首次处理英文语料,需要进行一些基础的NLP处理,首选工具当然是Stanford CoreNLP。由于Stanford CoreNLP官方示例的解析结果不宜直接使用,所以我在它的基础上进行修改,最终将解析结果转为json格式,并依照哈工大ltp的解析结果的格式,将依存句法的解析结果也添加到json中。
1、Stanford CoreNLP的安装
最新版的Stanford CoreNLP仅支持jdk1.8,这比较奇葩,因为目前多数机器的jdk还只是1.6或1.7,最以我下载了支持jdk1.6的最后一个版本,地址:http://nlp.stanford.edu/software/stanford-corenlp-full-2014-08-27.zip 。下载完成后,将解压后的所有内容放到(Eclipse)项目的根目录下,通过Build
Path将所有的jar包添加到项目库中,即可完成安装配置。解压后的目录中有一个名为StanfordCoreNlpDemo.java的示例文件,简洁地展示了如何使用此工具,但是它使用的结果显示方式是prettyPrint,这种结果只便于人来看,而不便于机器来获取。所以我以 http://www.cnblogs.com/tec-vegetables/p/4153144.html所示的例子来基础来改写代码。
2、代码,有详细解释
import java.util.ArrayList; import java.util.HashMap; import java.util.List; import java.util.Map; import java.util.Properties; import java.util.regex.Matcher; import java.util.regex.Pattern; import net.sf.json.JSONArray; import edu.stanford.nlp.dcoref.CorefChain; import edu.stanford.nlp.ling.CoreAnnotations.CharacterOffsetBeginAnnotation; import edu.stanford.nlp.ling.CoreAnnotations.CharacterOffsetEndAnnotation; import edu.stanford.nlp.ling.CoreAnnotations.LemmaAnnotation; import edu.stanford.nlp.ling.CoreAnnotations.NamedEntityTagAnnotation; import edu.stanford.nlp.ling.CoreAnnotations.PartOfSpeechAnnotation; import edu.stanford.nlp.ling.CoreAnnotations.SentencesAnnotation; import edu.stanford.nlp.ling.CoreAnnotations.TextAnnotation; import edu.stanford.nlp.ling.CoreAnnotations.TokensAnnotation; import edu.stanford.nlp.ling.CoreLabel; import edu.stanford.nlp.pipeline.Annotation; import edu.stanford.nlp.pipeline.StanfordCoreNLP; import edu.stanford.nlp.semgraph.SemanticGraph; import edu.stanford.nlp.semgraph.SemanticGraphCoreAnnotations.CollapsedCCProcessedDependenciesAnnotation; import edu.stanford.nlp.trees.Tree; import edu.stanford.nlp.trees.TreeCoreAnnotations.TreeAnnotation; import edu.stanford.nlp.util.CoreMap; public class TestCoreNLP { //参数text为需要处理的句子 public static void run(String text) { //创建一个corenlp对象,设置需要完成的任务。 //tokenize: 分词;ssplit:分句;pos:词性标注;lemma:获取词原型;parse:句法解析(含依存句法);dcoref:同义指代 Properties props = new Properties(); props.put("annotators", "tokenize, ssplit, pos, lemma, ner, parse, dcoref"); StanfordCoreNLP pipeline = new StanfordCoreNLP(props); // 创建一个基于参数句子的标注对象 Annotation document = new Annotation(text); // 将上述标注对象将对corenlp进行处理 pipeline.annotate(document); // 获取处理结果 List<CoreMap> sentences = document.get(SentencesAnnotation.class); //遍历所有句子,输出每一句的处理结果 for(CoreMap sentence: sentences) { //遍历句子中每一个词,获取其解析结果并构造json数据 JSONArray jsonSent = new JSONArray(); //创建一个json数组,用于保存当前句子的最终所有解析结果 int id=1;//当前词在句子中的id,从1开始,因为原始的解析结果就是从1开始的。 //先获取当前句子的依存句法分析结果 SemanticGraph dependencies = sentence.get(CollapsedCCProcessedDependenciesAnnotation.class); //遍历每一个词 for (CoreLabel token: sentence.get(TokensAnnotation.class)) { //获取每个词的分析结果 Map mapWord = new HashMap();//创建一个map对象,用于保存当前词的解析结果 mapWord.put("id", id);// 添加id值 mapWord.put("cont", token.get(TextAnnotation.class));//添加词内容 mapWord.put("pos", token.get(PartOfSpeechAnnotation.class));//添加词性标注值 mapWord.put("ner", token.get(NamedEntityTagAnnotation.class));//添加实体识别值 mapWord.put("lemma", token.get(LemmaAnnotation.class));//添加词原型 mapWord.put("charBegin",token.get(CharacterOffsetBeginAnnotation.class));//添加词在句子中的起始位置 mapWord.put("charEnd",token.get(CharacterOffsetEndAnnotation.class));//添加词在句子中的结束位置 //查找每个词对应的依存关系。由于原始的解析结果中,依存关系是单独地集中在另一个字符串变量中的,形如: 依存关系名(被依赖词-被依赖词id,依赖词-依赖词id)\n 依存关系名(被依赖词-被依赖词id,依赖词-依赖词id)\n......需要对其进行解析,这里采用的方法是依据\n进行分割,然后再用正则表达式进行匹配,来逐一获取每一个词的依赖词和依存关系名 int flag=0;//设置标志位,用于保存当前词的依存关系是否已经处理过,0未处理,1已处理 String[] dArray= (dependencies.toString(SemanticGraph.OutputFormat.LIST)).split("\n");//根据\n进行分割,结果保存为字符串数组 for (int i=0;i<dArray.length;i++) //遍历字符串数组 { if(flag==1) //检查当前词的依存关系是否已经处理过,如果已处理,则直接退出遍历过程 break; ArrayList dc=getDependencyContnet(dArray[i]);//获取数组中第i项,并从中获取依存关系名,被依赖词id和依赖词id,放到一个ArrayList中 if( Integer.parseInt(String.valueOf(dc.get(2)))==id) //如果当前词id等于当前依存关系中的依赖词id,则说明找到对应的关系结构 { mapWord.put("relation",dc.get(0));//添加依存关系名 mapWord.put("parent",dc.get(1));//添加被依赖词id flag=1; // 将当前词依存关系标志设为1 break;//退出遍历 } } jsonSent.add( mapWord );//将上述结果全部添加到当前句中 id++;//词id自增 } System.out.println(jsonSent); // // 获取并打印句法解析树 // Tree tree = sentence.get(TreeAnnotation.class); // System.out.println("\n"+tree.toString()); // // 获取并打印依存句法的结果 // System.out.println("\nDependency Graph:\n " +dependencies.toString(SemanticGraph.OutputFormat.LIST)); // // 获取并打印实体指代结果 // Map<Integer, CorefChain> graph = document.get(CorefChainAnnotation.class); // System.out.println(graph); } } //解析依存关系值的方法。如,从root(abc-1, efg-3)中获取一个ArrayList,值为[root,1,3] public static ArrayList getDependencyContnet(String sent) { String str=sent; ArrayList result=new ArrayList(); String patternName="(.*)\\("; String patternGid="\\(.*-([0-9]*)\\,"; String patternDid=".*-([0-9]*)\\)"; Pattern r = Pattern.compile(patternName); Matcher m = r.matcher(str); if(m.find()) { result.add(m.group(1)); } r=Pattern.compile(patternGid); m = r.matcher(str); if(m.find()) { result.add(m.group(1)); } r=Pattern.compile(patternDid); m = r.matcher(str); if(m.find()) { result.add(m.group(1)); } return (result); } }
以“Beijing is the capital of China.”为例,结果为:
[{"id":1,"lemma":"Beijing","relation":"nsubj","parent":"4","ner":"LOCATION","charEnd":7,"cont":"Beijing","charBegin":0,"pos":"NNP"},{"id":2,"lemma":"be","relation":"cop","parent":"4","ner":"O","charEnd":10,"cont":"is","charBegin":8,"pos":"VBZ"},{"id":3,"lemma":"the","relation":"det","parent":"4","ner":"O","charEnd":14,"cont":"the","charBegin":11,"pos":"DT"},{"id":4,"lemma":"capital","relation":"root","parent":"0","ner":"O","charEnd":22,"cont":"capital","charBegin":15,"pos":"NN"},{"id":5,"lemma":"of","ner":"O","charEnd":25,"cont":"of","charBegin":23,"pos":"IN"},{"id":6,"lemma":"China","relation":"prep_of","parent":"4","ner":"LOCATION","charEnd":31,"cont":"China","charBegin":26,"pos":"NNP"},{"id":7,"lemma":".","ner":"O","charEnd":32,"cont":".","charBegin":31,"pos":"."}]