Recursion Restrictions: Where Grammars Count

Abstract

Since recursion is a fundamental property of human languages, it is puzzling that we regularly find cases where recursion is impossible or restricted. In this paper we argue that these restrictions follow from an independently necessary property associated with individual lexical items, which encodes sensitivity to phonological properties. These restrictions must be stated on the output of the syntactic derivation, when syntactic structures are transferred to phonology as expected in current late spell-out models. The main idea is that phonological properties can be “grafts” on the structure-building requirement of a lexical item, referred to as an epp property, which can then be viewed as a repository of the finely grained knowledge speakers have of the phonological properties associated with local syntactic environments. In this view, restrictions on recursion, though accidental, can be straightforwardly and simply accounted for as arising from the way that independently necessary properties interact in specific local syntactic environments. This accounts for a number of well-known effects, including left branch restrictions, restrictions on center embedding, and complexity effects.

This paper builds on my presentation at the Recursion conference in Amherst, May 2009, and is a substantial revision of my 2010 paper On restricting recursion and the form of syntactic representation lingBuzz/001139. I would like to thank Natasha Abner, Melanie Bervoets, Chris Collins, Uli Sauerland, Ed Stabler, Dominique Sportiche and Anna Szabolcsi, for feedback and comments.

Appendix A: A Possible Linearization Account for the Recursion of Infinitives?

We add here a short discussion of Norvin Richards (2006) linearization account. Richards Proposes a general constraint on linearization, which imposes distinctness of linearization of categories within the same phase, a kind of modern incarnation of the OCP, and applies it to doubl-ing. Adapting parts of Kayne’s (1994) LCA, his proposal in essence prohibits two categories within the same phase from being spelled out, because they would yield a conflicting linearization statement. He applies this to a number of different phenomena (distinct markings on DPs, the doubl-ing filter (Ross 1972, but see Koopman 2002 for an explicit proposal along the lines in this paper). Applied to Dutch, this proposal would prohibit the linearization of two infinitives within the same phase, but not across a phase boundary. It is easy to adapt Richards (2006) to Dutch, given the syntactic analysis developed in Koopman and Szabolcsi (2000) and outlined here. Indeed, inverted infinitives end up occurring in the same V+ constituent (hence within the same phase), but infinitives in the VT1 Vinf2 Vinf3 order are each in CPs, hence separated by a phase boundary [_C V_inf [_C V_inf.. This would allow some understanding as to why Inf is subject to the restriction on recursion but other categories are not, which under the account in this paper is purely a historical accident. However, the proposal does not extend (easily) to any of the other observed and excluded patterns. It will not extend to exclude the 3 1 2 pattern with infinitives in Dutch (*zwemmen wil kunnen), as here the infinitives do not seem to be in the same phase as the selecting infinitive. Nor will it allow a simple understanding of variation, i.e. what happens with Dutch dialects or speakers that allow such patterns, or with German and Hungarian speakers? How do we let the right cases in? If Richards needs Koopman and Szabolcsi (2000) to understand Dutch, the problem then is how to allow German and Hungarian which allow surface constituents of this type. Richards proposal could force a double structural analysis where preverbal infinitives in German and Hungarian are or can be in their own CPs or within smaller constituents, but this would lose all the basic insights of Koopman and Szabolcsi’s analysis (see Koopman and Szabolcsi for an argument against this option), and any hope to model the observed variability between individual grammars. This analysis will have nothing to say about the Hungarian cases where it seems you can invert twice but not three times, and where the restriction on recursion does not seem sensitive to a particular category, just to dept of embedding in a specific location. And finally, it will have nothing to say about other cases of restrictions on recursion, and the general notion of [+ph] size that is independently needed, and that seems widely applicable.