SEMGREX规则优化求助：实现多AMOD成分匹配同一adj标签

Great question! The issue with your original rule is that ?>amod only matches zero or one amod dependency arc, which is why it’s only picking up one of your modifiers. To capture multiple amod modifiers (like both distinct and electromagnetic for field), we need to adjust the pattern to allow repeated amod arcs.

Modified SEMGREX Rule #

Here’s the updated rule that will capture all direct amod modifiers of your target noun:

((( {pos:/NN.*/}=target ?>compound {pos:/NN.*/}=compl ) (amod {pos:/JJ.*|VB.*/}=adj)* ) ?>/nmod.*/ {pos:/NN.*/}=adjn ) >det {word:/a|an/}=art

Key Changes Explained #

• Replaced ?>amod {pos:/JJ.*|VB.*/}=adj with (amod {pos:/JJ.*|VB.*/}=adj)*:
  • The parentheses () group the amod arc and its node pattern together, so the quantifier applies to the entire modifier pattern.
  • The * quantifier means "zero or more instances"—this preserves the optional behavior of the original rule but allows multiple amod modifiers to be matched.
  • SEMGREX automatically collects all matching modifier nodes under the adj label when the pattern repeats, so both distinct and electromagnetic will be included here.

Handling Indirect Modifiers (If Needed) #

If your dependency parse has modifiers in a chain (e.g., distinct modifies electromagnetic, which in turn modifies field via a compound arc), you can expand the rule to capture those too by including compound arcs in the modifier pattern:

((( {pos:/NN.*/}=target ?>compound {pos:/NN.*/}=compl ) ((amod|compound) {pos:/JJ.*|NN.*|VB.*/}=adj)* ) ?>/nmod.*/ {pos:/NN.*/}=adjn ) >det {word:/a|an/}=art

• Added |compound to the arc pattern to include compound modifiers alongside amods.
• Updated the POS regex to /JJ.*|NN.*|VB.*/ to cover nouns used as modifiers (in case electromagnetic is tagged as NN instead of JJ).

Testing the Rule #

For your sentence fragment a distinct electromagnetic field, this rule will:

• Set target to field
• Capture both distinct and electromagnetic under the adj label
• Keep matching the other components (art as a, etc.) as before

内容的提问来源于stack exchange，提问作者Oleg