Since Austin (1962) and Searle (1969) wrote their papers about speech acts, it was clear that the study of language must take into account the way people use it to move in the world. An utterance is an action, so it is made with some goals in mind. Among these goals, getting cooperation from the audience and maintaining a good relationship with them play a major role. The cooperation can range from simple attention (if you just want to chat), to providing information (in case of questions), to performing some general action (as closing a window if the speaker asks the hearer to do so). In all these cases, speech acts must be planned by taking into account the relation between the speaker and the hearer.
A major step in computational linguistics was made when the study of traditional fields as syntax and semantics was complemented with the computational study of pragmatics. However, this was accomplished by paying attention mainly to the first of the two goals mentioned above. In particular, it was recognized that goals and plans play a basic role in linguistic communication (Allen & Perrault 1980), but their study was centered on domain plans.
In the last fifteen years various models of recognition of the speaker's plans were developed, some of which gave fundamental formal accounts of the knowledge which it is based on (Cohen & Levesque, 1990; Cohen & Perrault, 1979), while others had a more computational bias (Carberry, 1988). More recently domain plans have been complemented with higher levels plans called discourse plans (Litman & Allen, 1987) and problem solving plans (Lambert, 1993).
While Litman and Allen's discourse plans dealt both with communication strategies and problem-solving activities, Lambert separates the discourse level in two parts: in her framework, communicating strategies are represented in the communicative level, while problem-solving plans model the activity of building the speaker's domain plans.
The present work addresses mainly the second goal mentioned in the first paragraph: what linguistic forms enable a speaker to manifest her/his choice to be more or less polite with the hearer? The desire of maintaining some harmony with the hearer is just one of the multiple goals of the conversation, so the problem of modeling this desire can be faced from a general perspective of modeling goals. However, the features that express the choices made are rather special; while the propositional content of a sentence enables the hearer, after some rather complex inferential activity, to understand the speaker's goals, it is the form in which that propositional content is expressed that makes the utterance more or less polite. For example, the following sentences have the same illocutionary force, but a different literal interpretation (e.g. 1b refers to the hearer's capabilities, 1c projects on a hypothetical perspective the hearer's action, 1d refers to her/his wants, while 1e simply mentions an unsatisfied precondition of the desired act):
1a) Dammi le chiavi della biblioteca!
[Give me the keys of the library!]
1b) Potresti darmi le chiavi della biblioteca?
[Could you give me the keys of the library?]
1c) Mi daresti le chiavi della biblioteca?
[Would you give me the keys of the library?]
1d) Ti dispiace darmi le chiavi della biblioteca?
[Do you mind giving me the keys of the library?]
1e) La biblioteca è chiusa
[The library is closed.]
Our goal is to get rid of these aspects of the literal interpretation, assuming
that their role is just to mark the politeness strategies the speaker has
adopted in communicating. This approach follows the guidelines drawn by the
research of many linguists, that have investigated the notion of politeness and
its implications in communication (Brown & Levinson, 1987; Kasper, 1990;
Leech, 1983): they have shown that the origin of many indirect forms of
expression lies in the necessity of smoothing the interaction for being polite.
As far as the notion of politeness is concerned, various more or less precise
explanations have been formulated. In our work, we will refer to
Brown and Levinson (1987), who motivate the use of indirect forms of expression
with the necessity to preserve some wants that every interlocutor has. In
order to characterize these wants they introduce the notion of
face as:
.
: A portion of the Speech-act Library
The public self-image that every member [of a society] wants to
claim for himself, consisting in two related aspects:
a) negative face: the basic claim of territories, personal preserves, rights to
non-distraction -i.e. to freedom of action and freedom from imposition
b) positive face: the positive consistent self-image or `personality' (crucially
including the desire that this self-image be appreciated and approved of)
claimed by interactants
Brown and Levinson interpret the behavior of speakers on the basis of a
taxonomy of linguistic strategies that enable a speaker to satisfy the goal of
preserving the negative face of the interlocutor.
For example, when a speaker wants the hearer to perform an action, s/he can
express her/his request directly, using an imperative form; however, in this
way, s/he does not preserve the hearer's negative face: in fact, she does
not hide the presupposition that s/he believes that the hearer
wants to execute the action. So, a safer strategy is to use an indirect
request such as 1d, which doesn't presuppose any hearer's attitude
towards the requested action (in fact, s/he is questioned about that). The
conditional mood in sentence 1c ( mi daresti: `would you give me')
has a similar role: in this case the presupposition is canceled by projecting
the utterance on an hypothetical world.
The various methods for modulating the strength of utterances are chosen according to the degree of familiarity, respect, relative social roles of the interactants, and the impact that the contents of the acts might have on the interlocutors (Brown & Levinson, 1987).
This paper takes into account the suggestions coming from the authors mentioned above to implement a method for processing and evaluating indirect speech acts as politeness forms. This is done within a framework of plan recognition that has already been applied successfully to the recognition of domain plans in an information-seeking environment (Ardissono etal, 1993; Ardissono etal, 1994; Ardissono & Sestero, 1995). It must be observed that Hinkelman and Allen (1989) challenged the possibility of facing this problem on the sole basis of planning structures. They argue that the variability of politeness forms among different languages calls for the introduction of knowledge about idioms. While we agree on the need of language-specific knowledge, we will show that the required information can be encoded within a plan formalism, so that the homogeneity of the representation is preserved.
The rest of the paper is organized as follows: the first section 2 describes the formalism used for representing the knowledge about speech acts; the second 3 describes how the speech acts library is used in the process of speech act recognition; the third section 4 shows the speech-act recognition process on an example. Finally, some brief conclusions 5 are presented.
