Page 157 - DCAP310_INTRODUCTION_TO_ARTIFICIAL_INTELLIGENCE_AND_EXPERT_SYSTEMS

Unit 8: Structured Representation of Knowledge

A similar inference mechanism in networks is spreading activation where instead of discrete Notes
symbolic markers, a continuous (numerical) activation level is propagated along the links of a
network. A model of lexical retrieval where a semantic network with labeled links is combined
with spreading activation.

8.1.2 Conceptual Dependency Structures and Conceptual Graphs

Many of the early symbolic AI research on natural language understanding used semantic
network or frame-based formalisms to represent its theoretical insights. Schank and his students
developed Conceptual Dependency Theory for the description of the meaning of sentences and
texts (Schank, 1975; 1980). This theory was based on semantic networks, but defined only a
limited number of node types and link types (conceptual primitives) that were deemed necessary
and sufficient as a language of thought to represent meaning unambiguously. Any implicit
information in the text (information that can be inferred by the reader) was to be made explicit
in the conceptual dependency representation.
This goal gave rise to the development of a large number of data structures and inference
mechanisms (often without a well-defined semantics). Data structures included causal chains
(a chain of states enabling or motivating actions which in turn result in, or initiate, other states),
scripts and scenarios (prepackaged sequences of causal chains), and MOPS (Memory Organization
Packages) (Schank & Abelson, 1977; Schank, 1982). These data structures enabled directed and
efficient inference mechanisms, based on following up causal connections and associations
between representations at the same and at different levels of abstraction.
One problem is that these models tend to focus on the data structure, and are vague on the
inference part.
Two sources of knowledge are indispensable for developing useful symbolic natural language
understanding systems: (1) knowledge about the intentions, plans and goals of different agents
in narratives or dialogue, and (2) knowledge about preceding discourse (discourse representation).
In work by Allen and Perrault (1980) and others, AI planning formalisms are combined with
speech act theory to model the recognition of intention, an approach which gave rise to research
on speech act planning, topic structure modeling, and user modeling. This AI work has influenced
psycholinguistic models of discourse comprehension and discourse production

Self Assessment

State whether the following statements are true or false:

1. A semantic network is regarded as a graphical notation for logical formulas.
2. In a semantic network, each link between four nodes represents a separate proposition.

8.2 Production Systems

Production systems are rule-based systems developed during the seventies as models for human
problem solving. They are common in models for many areas of knowledge. In this kind of
formalism, knowledge is expressed as rules taking the form of condition-action pairs: if X then
do Y.

Example: In a model for language production, one of the rules for producing questions
might be the following: if the intention is to query the truth of P, then produce a sentence about
P where the finite verb of the main clause is moved up front.

LOVELY PROFESSIONAL UNIVERSITY 151