Institut de Lingüística Aplicada

Third Mediterranean Meeting on Morphology (MMM3)


Greville Corbett (University of Surrey)
Agreement: Canonical instances and the contribution of inflectional morphology

Agreement is of interest not just within core linguistics but also for language acquisition, psycholinguistics and computational applications. Yet interdisciplinary work in this area is being hampered by muddled terminology, and important choices in analysis are sometimes made as much by tradition as by argument. This paper therefore has two goals. First, it attempts to draw out the common ground in the various views (sometimes conflicting) which linguists have adopted on the nature of agreement. We start from central (canonical) instances of agreement, characterized by several overlapping criteria. Various weakenings of these criteria result in more or fewer phenomena falling within the range of 'agreement', and show how the different views are related. I argue that being clear about the relatedness of these phenomena is more important that defining a precise line between agreement and non-agreement. The second goal is to show the centrality of inflectional morphology in attempting a coherent characterization of agreement. In particular, we shall see that certain views on what should be included as agreement reduce to the (often unstated) assumption that the 'best' agreement is expressed by regular inflectional morphology.

We call the element which determines the agreement (say the subject noun phrase) the controller. The element whose form is determined by agreement is the target. The syntactic environment in which agreement occurs is the domain of agreement. And when we indicate in what respect there is agreement, we are referring to agreement features. Factors which determine agreement, while not themselves marking agreement, are conditions. We define 'agreement' broadly, to include 'concord' (since this is defined in confusingly different ways). We begin with an instance of canonical agreement from Spanish: un cuadro hermos-o 'a beautiful picture', cf.: un-a cortina hermos-a 'a beautiful curtain'. We then consider various extensions, allowing in less canonical phenomena, starting from the five elements of our definition. In each one, we consider characteristics which are more and less canonical (indicated: more > less), with examples from a wide range of languages. Note that 'canonical' means conceptually central, irrespective of how frequently the type is attested.

In terms of controllers, we have: controllers which are present > absent (Russian: ja idu > SCB: idem 'I read'; those with overt features > covert features (French elle est contente > je suis content(e)). For targets the following are canonical: inflectional marking > other means of marking; obligatory > optional; morphologically bound > free; doubling > independent; immobile > mobile; regular > suppletive; alliterative > opaque; productive > sporadic; having one controller > having more than one controller. The terms 'controller' and 'target' suggest that agreement in an asymmetric relation. We treat this as a canonical property, allowing for arguably symmetric cases as less canonical. In terms of domains: we have: local > non-local, and: being one of a set of domains > single domain. Features provide the characteristics: lexical > non-lexical; matching > non-matching. Finally, for conditions, there is a simple characteristic, namely: no conditions > conditions.

These various claims can be collapsed into three metaprinciples:

  1. the closer the marking is to canonical inflectional morphology, the closer to canonical agreement.
  2. the more redundant (less informative) agreement is, the more canonical it is.
  3. the greater the reliance on formal properties the more canonical.

These conclusions are somewhat surprising. We may distinguish contextual inflection (which includes agreement) from inherent inflection, with contextual being 'the prototypical case of inflection' (Booij 1996: 14), see (1). And yet, the more canonical the instance of agreement, the more redundant the information conveyed (2), and the more 'formal' ('less semantic') its nature (3). These properties of canonical agreement help explain why the canonical type is rare among the languages of the world.