides the wn-ja DB schematic details from the programmer side. We provide both the simple API for general Java programmers and more fine-grained API for Natural Language Processing (NLP) application developers.

Just call methods in the façade class. You can see the list of available methods here.

Sample code:

public class SimpleDemo {
	private static void run( String word, POS pos ) {
		// Accessing Japanese WordNet from the façade class called JAWJAW
		Set<String> hypernyms = JAWJAW.findHypernyms(word, pos);
		Set<String> hyponyms = JAWJAW.findHyponyms(word, pos);
		Set<String> consequents = JAWJAW.findEntailments(word, pos);
		Set<String> translations = JAWJAW.findTranslations(word, pos);
		Set<String> definitions = JAWJAW.findDefinitions(word, pos);
		// Showing results. (note: polysemies are mixed up here)
		System.out.println( "hypernyms of "+word+" : \t"+ hypernyms );
		System.out.println( "hyponyms of "+word+" : \t"+ hyponyms );
		System.out.println( word+" entails : \t\t"+ consequents );
		System.out.println( "translations of "+word+" : \t"+ translations );
		System.out.println( "definitions of "+word+" : \t"+ definitions );		
	}
	public static void main(String[] args) {
		// Showing a demo for "買収"(verb) which means to acquire
		SimpleDemo.run( "買収", POS.v );
	}
}

Output:
hypernyms of 買収 : [支辨, 払出す, 会計, 払いだす, 払い出す, 出費, 支出, 出金, 支払う, 払いこむ, 支弁, 精算, 得る, 手に入れる, 入手, 買い上げる, 買いつける, 買求める, 召す, 買取り, 買収, 買いあげる, 買いもとめる, 買上げる, 買い入れる, 買込む, 購う, 購買, 買い付ける, 購入, 購求, 買う, 買いとる, 買入れる, 買いいれる, 買い求める, 買い取り, 買い取る, 買い受ける, 買い出し, 買いこむ, 買取, 買取る, 買い込む] hyponyms of 買収 : [買いうける, 買い上げる, 譲りうける, 買収, 買受ける, 買い取る, 譲受ける, 買取る] 買収 entails : [支辨, 払出す, 会計, 払いだす, 払い出す, 出費, 支出, 出金, 支払う, 払いこむ, 支弁, 精算, 採択, 択む, 選む, 選考, 選分, 選び取る, より取る, 選りわける, 選取, 選定, 選りすぐる, 択る, チョイス, 択ぶ, 選りどる, 選る, 選取る, 選り分ける, 選り抜く, より分ける, セレクト, 選抜, 選する, 精選, 選り取る, 選び出す, 選抜く, より抜く, 簡抜, 選択, 選り出す, 選分ける, 選りぬく, 選出す, より出す, 選りだす, 選ぶ] translations of 買収 : [corrupt, buy, bribe, grease_one's_palms, purchase, buy_out, take_over, buy_up] definitions of 買収 : [make illegal payments to in exchange for favors or influence; "This judge can be bought", obtain by purchase; acquire by means of a financial transaction; "The family purchased a new car"; "The conglomerate acquired a new company"; "She buys for the big department store", take over ownership of; of corporations and companies]

In this API, you can get the raw content from the DB through DAO (Data Access Objects).

Data model:
Here's the domain model diagram generated from the Japanese WordNet DB schema. The API provides each data class and its DAO. Domain attributes "pos", "link" and "lang" are implmented as Enum class.

Available concept relationships:
Here's a summary of concept relationship "links" stored in the synlink table. (As of wn-ja v0.9)


Total number of concepts/words are:

• 49,190 concepts (called synsets in WordNet)
• 85,966 words
• 156,684 word definitions (pairs of word and synset)

Sample code:

public class AdvancedDemo {
	private static void run( String word, POS pos ) {
		// Access the Japanese WordNet DB and process the raw data
		List<Word> words = WordDAO.findWordsByLemmaAndPos(word, pos);
		List<Sense> senses = SenseDAO.findSensesByWordid( words.get(0).getWordid() );
		String synsetId = senses.get(0).getSynset();
		Synset synset = SynsetDAO.findSynsetBySynset( synsetId );
		SynsetDef synsetDef = SynsetDefDAO.findSynsetDefBySynsetAndLang(synsetId, Lang.eng);
		List<Synlink> synlinks = SynlinkDAO.findSynlinksBySynset( synsetId );
		// Showing the result
		System.out.println( words.get(0) );
		System.out.println( senses.get(0) );
		System.out.println( synset );
		System.out.println( synsetDef );
		System.out.println( synlinks.get(0) );
	}
	public static void main(String[] args) {
		// Showing a demo for "自然言語処理"(noun) which means NLP
		AdvancedDemo.run( "自然言語処理", POS.n ); 
	}
}

Output:
{ "wordid":"201821", "lang":"jpn", "lemma":"自然言語処理", "pron":"null", "pos":"n" } { "synset":"06142412-n", "wordid":"201821", "lang":"jpn", "rank":"0", "lexid":"0", "freq":"0", "src":"mlsn" } { "synset":"06142412-n", "pos":"n", "name":"human_language_technology", "src":"eng30" } { "synset":"06142412-n", "lang":"eng", "def":"the branch of information science that deals with natural language information", "sid":"0" } { "synset1":"06142412-n", "synset2":"06142118-n", "link":"hype", "src":"eng30" }

Refer to this page.

Download the latest version from here. (First version released on Mar 23, 2009; v1.0.0 released on Ocr 6, 2011. License: Apache License, Version 2.0）

Download the DB from Japanese WordNet website and put it under the data directory, e.g. "src/main/resources/wnjpn-0.9.db". It works on JDK 5 or later. To compile and get libraries (i.e. sqlite-jdbc-3.6.11.jar, nestedvm-1.0.jar, junit-4.0.jar), we recommend you use Maven2. With the provided pom.xml file, you can easily compile and solve dependencies with "mvn compile" and sanity-check the code with "mvn test".

• Metrics for semantic similarity/distance between two synsets Released WS4J (WordNet Similarity for Java)
• Command line interface
• Web interface

Hideki Shima at Carnegie Mellon University
Email: hideki at cs.cmu.edu