我正在使用相同的斯坦福 CoreNLP (3.8.0) 管道分析以下两个句子。
我不明白为什么即使句子在语法上相同,依赖解析器也会构建不同的树。有没有办法强制一致性?
示例 1
S1: “Admin Account” means all or any account created in connection with this website.
S2: “Admin Account” means all or any user created in connection with this website.
S3: “Admin Account” means all or any cat created in connection with this website.
S4: “Admin Account” means all or any dog created in connection with this website.
这些被解析为以下内容:
示例 2
这是另一个使用引入名词短语的同一句子的变体的示例。
这是我运行 corenlp 服务器的方式
java -mx20g -cp "*" edu.stanford.nlp.pipeline.StanfordCoreNLPServer -port 9001 -timeout 35000 -parse.model edu/stanford/nlp/models/srparser/englishSR.beam.ser.gz -tokenize.language en -tagger edu/stanford/nlp/models/pos-tagger/english-bidirectional/english-bidirectional-distsim.tagger -depparse.model edu/stanford/nlp/models/parser/nndep/english_SD.gz