On Markedness Asymmetries in Person and Number

1. Introduction

Extensive typological research on systems of pronouns and agreement has led to a number of important discoveries about the representation of morphosyntactic features. For example, Harley and Ritter (2002; H&R) propose a universal geometry of person, number, and gender features (1), which captures a wide array of pronominal systems in the languages of the world (p. 486). H&R argue against the more traditional approach of using unstructured binary features to represent person and number categories, maintaining that such approaches can only stipulate certain implicational universals noted by Greenberg (1963).¹ H&R propose instead that pronominal categories are represented by a hierarchical organization of privative features (see also Bonet 1991, Béjar 2003, Nevins 2003). Under this approach, implicational universals can be encoded in terms of dependency relations.

(1) 

H&R’s analysis represents an important step toward capturing linguistic universals, and I assume that their proposed geometry is essentially correct. However, I argue that it needs to be modified in order to capture certain crosslinguistic markedness asymmetries in person and number.

H&R treat certain person and number categories as conjunctively specified. For example, if a language has a special category for inclusive person, this category has a Participant node with two dependent person features, [Speaker] and [Addressee] (p. 490).

(2) 

Likewise, if a language has a special category for dual number, the Individuation node of this category has two number features—[Minimal], also used for singular number, and [Group], also used for plural (p. 492).²

(3) 

This analysis of the dual predicts Greenberg’s (1963:94) observation that a language with a dual number category will also have a plural, on the straightforward assumption that features that occur in conjunction can also occur alone (H&R, p. 509). For example, if [Minimal] and [Group] can combine to form the dual, they can also occur separately to form the singular and the plural.

The [Minimal] and [Group] features have the same hierarchical position in H&R’s geometry: both are dependents of the Individuation node. Likewise, both [Speaker] and [Addressee] are dependents of the Participant node. If nothing more is said, then, the plural and first person representations in 4a and 4b below are structurally equivalent to their singular and second person counterparts in 5a and 5b.

(4)

a.

b.

(5)

a.

b.

Despite this structural equivalence, there appears to be an important asymmetry between the representations in 4 and those in 5. This asymmetry emerges in languages lacking conjunctively specified representations.

In a language with no dual category, the dual is conflated with the plural, not with the singular. For example, English makes no systematic distinction between dual and plural (Table 1, row a). In this case, H&R assume that dual and plural number are represented with the feature [Group], as in 4a (p. 489). However, if [Group] and [Minimal] are equivalent, languages should equally allow all minimal sets—sets of one or two—to be represented as [Minimal], as in Table 1, row b. None of the systems described in Corbett’s (2000) survey of about 250 number systems conflates singular and dual into a single category. Thus, it appears that [Group] and [Minimal] are not completely equivalent.

Table 1.

Number contrasts predicted if [Group] and [Minimal] are equivalent.

Similarly, in a language without an inclusive category, the inclusive is conflated with first person, not with second. For example, English lacks a distinction between inclusive and exclusive first person (Table 2, row a). Again, if [Speaker] and [Addressee] are equivalent, it should be equally possible to conflate inclusive with second person, as in Table 2, row b. In fact, however, Table 2, row b is unattested (Zwicky 1977, Noyer 1992). It appears that, like [Group] and [Minimal], [Speaker] and [Addressee] are not completely equivalent.

Table 2.

Participant contrasts predicted if [Speaker] and [Addressee] are equivalent.

It is important here to distinguish between syncretism and what I call conflation. Syncretism arises when a distinction between two syntactic representations is neutralized...