Fragments in Information Packaging

Invited talk at the  ESSLLI 2001 Workshop on Information Structure, Discourse Structure and Discourse Semantics

Papers on information structure often make use of question-answer pairs like (1) and (2) to illustrate the sentential partition between rheme (focus, new info) and ground (theme, presupposition, old info, topic):

(1) a. Who likes beer?
       b. [R JOHN] [G likes beer]
(2) a. What does John like?
       [G John likes ] [R BEER ]

The idea is that the question is the exponent of some underspecified description in a context or information state. In (1a), for instance, an instantiation for the experiencer parameter is needed. The rheme in (1b) is the structural element that "answers the question" in the sense that it expresses a value for this uninstantiated parameter, while the ground simply expresses the pieces of meaning in this description that are already available from context. Let us call the pieces of meaning that make a contribution to context the "update (potential)" and the pieces of meaning that are already available in context the "base".

The fact is often neglected that (1b) and (2b) are not discourse-natural answers to (1a) and (2a). In fact, the most natural answers to (1-2a) are those in (3) (and also those in (4), see below):

(3) a. JOHN (answer to (1a))
       b. BEER (answer to (2a))
(4) a. JOHN does. (answer to (1a))
       b. He likes BEER (answer to (2a))

A full account of information packaging should take this dialogical evidence into account. The difference between (1-2b) and (3) on the other is clearly not in their update potential. Still, we need to account for the fact that (3) is a more natural answer to (1-2a) than (1-2b).

The proposal defended here is that the difference between (1-2b) and (3) is in the nature of the base, namely, in the accessibility status of the base. In (1-2) the immediate, explicit questions make the base fully accessible at the time of utterance, so a ground is not needed. This follows from a view of grounds as context-anchoring elements. The ground is the structural expression of a base, but such expression is only needed when the anchoring of the update to a given base is not self-evident at the time of utterance. In question-answer exchanges, where the base is fully accessible from context, grounds are superfluous; this is why rheme-only fragment utterances, which express only the update, are so common as direct answers to questions. Of course, the ground may be redundantly used as an anchor in contexts where it is not needed, but this often gives rise to secondary interpretive effects, sometimes triggered by the violation of Gricean maxims.

What we need is a theory that allows us to establish some notion of accessibility for propositional descriptions in context, akin to the notion of accessibility assumed in accounts of anaphora resolution. Grounds are unnecessary/disfavored when the base is within some local attentional domain: the link between the update potential expressed as rheme and the base in the context proceeds as the result of certain default dynamics of dialogue. However, grounds become necessary when the base of a given update is outside this local attentional domain: the unification of update and base cannot take place without the explicit anchoring effected by the ground. In a sense, then, rheme-only fragments are to rheme-ground sentences what pronouns are to full definite descriptions in a theory of referential accessibility.

The answer sentences in (4) above behave more like sentence fragments than full answers with respect to information packaging (their dialogic distribution is closer to (3) than (1-2b)). Clearly, the examples in (4) are not syntactic fragments, but their behavior is accounted for if they are treated as informational fragments; in other words, the answers in (4) do not have a ground. The weak pronoun and the verbal elements that appear in (4) are there exclusively for morphosyntactic reasons and do not carry out the anchoring function that grounds carry out. This leads us to propose too that (5b) is as much of an informational fragment as (6b) is (although (5b) is obviously not a syntactic fragment):

(5) a. How does he feel about Bill?
       b. He LOVES him.
(6) a. Who does John love?
       b. BILL.

Both (5b) and (6b) are answers that do not need a ground to make the connection explicit between their update potential and the pertinent base in context.