In grammatical analysis, a Pattern can help identify structures that may be present. However, similarity in surface form is not sufficient to prove that two expressions have the same syntactic structure.
A Pattern initially provides only a Candidate Parse.
For example, given:
Until these questions are resolved, [B + C] remains only a candidate structure.
This is where Semantic Analysis becomes an important part of grammatical analysis.
1. Pattern Identifies Form; It Does Not Prove Structure
Suppose we already know a structure:
Two expressions may look similar on the surface while differing in:
Constituency
Syntactic Relations
Semantic Roles
- Semantic Relations
Therefore:
A Pattern is useful for generating possible analyses. It is not, by itself, proof of structure.
2. A Compiler Perspective: Parser Acceptance Is Not Enough
The distinction between structure and semantics is particularly clear in a Compiler.
Consider the hypothetical expression:
The Parser may therefore successfully construct an Abstract Syntax Tree :
At the syntactic level, the structure has been identified:
But the analysis is not finished.
Semantic Analysis and Type Checking must still determine whether these components satisfy the constraints imposed by the operator.
Suppose - is defined only for numbers:
Therefore:
3. Why Is This Comparison Useful for Natural Language?
Natural language is not a programming language, nor does it operate under a formally specified Type System.
Nevertheless, both fields share a methodological problem:
Given a linear sequence of units, which elements combine with one another, and what relations exist among them?
A natural-language sentence may permit several Candidate Parses:
The fact that Parse A matches a familiar Pattern establishes only that:
Parse A is worth considering.
It does not establish that:
Parse A is the structure of the sentence.
The analysis must continue through:
In particular, if a proposed segmentation loses or changes an important semantic relation in the sentence, that is a reason to re-examine the analysis.
4. An Example: 我送他回国
Consider:
On the surface, we can identify the sequence:
If we rely only on Pattern, one possible Candidate Parse is:
At the level of surface form, this analysis may initially appear plausible.
Semantic Relations, however, reveal an additional problem.
5. “Whom Does 我 送?” and “Who Returns Home?”
We can test two relations separately.
送谁？
The first relation is:
Now consider the second relation.
谁回国？
This gives:
They can be visualized as:
The same element, 他, participates in two relations.
6. 兼语: One Element Participating in Two Relations
In traditional Chinese grammatical analysis, this type of construction is associated with 兼语 .
A simplified representation is:
The crucial element is 他:
A more explicit representation is:
他 participates in a relation with 送, while simultaneously serving as the semantic subject associated with 回国.
This dual relation is what the traditional analysis of 兼语 attempts to capture.
The important point here is not that the term 兼语 must necessarily be adopted.
The methodological point is:
A syntactic analysis should preserve and explain the important relations actually present in the sentence.
7. Why Is [他回国] Not Enough to Explain the Sentence?
In:
This illustrates an important principle:
A sequence may form a constituent in environment A without necessarily having the same structure in environment B.
8. Semantic Relations Can Constrain a Candidate Parse
The analytical process can be represented as:
Suppose a Candidate Parse proposes:
or shows that an element inside [A + B] maintains an important direct dependency with something outside the proposed constituent.
We then have reason to reconsider:
- the constituent boundary;
- the type of structure;
- the dependency relations;
- or the analytical model itself.
Semantics does not replace Syntax.
Rather:
Semantic Relations constrain competing Candidate Parses and provide evidence for evaluating them.
9. Semantic Constraints and Type Constraints
At this point, another comparison with computer science becomes useful.
In a typed programming language:
重要 can evaluate an event or proposition.
By contrast:
好听 normally establishes a semantic relation with something that can be perceived auditorily.
A Predicate cannot combine entirely arbitrarily with every possible Semantic Argument.
If an analysis proposes:
This is one of the most important limitations of the compiler analogy.
11. Semantics Does Not Replace Syntax
If Pattern is insufficient, this does not mean that Syntax can be discarded and structure inferred entirely from meaning.
The fact that a sentence is understandable does not prove that every proposed hierarchical analysis is correct.
Given an interpretation:
without structural evidence.
A complete analysis must consider:
The more appropriate relationship is therefore:
A complete analysis must account for both.
If a Parse explains only:
13. Pattern Is Still Important
Saying that Pattern is insufficient does not mean that Pattern is useless.
Patterns help us:
- identify possible structures;
- locate similar linguistic data;
- generate Candidate Parses;
- formulate grammatical rules;
- detect unusual forms;
- compare distributions in a Corpus.
Grammar Rules are equally important in a Compiler.
Without Grammar, a Parser would not know which structures it can construct.
But:
is not the entire compiler pipeline.
Likewise, in natural-language analysis:
is only one stage.
The methodological error occurs when:
14. From Pattern Matching to Grammatical Analysis
The overall process can be summarized as:
This does not mean that natural-language analysis literally follows these steps in a rigid sequence.
It is a methodological model, not a formal algorithm for natural language.
Its purpose is to prevent an overly rapid inference:
A more defensible inference is:
15. Three Distinctions That Must Be Preserved
The entire problem can be condensed into three propositions.
Pattern ≠ Structure
Natural language does not have a rigid pipeline that corresponds exactly to this architecture.
Nevertheless, the layered mode of reasoning remains useful:
The common point is not:
Natural language works like a programming language.
It is:
A formal structure must still be tested against the relations it claims to represent.
A compiler makes this principle particularly visible. An AST may be syntactically constructible while its operands fail the semantic or type constraints imposed by an operator.
Natural language is much more flexible, but the methodological question remains:
17. Conclusion
Pattern is an important tool in grammatical analysis. It helps identify potentially relevant structures and generate Candidate Parses for further examination.
But Pattern does not prove Structure.
Structure, in turn, cannot be evaluated entirely independently of Semantic Relations.
Consider again:
reveals that 他 simultaneously participates in relations with both 送 and 回国.
This relationship is central to the analysis of 兼语 and demonstrates why surface Pattern cannot replace relational analysis.
A more complete analytical procedure is:
Therefore:
The next article will apply this framework directly to 他学习很好 and 他唱歌很好听.
The question will no longer be merely whether these expressions occur or are understandable. Instead, it will examine whether 学习 and 唱歌 have the same syntactic properties, what 很好 and 很好听 are actually predicated of, and whether similar surface Patterns provide sufficient evidence for assigning the two sentences the same grammatical analysis.