The labeling algorithm determines the category of a newly merged object when merge combines two elements. The label is typically inherited from the most prominent element (highest head or structurally dominant element). This algorithm eliminates the need for pre-specified phrase structure rules and explains how categories emerge from the primitive merge operation, unifying syntax generation across languages.
Apply the labeling algorithm to simple merges (head-complement, specifier-head) and examine how different labeling outcomes affect structural properties. Explore edge cases where labeling is ambiguous or fails.
You already know that Merge is the primitive structure-building operation in Minimalist syntax: it combines two syntactic objects into a new set {X, Y}. But Merge by itself creates only unlabeled sets — it does not specify what kind of thing the resulting object is. The question the labeling algorithm answers is: when we merge two elements, what is the syntactic category of the resulting structure? Is it a verb phrase, a noun phrase, a clause? The answer matters because the rest of the derivation — movement, agreement, interpretation — depends on knowing what category each syntactic object belongs to.
The simplest case gives the clearest intuition. When Merge combines a head (like a verb *V*) with its complement (say, a noun phrase *NP*), the result is a syntactic object dominated by the head: it inherits the category *V* and projects as a verb phrase. This is why we call it "VP" — the verb is the most prominent element, in the technical sense that it has features other elements are looking for (the verb has tense features, case-assigning properties, etc.). The labeling algorithm formalizes this prominence: the label of {X, Y} is the label of whichever element makes its features "visible" to the rest of the structure. In straightforward head-complement structures, this is always the head, which is why heads project and complements do not.
The more revealing cases arise when you merge a specifier with a phrase. In the structure {DP, VP} — a subject merging with a verb phrase — neither element is the head of the other. Chomsky's proposal is that in such cases, labeling succeeds through feature sharing: if the DP and the head of the VP share a feature (like φ-features in agreement), that shared feature can serve as the label. This is why specifier-head agreement is not just a surface phenomenon — it is the mechanism by which specifier-phrase structures become properly labeled and interpretable. Without agreement, the structure would be unlabeled and the derivation would crash.
The labeling algorithm eliminates something older generative frameworks took for granted: phrase structure rules. In classical Transformational Grammar and early Government-Binding theory, you wrote rules like VP → V NP, NP → Det N, etc., stipulating that verbs take noun-phrase complements and that noun phrases consist of determiners and nouns. These rules worked but were not explanatory — they described the patterns without deriving them from anything deeper. The labeling algorithm, combined with Merge, derives these patterns: we do not need to be told that a verb plus its complement forms a verb phrase, because the algorithm computes the label from structural and feature-based properties. This is what Minimalism means by eliminating stipulations: every property of phrase structure that can be derived from simpler mechanisms should be, and the labeling algorithm is the mechanism that handles category projection. The connection to phases matters here too — labeling interacts with phase theory because the algorithm applies at the phase level, and its failure at phase edges creates the conditions for movement operations that rescue unlabeled structures.