关于Stanford Parser 4.0.0中NML标签的技术问询

Great question! Let’s break down everything you need to know about NML labels, including their purpose, how they work, and why you’re seeing different behavior with those two noun phrases.

What does NML stand for? #

NML stands for Nominal Modifier Labeled—it’s a special tag in the Stanford Parser used to mark noun-like phrases that act as modifiers to a core noun phrase (NP). Unlike a standard NP (which represents a core noun or noun phrase), NML explicitly signals that the phrase is functioning to describe or restrict another noun.

Why was NML introduced? #

The core motivation behind adding NML was to add finer granularity to syntactic parsing, especially for compound noun structures. Before NML, many noun modifiers were lumped into NP tags, which made it harder to distinguish between "core" noun phrases and the phrases that modify them. By introducing NML, the parser can clearly indicate which parts of a noun phrase are acting as modifiers, improving the precision of syntactic analysis for complex noun sequences.

From the ModCollinsHeadFinder documentation you referenced:

新增NML中心词规则，与NP规则相同。
NP中心词规则：NP与NML处理方式几乎一致（NP优先级更高）
NAC中心词规则：NML优先级高于NN/NNS，但低于NNP/NNPS

These rules confirm that NML is designed to align closely with NP but with strict priority hierarchies to ensure correct labeling.

Why "Income tax proposal" and "Fish tank water" get different treatment? #

Let’s break down the semantic structure of each phrase:

• "Income tax proposal": Here, "income tax" is a modifier that describes what the "proposal" is about. The core of the phrase is "proposal" (marked as NP), so "income tax" is tagged as NML to signal it’s a modifying nominal phrase. This fits perfectly with the parser’s priority rule—since NP takes precedence, the core noun gets the NP tag, and the modifier gets NML.
• "Fish tank water": This phrase’s core is "water", but "fish tank" is a fixed compound noun (a container for fish) that acts as a modifier. The parser likely tags "fish tank" as NP instead of NML for two reasons:
  1. Fixed compound noun status: "Fish tank" is a common, established compound noun, so the parser treats it as a core NP rather than a temporary modifying phrase.
  2. Priority rules: If a phrase can be interpreted as a core NP (even when used to modify another noun), the parser will prioritize NP over NML unless the context clearly signals a modifying role.

Is the Stanford Parser 4.0.0 correctly handling NML labels? #

Based on your example and the official documentation, yes—it’s working as intended. Your parsed sentence:

(ROOT (S (NP (DT The) (NML (NN income) (NN tax)) (NN proposal)) (VP (VBD was) (VP (VBN rejected))) (. .)))

Perfectly follows the rules: "income tax" is correctly marked as NML to show it modifies the core NP "proposal". The parser adheres to the priority rules outlined (NP > NML, NML > NN/NNS in NAC structures), so it’s correctly applying the NML tag where it makes syntactic and semantic sense.

If you encounter edge cases where the labeling seems off, double-check the semantic role of the phrase—NML is only meant for nominal phrases that act as modifiers, not core noun phrases.

内容的提问来源于stack exchange，提问作者Vineet Chaitanya