Person and Number in Pronouns: A Feature-Geometric Analysis

The set of person and number features necessary to characterize the pronominal paradigms of the world’s languages is highly constrained, and their interaction is demonstrably systematic. We develop a geometric representation of morphosyntactic features which provides a principled explanation for the observed restrictions on these paradigms. The organization of this geometry represents the grammaticalization of fundamental cognitive categories, such as reference, plurality, and taxonomy. We motivate the geometry through the analysis of pronoun paradigms in a broad range of genetically distinct languages.*

Introduction

It is generally accepted that syntactic and phonological representations are formal in nature and highly structured. Morphology, however, is often seen as a gray area in which amorphous bundles of features connect phonology with syntax via a series of ad hoc correspondence rules. Yet it is clear from the pronoun and agreement paradigms of the world’s languages that Universal Grammar provides a highly constrained set of morphological features, and moreover that these features are systematically and hierarchically organized.¹ In this article we develop a structured representation of person and number features intended to predict the range and types of interactions among them. More specifically, we will motivate the claims in 1.

(1) Claims

a. The language faculty represents pronominal elements with a geometry of morphological features.

b. The organization of this geometry is constrained and motivated by conceptual considerations.

c. Crosslinguistic variation and paradigm-internal gaps and syncretisms are constrained by the hierarchical organization of features in the universal geometry.

d. The interpretation of subtrees of the geometry may be relativized in tightly constrained ways so that language-specific interpretation of a given feature will depend in part upon the contrasts available within that language.

In order to substantiate these claims we demonstrate that a geometric analysis can shed light on children’s acquisition of pronouns and on paradigms that manifest a range of unusual properties. In the course of this demonstration we draw on our own database of 110 languages, as well as descriptions of other exceptional languages discussed in the literature.

1. Feature geometries

Linguists agree that there are natural classes of morphological features. This is reflected in the universal classificatory use of the terms person, number, a nd gender, as well as other classes of features not discussed here. At least this much organization is unconditionally assumed, and further organized relationships among these classes of features have often been noted, although not often treated theoretically. For instance, Greenberg (1963) observed a number of crosslinguistic generalizations about the clustering of features, describing, for example, the dependence of gender on number (Universal 32: ‘Whenever a verb agrees with a nominal subject or object in gender it also agrees in number.’) and the dependence of dual number on plural number (Universal 34: ‘No language has a dual [number] unless it has a plural.’ Greenberg 1963:94).

Although these dependencies must be the consequence of some aspect of universal grammar, morphological theory has not in general attempted to provide an account of them. It seems to be widely assumed that features are collected in unstructured bundles and that morphological rules may freely refer to any feature or group of features, regardless of whether or not they form a natural class.

1.1. The problem with unstructured bundles of features

Many modern theories of morphology claim that morphosyntactic features are grouped in unstructured bundles and even make a virtue of it—despite the fact that a certain amount of organization is implicit in the very terminology common to all morphological theories. Consider, for example, the discussion of morphosyntactic representations (MSRs) in Anderson (1992:92): ‘The minimal (and thus the most desirable) theory of MSRs . . . is one that would assign them no internal structure at all.’ Yet even given this explicit assertion, Anderson does not assume that MSRs representing agreement features are entirely unstructured. He implicitly assigns attribute-value structure to them: There is at least a person attribute that may bear the values {[±me], [±you]}; a number attribute that may bear the values {[±pl]}; and a gender attribute that may bear a series of familiar values...