Knowledge Representation Questions and Answers

Answer: A

A representation must be able to say what needs saying, allow conclusions to be drawn in reasonable time, and be practical for a human or a program to populate, and improving any one generally worsens another. Logical properties and engineering qualities matter but are not the governing trade. Recognising that the choice of representation is a compromise between three incompatible goods, rather than a search for the best one, explains why so many representations coexist.

Answer: B

Full first order logic can express almost anything but its entailment problem is only semidecidable, while a propositional or a restricted description logic gives decidable and often polynomial reasoning. Storage and readability are secondary. Designers therefore choose the least expressive language that says what must be said. That the difficulty of reasoning rises with what can be stated is the single most important fact about representation, and it shapes every language in the field.

Answer: C

A semantic network represents objects and categories as nodes and relations such as is a and has part as labelled edges, which makes inheritance a matter of following links. Its appeal was diagrammatic clarity and its weakness a lack of precise semantics until later work supplied one. Distributed systems, neural networks and probabilistic graphs are different structures. A notation that is intuitive before it is formalised often needs the formalisation before its inferences can be trusted.

Answer: D

Early networks used links whose meaning varied between systems and sometimes within one, so what a diagram entailed was unclear. Giving them a translation into first order logic settled the matter and showed they were a notational variant of a fragment of it. Relations were exactly what they represented, and memory and scale were practical issues. A representation whose inferences cannot be stated precisely cannot be debugged, which is why formal semantics matters even for an informal looking notation.

Answer: A

A frame groups everything known about an object or a category into named slots with values, defaults and attached procedures, which matches how people organise knowledge and makes retrieval efficient. Frames and semantic networks are largely notational variants of each other. Temporal windows, scope boundaries and logical sentences are different notions. Organising knowledge around objects rather than around sentences is a difference of packaging that nevertheless changes what is easy to do with it.

Answer: B

Stating that birds fly once, and that a canary is a bird, saves stating it for every species, and the inference is a matter of following the is a links upward. It is the principal economy such representations offer. Run time transfer, system migration and caching are different notions. Storing a property at the most general level at which it holds is an economy of expression whose price is the difficulty of handling exceptions.

Answer: C

Penguins are birds and do not fly, so the inherited default must be overridden, which means the representation is no longer monotonic and adding a fact can retract a conclusion. Handling that properly requires default or non-monotonic logic. Exceptions are representable, circularity is a separate hazard, and new categories are one workaround among several. A representation that is convenient until exceptions appear, and exceptions always appear, is the recurring difficulty of this area.

Answer: D

Classical logic is monotonic: adding premises never invalidates a conclusion, whereas real reasoning from defaults must retract when an exception is learned. Learning that Tweety is a penguin withdraws the conclusion that Tweety flies. Certainty decay, directionality and operator properties are different notions. Departing from a property so basic that classical logic takes it for granted is what makes this family of logics both necessary and difficult.

Answer: A

Circumscription formalises the assumption that things are as normal as the facts permit, by minimising the set of objects satisfying an abnormality predicate. Birds fly unless abnormal, and only known exceptions are abnormal. Scope limitation, fragment restriction and inference bounding are different techniques. Making the assumption that nothing is unusual unless stated into a formal minimisation is what turns an informal habit of thought into an inference rule.

Answer: B

Default logic adds rules of the form: if this holds and it is consistent to assume that, then conclude that, which lets birds fly by default while permitting exceptions. Multiple defaults may conflict, giving several extensions and requiring a choice. Truth values, axiom libraries and typicality restriction are different notions. A rule whose applicability depends on what cannot be proved, rather than on what can, is what makes such logics computationally awkward and expressively necessary.