Abstract
Every once in a while, a theory needs to reevaluate its basic concepts. This reevaluation is necessary because, over time, as more empirical data is covered, one tends to lose sight of the original intent. In the case of LFG, one of the most basic concepts is the grammatical function. LFG differs from almost every other generative theory in recognizing the centrality of grammatical functions. Other theoretical approaches recognize grammatical relations, but not grammatical functions. The concept of grammatical relation is vague; it just means some relation that is relevant to the grammar (or more precisely, the syntax). A grammatical function is something much more specific--it is a link between structure and function. We can identify grammatical relations by their (apparently arbitrary) properties; with grammatical functions, the properties are the result if we correctly understand the functions. Grammatical relations have no role in explanation; grammatical functions should be the centerpiece of explanation.
One of the most problematic grammatical functions, perhaps the most problematic, is the function subj. The usual LFG view is that subj is an argument function, specifically, the most prominent argument function on the relational hierarchy. We will call this function ̂gf, parallel to the LFG notation θ̂ for thematically most prominent argument. This accounts for “argumenthood” properties of subjects, the ones typically associated with agentive arguments even in syntactically ergative and Philippine-type languages. But the more interesting properties of subjects are the ones which are not related to argumenthood, and which in these non-nominative-accusative languages are not associated with the agent. We propose that these properties are associated with a grammatical function which we call piv (pivot). The function of the piv function is primarily cross-clausal continuity, a kind of sentence-internal topic. This function is not an argument function, but rather an overlay (or “grammaticized discourse”) function. We show that such a theory, by relating formal properties to function, provides explanations for the properties of subjects.