Treatment of recursion in linguistic description
Recursion—more precisely the phenomenon that this term is usually taken to refer to—is not an unproblematical notion. Chomsky himself, who has popularized its use in linguistics (Chomsky 1957), observes that ‘‘[T]he possibility that languages are nonrecursive, however, is granted by everyone who has seriously discussed the subject, and the question whether this possibility is realized remains an open one’’, and he concludes that ‘‘while languages may be recursive, there is no reason to suppose that this must be so’’ (Chomsky 1980: 120, 122).
As we noted above, recursion is not a phenomenon of language but rather of a theory that postulates it as a useful device to describe or explain certain properties of language structure. Accordingly, the way it has been treated in linguistics differs greatly from one model to another. First, there are models where its relevance is ignored, or denied. An extreme position is maintained in the paratactic theory of Davidson (2004), where there is norecursion and no embedding; rather, propositions are understood to be paratactically ordered. Accordingly, the English sentence Pierre believes that snow is white is said to consist of two distinct utterances linked by parataxis: Pierre believes that. Snow is white.
Second, quite a number of linguists acknowledge that there are phenomena that can be described in terms of recursion but that there are also alternative ways of describing such phenomena. And third, there are in fact alternative models that have been proposed to deal with recursive phenomena, perhaps the most noteworthy one being that of endocentricity as used in the early structuralist tradition, which is based on observations of the distributional equivalence of constituents. According to Bloomfield (1933: 194–7), a construction (e.g. poor John) is defined as endocentric if it
Belongs to the same form-class as one (or more) of its constituents. The latter constituent is called the head (John) while the other constituent is the attribute (poor). Endocentric constructions contrast with exocentric constructions (e.g. John ran), which belong to a form-class other than that of any of its constituents. Bloomfield observes that most constructions in any language are endocentric, while exocentric constructions are few. Endocentricity is productive (‘‘there can be several ranks of subordinative position’’ in his terminology) in that an endocentric construction ([fresh] milk) can be the head of another endocentric construction ([[very] fresh] milk).
The difference between embedding and iterating recursion (see below) is reflected in Bloomfield’s distinction between two kinds of endocentric constructions, namely subordinative or attributive (poor John) and coordinative or serial constructions (boys and girls), respectively. While Bloomfield does not extend the notion of endocentricity to the relation between main and subordinate clauses,1 later authors did, treating clause subordination in terms of endocentricity. Thus, a noun phrase containing an embedded relative clause, as in the man who came to tea, is interpreted as endocentric since it has the same distribution as the noun phrase the man (Lyons 1977: 391), and Lehmann (1988: 182) proposes to define clause subordination more generally as an endocentric construction between two clauses where one is the head and the other its dependent. In most formal models, recursion is simply a formal means of capturing (one aspect of) endocentricity (Fritz Newmeyer, p.c.).
1 Bloomfield (1933:194) notes that subordinate clauses are exocentric since the resultant phrase (e.g. if John ran away) has the function of neither of its constituents (if or John ran away). But he does not elaborate on how subordinate clauses relate to main clauses.
