Page 156 - DCAP310_INTRODUCTION_TO_ARTIFICIAL_INTELLIGENCE_AND_EXPERT_SYSTEMS

Introduction to Artificial Intelligence & Expert Systems

Notes Inheritance

Given the enormous number of linguistic facts that are brought to bear in language processing,
it is clearly important to represent this knowledge in an efficient and general way. Inheritance
is a powerful technique to represent generalizations over descriptions in a way similar to is-a
relations in semantic networks, and to use these generalizations to make inferences. Daelemans
et al. (1992) motivate the use of inheritance for linguistic knowledge as follows. Imagine that we
are setting out on the job of building a lexicon for English. We begin by encoding everything we
know about the verb love, including its syntactic category and how it is conjugated. Next, we
turn our attention to the verb hate. Although these words are antonyms, they nevertheless have
a lot in common: for example, they are both transitive verbs and have past participle ending on
-ed. To save time and space, both can be categorized as transitive verbs and the common
information about these kinds of words can be encoded in just one place called, say, transitive
verb. The verb love then inherits information from transitive verb. Similarly, when we hit upon
intransitive verbs, we collect all information about this kind of verbs in a place intransitive
verbs. The next step is to extract from both classes of verbs the common information which is
then stored in an even more abstract category verb.
Figure 8.1: Inheritance Example

Suppose, we discover the verb beat, which is transitive but not regular: it has a past participle
ending on -en. If we let beat inherit from transitive verb, we still need to specify the exceptional
information, but then we get an inconsistency with the inherited information. The obvious
solution is to let the exceptional information override inherited information. This mechanism is
called default (or non-monotonic) inheritance. Default inheritance is incorporated in semantic
networks and frame-based representation languages used in AI.

Did u know? The sentence production model IPF, is implemented in a frame-based
language using default inheritance. Default inheritance is also used in specific linguistic
formalisms, for example DATR, a formalism for lexical representation.

8.1.1 Market Passing and Spreading Activation

One of the inference mechanisms available in semantic networks is marker passing, a process
with which intersection search can be implemented. This type of search starts from two nodes
which pass a marker to those nodes which are linked to them, a process which is repeated for
each of those nodes. Whenever a marker originating from one of the original nodes encounters
a marker originating from the other node, a path is found representing a association between
the two concepts represented by the nodes. This way, semantic associations can be modeled, but
marker passing can also be used in networks representing other types of linguistic knowledge.

150 LOVELY PROFESSIONAL UNIVERSITY