17-03-2014, 06:20 PM
In a mixed-initiative natural language interface] questions are essentially requests
for information. Such questions entail accompanying expectations about the nature
of information requested. When the user replies to an information-seeking
question posed by the system, his response may not always be a “good” answer to
that question. Thus, the system needs a way of determining whether the user has
adequately answered its question. An expectation-based approach to this aspect
of question-answer understanding is a viable strategy since questions entail accompanying
expectations. In this paper, we discuss representing and extracting these
expectations from system questions and then subsequently using them to evaluate
user answers to these questions.
Introduction
Natural language understanding systems serving as frontends
to databases or expert systems process questions which
are requests for information. Since the user of a database or
expert system may be unfamiliar with system internals, a
mixed-initiative interface can be employed to interact with
the user. The interface can apply expectation-based processing
to its question and answer understanding procedures
in order to cooperate with its user [15]. Expectation-based
processing, in this context, is based on the assumptions:
that the user expects a certain sort of answer to his query;
that the system is able to interpret the user’s query in such
a way that expectational information is extracted; and that
the system develops its own expectations as processing proceeds.
Thus, the interface’s dialogue management capabilities
can be enhanced by determining whether or not user
responses adequately answer system questions, a process
which we term ezpectation resolution.
To incorporate expectation in this capacity, it is necessary
to use expectations which have been generated “on the
fly” rather than determined a pion ’. The potential fatal
flaw with canned expectations is that it is both difficult
and costly to anticipate every possible situation, many of
which may not necessarily be embedded in the form of a
subplan or a slot in a frame. In addition, determining expectation
fulfillment cannot be soundly done solely through
a simple matching scheme. Expectation resolution should
rely on fine granularity in semantics processing since deciding
whether an expectation has been fulfilled is not a purely
binary decision.
In this paper, we maintain that mixed--initiative natural
1angua.ge interfaces must be capable of understanding a wide
variety of user input. We focus on user answers to system
questions since these responses may not always be “good”
answers to their corresponding questions. We posit that
an expecta.tion-based approach to this aspect of questionanswer
understanding is a viable strategy since questions
entail accompanying expectations. We discuss representing
and extracting these expectations from questions and then
using them subsequently to evaluate answers to the questions.
for information. Such questions entail accompanying expectations about the nature
of information requested. When the user replies to an information-seeking
question posed by the system, his response may not always be a “good” answer to
that question. Thus, the system needs a way of determining whether the user has
adequately answered its question. An expectation-based approach to this aspect
of question-answer understanding is a viable strategy since questions entail accompanying
expectations. In this paper, we discuss representing and extracting these
expectations from system questions and then subsequently using them to evaluate
user answers to these questions.
Introduction
Natural language understanding systems serving as frontends
to databases or expert systems process questions which
are requests for information. Since the user of a database or
expert system may be unfamiliar with system internals, a
mixed-initiative interface can be employed to interact with
the user. The interface can apply expectation-based processing
to its question and answer understanding procedures
in order to cooperate with its user [15]. Expectation-based
processing, in this context, is based on the assumptions:
that the user expects a certain sort of answer to his query;
that the system is able to interpret the user’s query in such
a way that expectational information is extracted; and that
the system develops its own expectations as processing proceeds.
Thus, the interface’s dialogue management capabilities
can be enhanced by determining whether or not user
responses adequately answer system questions, a process
which we term ezpectation resolution.
To incorporate expectation in this capacity, it is necessary
to use expectations which have been generated “on the
fly” rather than determined a pion ’. The potential fatal
flaw with canned expectations is that it is both difficult
and costly to anticipate every possible situation, many of
which may not necessarily be embedded in the form of a
subplan or a slot in a frame. In addition, determining expectation
fulfillment cannot be soundly done solely through
a simple matching scheme. Expectation resolution should
rely on fine granularity in semantics processing since deciding
whether an expectation has been fulfilled is not a purely
binary decision.
In this paper, we maintain that mixed--initiative natural
1angua.ge interfaces must be capable of understanding a wide
variety of user input. We focus on user answers to system
questions since these responses may not always be “good”
answers to their corresponding questions. We posit that
an expecta.tion-based approach to this aspect of questionanswer
understanding is a viable strategy since questions
entail accompanying expectations. We discuss representing
and extracting these expectations from questions and then
using them subsequently to evaluate answers to the questions.