Given the sentence:
3) Posso chiederti di darmi le chiavi della biblioteca?
[May I ask you to give me the keys of the library?]
The surf-yn-question action is activated on the
basis of the interrogative form; the third parameter of the action
is instantiated with its propositional content, that refers to the node
of the semantic representation ( sem1), associated to potere (`may').
The instantiated surface speech act is:
surf-yn-question (User, System, Sem1)
After the identification of this speech act, the analysis proceeds with the activation of the speech acts of which it is a substep or a specialization (upward expansion in the speech acts library (Carberry, 1988)): the direct-ask-if and then the ask-if actions are activated. Note that surface-yn-question could be considered as a direct substep of obtain-info (in a `generation' relation). However, the net specifies that a surface-yn-question generates an ask-if, which in turn generates obtain-info. In this way, we are able to factorize an effect (the Cint effect of ask-if) that is shared by obtain-info and the other actions that are generated by ask-if (e.g. ind-req1 or hedged-perform as shown in Figure 1 and 2); on the contrary, the peculiarity of the obtain-info (i.e. the precondition of not knowing the answer) is kept separate (in fact, in indirect acts performed by means of a question, the speaker almost always knows the answer to the question). Moreover, this effect is inherited both by indirect-ask-if and direct-ask-if through the specialization hierarchy.
When an action is in the decomposition of more than one speech act, more than
one alternative hypothesis can be built (in the example, for the sake of
simplicity, we only consider obtain-info, ind-req1 and
hedged-perform). However, the domain-level processing rejects the
obtain-info since here, as usual, it does not make sense that the speaker
questions the hearer about her/his own capabilities; ind-req1 can not be
instantiated because the node associated with potere (`may') should have
the hearer as semantic agent, while in the example the agent is the speaker
(compare with sentence 1b). So, only
hedged-perform is activated, because all its restrictions are satisfied.
Since only one higher-level action has been instantiated, no ambiguity arises
in the interpretation of the user's utterance and the upward expansion goes on,
extending the unique hypothesis. The on-record-req and request
actions are activated, so interpreting the sentence as a request by the user to
perform the domain action:
give (System, Keys, User),
that is identified by means of an action identification phase. Here, this
phase is carried out easily, because the request is posed explicitly
and the identified action coincides with the one expressed by the user.
The situation is very different for the so called off-record requests
(Brown & Levinson, 1987), where the speaker doesn't express in an explicit way
the requested action, but s/he states one of her/his goals or s/he asks whether
some precondition of the action is satisfied: e.g. ``I would like to open the
library" or ``Do you have the keys of the library, please?". In this cases,
the requested actions must be inferred from the utterance using the knowledge
about domain actions (the task is performed by the domain plan recognition
process).