Research activity in 1994
Our research focuses mainly on 4 areas: dependency syntactic formalisms, plan
recognition, cognitive architectures for natural language processing, lexical
semantics.
1) To treat complex syntactic phenomena, a mechanism based on features
unification has been introduced. The features represent a factorization of the
syntactic information in terms of elementary units that are propagated along
the syntactic structure, relating long-distance terms of a sentence. Moreover,
a hierarchic description of the syntactic constraints has been introduced,
avoiding redundancies by means of inheritance mechanisms. The parser, that
produces a compact representation of the syntactic structures associated to a
sentence, has been enhanced with the feature constraints check for the
recognition of complex syntactic phenomena. Finally, we have started an
empirical work to develop an extended grammar covering a large amount of
linguistic phenomena occuring in real texts.
2) The recognition of a user's goals is essential to allow a natural and
friendly interaction. Such a recognition is based on an a-priori knowledge of
the possible goals and of the possible plans to obtain them. In this area, our
activity has been concerned with modeling knowledge related to the plans in a
restricted domain (we considered information-seeking dialogues between the
consultant system and the students about the activities in a Computer Science
Department) and with the development of procedures working on them. In
particular, the data structures necessary to carry out the recognition of the
user's plans (Context Models) are incrementally built during the interaction
with the users and then used to provide cooperative answers to them. A research
area that has been explored recently concerns the interaction of the plan
recognition model with a user model. Such an interaction has a double effect:
on one hand, it allows for making use of the specific knowledge of a particular
user to favour some of the possible interpretations of the user's sentences; on
the other, it allows for enriching the user model on the grounds of recognized
plans, adding to it all the intentions common to all the alternative
interpretations of the given sentences.
3) Research on cognitive architectures for natural language systems have lead
to the definition of an environment to check psycholinguistic hypotheses on the
human comprehension mechanism. In particular, we started an experimentation to
test the validity of psycholinguistic theories based on the efficiency of
parsing when applied to real texts, that are usually read and understood by
human users without particular effort. Besides, we have refined the mechanism
for the treatment of errors, in order to correct possible misunderstandings,
due to incorrect interpretations provided by the system because of their high
reliability in a local context.
4) The work on lexical semantics started from the acquired consciousness that
any natural language comprehension system cannot be realized without an
adequate lexical knowledge base, containing highly structured information. In
particular, our preliminary study on a restricted domain has focused on the
analysis of about 500 Italian communication verbs, on which we have been
individualizing the semantic components useful to build up their lexical
semantic representations. The working hypothesis is the definition of the
linking rules that determine the superficial syntactic behaviour on the grounds
of the semantic features.
The theoretical activity we have just illustrated comes along with the
extension of an existing prototype. In particular, the architecture has been
enhanced by a flexible instrument allowing to test various solutions to the
specific problems we have just described.
The results of our activity have been published on reviews and proceedings of
international conferences.
