Information Technology Reference
In-Depth Information
these actions, whose results are visible on screen. This task puts the accent on
object reference phenomena: what object is referred to by “the pyramid”? To
correctly interpret such a reference, the system must find among the objects
on display which one is correct, meaning which one corresponds to the user's
intent.Iftwoorthreepyramidsarevisible,thesystemcanthusanswer“Idonot
understand which pyramid you mean”. After clarification, it does what it must
do, that is carry out the actions and answer questions. Many of the possible
questions revolve around the physical world of objects: “what does the box
contain?” and “what is the pyramid supported by?”. Each time, SHRDLU is
able to analyze the image, identify the spatial relations between objects, count
and answer. Certainly, a world of geometric objects remains simple. But all
these implemented automatic understanding processes are impressive, as well
as the matching knowledge modeling: the system is able to solve complex
references, such as “a block which is taller than the one you are holding”, to
solveanaphorasuchas“putitinthebox”, toidentifyspeechacts.Theresulting
dialogue is focused on the essential. There may be a lack of fluidity, but the
goal is to satisfy the task, and indeed, all is done for this to happen.
As for GUS [BOB 77], it takes an additional step into the utilitarian
MMD, with a flight reservation task. To demonstrate this research prototype,
the database only comprises a single flight in California. Beyond this
limitation, the linguistic modeling, the computer modeling and the
methodological aspects give an idea of what the MMD domain will look like
in a few years. Just as SHRDLU, the system is able to solve object and
anaphora references, at least when they directly concern the task's objects,
that is the flights, days and timetables. For example, it manages to allocate a
date-type reference to the referential expression “Friday” used as a return date
after specifying “May 28th” as an outward flight. The interpretation of the
user's utterances triggers a syntactic and semantic analysis that can be partial,
and thus operate on other linguistic materials than just full sentences. It also
triggers a recognition of speech acts, notably with the understanding of
indirect answers to some questions. The great results of linguistic works on
the dialogue structure and the information structure are used, which leads to
the system managing a great deal of knowledge on language: lexicon (3,000
roots recorded, which is greater than the precedent systems), morphological
rules, syntactic constructions, simplified principles of the informational
structure, patterns for the dialogue structure, conceptual models for the travel
plans and dates, and finally the agenda model: central structure that will allow
the system to manage events and know at any moment what task to carry out.
Search WWH ::




Custom Search