The knowledge about speech acts and the way they relate to each other is stored
in the speech acts library, represented in an action hierarchy based on a
formalism similar to that by Kautz (1990). In particular, we set
apart the decomposition hierarchy (boxed arcs in the figures) and the
generalization-specialization hierarchy (thick arrows).
When the decomposition includes a single step, the relation between the two
actions is a generation relation (Pollack, 1990).
The leaves of the hierarchy, surf-imperative, surf-yn-question,
surf-wh-question, surf-assertion correspond to the different syntactic types
of sentence, namely imperatives, declaratives and interrogatives (two small
portions of the library are reported in Figure 1 and
2. There, the surface types are circled by thick ovals).
The actions of the hierarchy are characterized by the following features:
: Representation of the ask-if speech-act
and a reference to the speech act. The third parameter has different meanings in the various actions of the speech act library: since the interpretation of surface speech acts starts with the analysis of the linguistic aspects in the input utterances (e.g. the detection of politeness features), the actions related to that phase refer to the semantic representation of the input sentences (e.g. consider ask-if, ..., indirect-req in Figure 1). On the other hand, after considering the linguistic aspects, the analysis goes on taking into account the knowledge about domain actions (in order to relate the speaker's utterance to domain goals). So, the third parameter of actions referring to this phase of the analysis refers to an instance of a domain action involved by the speaker's utterance.
The domain action is recognized from the semantic representation by a plan recognition phase ( action identification (Carberry, 1990), shown in the figures as act-id).
The recursiveness of natural language implies that illocutionary force indicating devices can be nested inside each other; so, complex utterances including different speech acts can be built and interpreted in a compositional way. For example, the sentence:
2) Vorrei chiederti se puoi dirmi dove si trova la biblioteca.
[I would like to ask you whether you can tell me where is the library.]
is composed of an external surface statement with conditional mood
( vorrei, ``I would like"), an explicit performative ( chiedere,
``to ask") and an indirect request expressed by an inner
yes/no question ( se ..., ``whether ...").
Because of the freedom in the composition of sentences, the speech acts
library contains some cyclic paths (see the ask-if action that, in
figure 2, occurs in its own definition).