Rhetoric and Optimality Theory

by Helen de Hoop

The circumstances under which Parmenion was killed, are well-known. An envoy was sent out to offer Parmenion two letters. The first one was from Alexander the Great himself. The second one was sealed as if from Parmenion's condemned son. He began to read the letter with pleasure, when they stabbed him (1).

Consider the last sentence of the previous paragraph again. Note that the sentence consists of a main clause and a subordinate when-clause. The interpretation obtained is that there is a temporal overlap between the two events described by these two clauses. Crucially, there is no causal relation between the two events. Reading the letter with pleasure does not cause the stabbing, nor the other way around. When we would prepose the when-clause, however, we would get such a causal reading: When they stabbed him, he began to read the letter with pleasure (2). This interpretation (the stabbing causes his reading the letter with pleasure) strikes us as infelicitous in the context of our present world. We can also establish a reversed causal relation by reversing the events in the when-clause and the main clause: When he began to read the letter with pleasure, they stabbed him (3). In sentence (3), the reading the letter with pleasure causes the stabbing. Strikingly, when we change the order of the two clauses, we get two possible interpretations. Either there is no causal relation, only temporal overlap, just like the reading we got for sentence (1) above, or the causal relation reading as established in sentence (3) is maintained: They stabbed him, when he began to read the letter with pleasure (4). How come?
The central hypothesis of this talk is that optimization plays a crucial role in natural language interpretation. In order to communicate their speculative thoughts, speakers effectively use their implicit knowledge of how hearers arrive at the optimal interpretation of a speaker's utterance in a certain context. The optimal interpretation of an expression is computed by evaluating candidate interpretations against a set of potentially conflicting constraints.
In principle, when a main clause and a when-clause denote two events, we never get a rhetorical relation between the two events such that the event described by the main clause is the antecedent of the event described by the when-clause. This is independent of the position of the when-clause, whether it is in preposed or postponed position. On the other hand, both preposed and postponed when-clauses may express a rhetorical relation such that the when-clause event constitutes the antecedent for the main clause event. The latter reading is the only possible reading for a preposed when-clause. A postponed when-clause in addition allows for a reading in which no rhetorical relation is established at all between the main clause event and the when-clause event. The relation between the two events is merely temporal then. That a merely temporal reading is not possible in case of a preposed when-clause is shown by the infelicity of sentence (2).
In order to account for the relevant set of interpretations, I will discuss three constraints that appear to play a role in the interpretation of preposed and postponed when-clauses. The constraints are very different in nature. The first one is a general pragmatic constraint that favours anaphoric interpretations. Secondly, a construction-specific constraint about when- and other temporal adjunct clauses is introduced. The third constraint is a constraint at the syntax-discourse interface that links syntactic word order to the order of the antecedent and the anaphor in a rhetorical relation. I develop a cross-modular approach which integrates these various types of constraints into a domination hierarchy. This optimality theoretic analysis construes the optimal interpretations for inputs that contain either proposed or postponed when-clauses.