Separate off the NML part of the nn relationship so that we can do it without the comma restriction. This especially finds lists of NML as nn modifiers of a noun

Author: AngledLuffa

src/edu/stanford/nlp/trees/EnglishGrammaticalRelations.java

@@ -1018,7 +1018,10 @@ private EnglishGrammaticalRelations() {}
   public static final GrammaticalRelation NOUN_COMPOUND_MODIFIER =
     new GrammaticalRelation(Language.English, "nn", "nn modifier",
         MODIFIER, "(?:WH)?(?:NP|NX|NAC|NML|ADVP|ADJP)(?:-TMP|-ADV)?", tregexCompiler,
-            "/^(?:WH)?(?:NP|NX|NAC|NML)(?:-TMP|-ADV)?$/ < (NP|NML|NN|NNS|NNP|NNPS|FW|AFX=target $++ NN|NNS|NNP|NNPS|FW|CD=sister !<<- POS !<<- (VBZ < /^[\'’]s$/) !$- /^,$/ !$++ (POS $++ =sister))",
+            "/^(?:WH)?(?:NP|NX|NAC|NML)(?:-TMP|-ADV)?$/ < (NP|NN|NNS|NNP|NNPS|FW|AFX=target $++ NN|NNS|NNP|NNPS|FW|CD=sister !<<- POS !<<- (VBZ < /^[\'’]s$/) !$- /^,$/ !$++ (POS $++ =sister))",
+            // same thing as the above, but without the comma.  NML in such a situation is typically a noun phrase modifying a noun,
+            // whereas other nodes such as NN can be parts of lists or otherwise unsuitable for the nn relationship
+            "/^(?:WH)?(?:NP|NX|NAC|NML)(?:-TMP|-ADV)?$/ < (NML=target $++ NN|NNS|NNP|NNPS|FW|CD=sister !<<- POS !<<- (VBZ < /^[\'’]s$/) !$++ (POS $++ =sister))",
             "/^(?:WH)?(?:NP|NX|NAC|NML)(?:-TMP|-ADV)?$/ < JJ|JJR|JJS=sister < (NP|NML|NN|NNS|NNP|NNPS|FW=target !<<- POS !<<- (VBZ < /^[\'’]s$/) $+ =sister) <# NN|NNS|NNP|NNPS !<<- POS !<<- (VBZ < /^[\'’]s$/) ",
             // in vitro, in vivo, etc., in Genia
             // matches against "etc etc"