Information Technology Reference
In-Depth Information
leaves, I don't have any seats left for the Paris train”, which has the same
illocutionary value with a slight overtone that warns the user that he/she
should have done this earlier and a perlocutionary value aiming to modify the
user's beliefs on how to reserve a ticket. In a similar vein, “I do not have any
seats for Paris left, only for Massy-Palaiseau” carries a perlocutionary value,
which tries to modify the user's initial goal by inciting him/her to consider an
alternative goal, but not expressing it openly, which would have been “are you
ready to change your destination to Massy-Palaiseau?”.
In natural language, managing the perlocutionary value is close to
inference management and planning in dialogue. In other modalities, such as
in an MMI, it can be to simply make clear the different possibilities that the
user has following one of his actions. In an MMI, buttons are pressed, text is
typed in areas that resemble input fields and table cells are clicked on: we
know that any displayed element has a function, and we explore this function
with the means offered by the keyboard-mouse interaction. Thus, the
multimedia presenter managing the MMI at the same time as the MMD has to
take into account the MMI element functions in the different presentation
phases: for each element involved in the previous or current action, it has to
know the input interaction possibilities, eventually block them (grayed out
button, table cell displayed in a certain color) and, if needed, let the input
manager module know.
This strategy consists of anticipating the user's future actions depending
on a specific perlocutionary value and can have numerous complex
consequences. When the MMD system asks the user a question, and expects
an answer from a set of clearly identified alternatives, when it incites the user
to talk about a new topic and when it presents information in a specific order,
and thus incites the user to refer to this information, it leads to a reduction in
the interaction possibilities. The user's next utterance can thus almost be
predicted. Thus, the speech recognition and linguistic analysis modules
benefit greatly from being warned. Taking into account the difficulties in open
domain speech recognition, several speech models can be involved: a
generalist model can be used at the beginning of the dialogue, but a model
oriented toward numbers, dates and values can be used when the user has to
answer a question asked by the system and concerning such data. In the same
way, following an information presentation in a significant order, the user can
be led to use mentional references such as “the first”, “the second” and “the
last two” or even quantified expressions such as “each...”or“all the...”.The
Search WWH ::
Custom Search