Semantics usually addresses the literal meanings of individual clauses. When we start to string these clauses together into larger bodies, other issues come to the fore.
[ Up to Discourse and Pragmatics]
Pragmatics involves the ability to determine meanings implied from the social context in which a conversation takes place. We'll have more to say about that later, but for now, most pragmatics falls under the rubric of plan inference, or being able to figure out what goals the speaker is trying to attain by observing his or her behavior.
[ Up to Discourse and Pragmatics]
At the level of discourse, sentences are our building blocks, and the structures we use to build meaning become much less rigid.
There are hardly any formal constraints on which sentences appear in which order, except as determined by the needs of the speaker and his perception of the listener.
People start out talking about one subject, then digress, then return from the digression, then change the subject altogether, and so on. Somehow the hearer follows the focus of the conversation (usually) with little or no effort. This refers to discourse structure.
Here's an example of a discourse (taken from Allen):
In general, a discourse or conversation can be summarized thus:
H.P. Grice speaks of conversational implicature, which refers to a set of conventions which hearers expect speakers to adhere to.
As mentioned earlier, speakers perform various speech acts in an effort to accomplish his or her goals.
Some typical speech acts:
Within the context of a discourse or conversation, one of the discourse entities serves as the focus. This focus can shift in the course of the discourse:
Note that the focus in intimately related to our ability to figure out what pronouns like 'he' and 'it' mean. This is called anaphora resolution, discussed below.
We can cause confusion if we shift the focus abruptly (example from Allen):
Longer discourses can be divided into discourse segments.
Often they are marked by cue words and phrases like 'anyway,...' 'by the way...', etc.
Discourse segments are characterized by:
Often the relationships between segments can be characterized as a shifting attentional stack. The usual operations of push and pop apply to it. In the lawnmower example above, statements C-E, and H-K can be regarded as elaborations and digressions, respectively, which were pushed, then popped in turn.
[ Up to Discourse and Pragmatics]
Discourses are about specific things in the world of speaker and hearer. In general, we can use four ways of denoting entities:
Names
Definite descriptions ('the tall woman')
Indefinite descriptions ('a tall woman')
Pronouns
Expressions which make denote unique, specified things are called definite descriptions. Expressions which denote some member of a class of entities are indefinite descriptions. 'The' and 'a' are the definite and indefinite articles, respectively.
Descriptions can be existential, in that they introduce new discourse entities. some of referential in that they refer to objects already in the context. In:
In general, indefinite descriptions tend to be existential, while definite descriptions tend to be referential. Compare:
He owns a cat and feeds a cat.
He owns a cat and feeds the cat.
These expressions tend to be more powerful than pronouns, and can 'reach back' further to other parts of the discourse.
[ Up to Discourse and Pragmatics]
People use terms like 'it', 'he', 'her', and other pronouns which typically refer anaphorically back to entities introduced earlier in the discourse. Figuring out which entities are being referred to is called anaphora resolution.
Consider:
'He is making an anaphoric reference back to 'John'. 'John' is the antecedent. It is not hard to figure out what the antecedent is in this case, since it's the only discourse entity.
Now consider:
It is (somewhat)less clear whether it is referring to the bird or the plane. Only pragmatic constraints can resolve the ambiguity.
As out texts get larger, the number of antecedent candidates increases, and anaphor resolution becomes harder. We can go a long way by mainaining a history list which lists the discourse entities which have been mentioned in the discourse, in order of most recent reference. Then we can select the most recently mentioned entity that fits whatever constraints we can encode for the anaphor in question.
Suppose we continue the story:
We can resolve the reference by matching the features [gender = neuter] and [+ animate] [+ eats seeds] to each of {dog, delivery, Mary, New York, plane, bird, ....}.
Features that aid in anaphor resolution:
The first two have already been dealt with. The third is really a source of a richer set of constraints. Consider:
Suppose it was preceeded by:
Knowledge of the deathbed script (and Roman Catholicism) is critical to assigning the reference.
Discourses and conversations whose focus ranges over a number of events and topics have a more complex set of anaphoric references, and in fact the coherence of the discourse is marked by the scope of anaphoric references.
Recall that discourse can be modeled using an attentional stack. Anaphoric references generally do not apply to segments which have been popped.