Main constituent order

Given the Penthouse Principle, a language cannot acquire a different branching direction in lower-level constructions by simply inverting the order of their components, e.g. by shifting suffixes into a prefix position. The way for a language to change its branching pattern – also called its serialization direction – is to start from top, where order is free. Once main constituent order has been changed, lower level constructions which originate by grammaticalization of higher-level constructions inherit the new order, and gradually the new serialization appears at lower levels.

A historical example is provided by main constituent order in the Germanic and Romance languages. The order inherited from Proto-Indo-European was rather free, with a preference for verb-final order. In accordance with the Penthouse Principle, it was freer in independent than in subordinate clauses. illustrates the Old English main constituent order in an independent declarative clause; illustrates it in a subordinate clause.

In both examples, the auxiliary hæfdon is the finite verb. It occupies the second position in the independent clause of and final position in the subordinate clause of . At some Proto-Germanic stage, the language had changed the inherited main constituent order in main clauses to verb-second order, whereas it had preserved the inherited order at the lower level. It could not be the other way around.

On its way to modern English, the language also changed main constituent order in subordinate clauses. The translation of the two examples shows verb-second order not only in the independent, but also in the dependent clause. Two pathways for this change are conceivable: Either subordinate clauses were remade by recruiting earlier main clauses into subordinate function; or subordinate clauses changed their word-order pattern in analogy with main clauses.

Possessive noun phrases and adpositions

There is an implicational generalization to the effect that if a language has postpositions, its possessive attribute has prenominal position; and if it has prepositions, its possessive attribute has postnominal position. Since we have seen postpositions before, here are examples of the latter kind.

The basis of this typological correlation is, again, grammaticalization. As we have seen, relational nouns in possessive constructions get grammaticalized to adpositions with their complement. Grammaticalization shifts a construction down the complexity levels. It does not, however, interfere with the internal word and formative order of the constructions. Therefore, if the underlying relational noun governs its nominal complement to the left, so does the resulting adposition; and analogously for government to the right.

This also prescribes the ways that a language can change the order of its adpositions. Given the low grammatical level of the adpositional phrase, the adposition cannot simply hop over its nominal complement. Instead, a new set of adpositions must be grammaticalized. For instance, Pre-Proto-Indo-European must have had postpositions, because this is the only possible source of its case suffix morphology. Some Indo-European languages, including Proto-Germanic, abandoned this construction. They introduced right-branching order at higher syntactic levels, including a postnominal possessive attribute. Languages like English then grammaticalize the relational noun in this latter construction to a case relator. Which produces the prepositions which the language has today.