斯坦福stanford coreNLP 宾州树库汉语短语类别表23个

灰太狼 2022-06-12 02:52 199阅读 0赞

## 短语标记17个 ##

<table> 
 <tbody> 
 <tr> 
 <td> 标注 </td> 
 <td> 英文说明 </td> 
 <td> 中文说明 </td> 
 </tr> 
 <tr> 
 <td> ADJP </td> 
 <td> Adjective phrase </td> 
 <td> 形容词短语，由JJ投射 </td> 
 </tr> 
 <tr> 
 <td> ADVP </td> 
 <td> Adverbial phrase headed by AD </td> 
 <td> 由副词开头的副词短语、状语 </td> 
 </tr> 
 <tr> 
 <td> CLP </td> 
 <td> Classifier phrase </td> 
 <td> 量词短语 </td> 
 </tr> 
 <tr> 
 <td> CP </td> 
 <td> Clause headed by C（complementizer） </td> 
 <td> 由补语引导的补语从句，关系从句 </td> 
 </tr> 
 <tr> 
 <td> DNP </td> 
 <td> Phrase formed by “XP+DEG” </td> 
 <td> XP+DEG结构构成的短语 </td> 
 </tr> 
 <tr> 
 <td> DP </td> 
 <td> Determiner phrease </td> 
 <td> 限定词短语 </td> 
 </tr> 
 <tr> 
 <td> DVP </td> 
 <td> Phrase formed BY ‘’XP+DEB“ </td> 
 <td> XP+DEV结构构成的短语 </td> 
 </tr> 
 <tr> 
 <td> FRAG </td> 
 <td> fragment </td> 
 <td> 片段 </td> 
 </tr> 
 <tr> 
 <td> IP </td> 
 <td> InflectionPhrase </td> 
 <td> Simple clause headed by I（INFL或其他曲折成份） </td> 
 </tr> 
 <tr> 
 <td> LCP </td> 
 <td> Phrase formed by ”XP+LC“ </td> 
 <td> 处所词为中心语的短语 </td> 
 </tr> 
 <tr> 
 <td> LST </td> 
 <td> List marker </td> 
 <td> 用于解释说明性的列表标记短语 </td> 
 </tr> 
 <tr> 
 <td> NP </td> 
 <td> Noun phrase </td> 
 <td> 名词短语 </td> 
 </tr> 
 <tr> 
 <td> PP </td> 
 <td> Preposition phrase </td> 
 <td> 介词短语 </td> 
 </tr> 
 <tr> 
 <td> PRN </td> 
 <td> Parenthetical </td> 
 <td> 插入语 </td> 
 </tr> 
 <tr> 
 <td> QP </td> 
 <td> Quantifier phrase </td> 
 <td> 数词短语，由数量词构成的短语结构 </td> 
 </tr> 
 <tr> 
 <td> UCP </td> 
 <td> Unidentical coordination phrase </td> 
 <td> 非一致性并列短语 </td> 
 </tr> 
 <tr> 
 <td> VP </td> 
 <td> Verb phrase </td> 
 <td> 动词短语 </td> 
 </tr> 
 </tbody> 
</table>

## 动词复合6个标记 ##

VCD  并列动词复合 (VCD (VV 投资 ) (VV 办厂 ))  
VCP VV+VC 动词+是  
VNV A不A，A一A，(VNV(VV 能) (AD 不) (VV 能))  
VPT V的R，或V不R (VPT (VV 得) (AD 不) (VV 到))  
VRD 动词结果复合，第二个成份是第一个成份的结果(VRD (VV 呈现) (VV 出))；(VP(VRD(VV 联合) (VV 起来)))  
VSB 定语+核心复合，第一个成份为不及物动词，两个成份之间没有附加语或者体标记，VSB (VV 加速) (VV 建设)) (VP(VSB(VV 仰头)(VV 望去)))

### NP ###

中心词为名词构成的短语。从语法角度看，有两种含义：（1）按句法成份构成的短语，如组块在句子中充当主语、宾语等，可以增加辅助标签，NP-Sbg，NP-Obj；（2）知识库中的实体和属性，这种组块称为baseNP。

### VP ###

以动词为中心，与其修饰、限定、并列成份共同构成的一种语义组块。

## CoreNLP中源码 ##

nonTerminalInfo.put("ROOT",new String[][]{
 {left, "IP"}});
 nonTerminalInfo.put("PAIR",new String[][]{
 {left, "IP"}});
 
 // Major syntactic categories
 nonTerminalInfo.put("ADJP",new String[][]{
 {left, "JJ","ADJP"}}); // there is one ADJP unary rewrite to AD but otherwiseall have JJ or ADJP
 nonTerminalInfo.put("ADVP",new String[][]{
 {left, "AD","CS", "ADVP","JJ"}}); // CS is a subordinating conjunctor, and there are acouple of ADVP->JJ unary rewrites
 nonTerminalInfo.put("CLP",new String[][]{
 {right, "M","CLP"}});
 //nonTerminalInfo.put("CP", newString[][] {
 {left,"WHNP","IP","CP","VP"}}); // this iscomplicated; see bracketing guide p. 34. Actually, all WHNP are empty. IP/CP seems to be the best semantic head; syntax would dictate DEC/ADVP.Using IP/CP/VP/M is INCREDIBLY bad for Dep parser - lose 3% absolute.
 nonTerminalInfo.put("CP",new String[][]{
 {right, "DEC","WHNP", "WHPP"},rightExceptPunct}); // the (syntax-oriented) right-first head rule
 // nonTerminalInfo.put("CP", new String[][]{
 {right, "DEC","ADVP", "CP", "IP", "VP","M"}}); // the (syntax-oriented) right-first head rule
 nonTerminalInfo.put("DNP",new String[][]{
 {right, "DEG","DEC"}, rightExceptPunct});//according to tgrep2, first preparation, all DNPs have a DEG daughter
 nonTerminalInfo.put("DP",new String[][]{
 {left, "DT","DP"}}); // there's one instance of DP adjunction
 nonTerminalInfo.put("DVP",new String[][]{
 {right, "DEV","DEC"}}); // DVP always has DEV under it
 nonTerminalInfo.put("FRAG",new String[][]{
 {right, "VV","NN"}, rightExceptPunct});//FRAGseems only to be used for bits at the beginnings of articles:"Xinwenshe<DATE>" and "(wan)"
 nonTerminalInfo.put("INTJ",new String[][]{
 {right, "INTJ","IJ", "SP"}});
 nonTerminalInfo.put("IP",new String[][]{
 {left, "VP","IP"}, rightExceptPunct}); // CDM July 2010 following email from Pi-Chuanchanged preference to VP over IP: IP can be -SBJ, -OBJ, or -ADV, and shouldn'tbe head
 nonTerminalInfo.put("LCP",new String[][]{
 {right, "LC","LCP"}}); // there's a bit of LCP adjunction
 nonTerminalInfo.put("LST",new String[][]{
 {right, "CD","PU"}}); // covers all examples
 nonTerminalInfo.put("NP",new String[][]{
 {right, "NN","NR", "NT","NP", "PN","CP"}}); // Basic heads are NN/NR/NT/NP; PN is pronoun. Some NPs are nominalized relative clauseswithout overt nominal material; these are NP->CP unary rewrites. Finally, note that this doesn't give any specialtreatment of coordination.
 nonTerminalInfo.put("PP",new String[][]{
 {left, "P","PP"}}); // in the manual there's an example of VV heading PP butI couldn't find such an example with tgrep2
 // cdm 2006: PRN changed to not choose punctuation. Helped parsing (if not significantly)
 // nonTerminalInfo.put("PRN", new String[][]{
 {left,"PU"}}); //presumably left/right doesn't matter
 nonTerminalInfo.put("PRN",new String[][]{
 {left, "NP","VP", "IP","QP", "PP","ADJP", "CLP","LCP"}, {rightdis, "NN","NR", "NT","FW"}});
 // cdm 2006: QP: add OD -- occurs some;occasionally NP, NT, M; parsing performance no-op
 nonTerminalInfo.put("QP",new String[][]{
 {right, "QP","CLP", "CD","OD", "NP","NT", "M"}});//there's some QP adjunction
 // add OD?
 nonTerminalInfo.put("UCP",new String[][]{
 {left, }}); //an alternative would be"PU","CC"
 nonTerminalInfo.put("VP",new String[][]{
 {left, "VP","VCD", "VPT","VV", "VCP","VA", "VC","VE", "IP","VSB", "VCP","VRD", "VNV"},leftExceptPunct}); //note that ba and long bei introduce IP-OBJ smallclauses; short bei introduces VP
 // add BA, LB, as needed
 
 // verb compounds
 nonTerminalInfo.put("VCD",new String[][]{
 {left, "VCD","VV", "VA","VC", "VE"}});//could easily be right instead
 nonTerminalInfo.put("VCP",new String[][]{
 {left, "VCD","VV", "VA","VC", "VE"}});// notmuch info from documentation
 nonTerminalInfo.put("VRD",new String[][]{
 {left, "VCD","VRD", "VV","VA", "VC","VE"}}); // definitely left
 nonTerminalInfo.put("VSB",new String[][]{
 {right, "VCD","VSB", "VV","VA", "VC","VE"}}); // definitely right, though some examples lookquestionably classified (na2lai2 zhi1fu4)
 nonTerminalInfo.put("VNV",new String[][]{
 {left, "VV","VA", "VC","VE"}}); // left/right doesn't matter
 nonTerminalInfo.put("VPT",new String[][]{
 {left, "VV","VA", "VC","VE"}}); // activity verb is to the left
 
 // some POS tags apparently sit where phrases are supposed to be
 nonTerminalInfo.put("CD",new String[][]{
 {right, "CD"}});
 nonTerminalInfo.put("NN",new String[][]{
 {right, "NN"}});
 nonTerminalInfo.put("NR",new String[][]{
 {right, "NR"}});
 
 // I'm adding these POS tags to doprimitive morphology for character-level
 // parsing. It shouldn't affect anythingelse because heads of preterminals are not
 // generally queried - GMA
 nonTerminalInfo.put("VV",new String[][]{
 {left}});
 nonTerminalInfo.put("VA",new String[][]{
 {left}});
 nonTerminalInfo.put("VC",new String[][]{
 {left}});
 nonTerminalInfo.put("VE",new String[][]{
 {left}});
 
 // new for ctb6.
 nonTerminalInfo.put("FLR",new String[][]{rightExceptPunct});
 
 // new for CTB9
 nonTerminalInfo.put("DFL",new String[][]{rightExceptPunct});
 nonTerminalInfo.put("EMO",new String[][]{leftExceptPunct});//left/right doesn't matter
 nonTerminalInfo.put("INC",new String[][]{leftExceptPunct});
 nonTerminalInfo.put("INTJ",new String[][]{leftExceptPunct});
 nonTerminalInfo.put("OTH",new String[][]{leftExceptPunct});
 nonTerminalInfo.put("SKIP",new String[][]{leftExceptPunct});