Knowledge Representation Questions and Answers

Answer: A

Deciding what to tell someone, or predicting how a competitor will act, requires modelling what they already believe rather than what is true. That is why speech act planning and game playing both need the machinery. Consistency with others, storage and truth evaluation are different activities. A representation adequate for a single agent alone turns out to be inadequate as soon as another agent enters the environment, which is a good argument for building it in from the start.

Answer: B

The standard treatment makes knowing p entail p while believing p does not, so an agent may believe something false but cannot know it. That difference has consequences for what may be inferred from a report. Certainty, ownership and derivation are different distinctions, though probabilistic treatments exist alongside. A logical property that seems merely philosophical turns out to determine what a system may conclude from another agent's assertion, which makes it a practical matter.

Answer: C

Two systems can exchange triples freely and still misunderstand each other unless they agree what the terms mean, and an ontology fixes that agreement. Format conversion and transport are separate layers that assume the agreement rather than providing it. Answer verification is a different exercise. That syntactic interoperability is easy and semantic interoperability is hard is the lesson every integration project rediscovers.

Answer: D

One system's customer may include prospects while another's does not, so merging their data produces confident nonsense although every message parsed correctly. Only shared definitions prevent it. Arity, parsers and validation are engineering matters that are comparatively easy. A failure that occurs after every technical check has passed, because the checks address the wrong level, is the characteristic difficulty of data integration.

Answer: A

Since a single shared ontology has proved unattainable, mapping one vocabulary to another, wholly or partially, is how heterogeneous sources are combined in practice. The mappings are often approximate and require human judgement. Hierarchy construction, consistency and formatting are different tasks. Accepting that agreement will be partial, and building machinery to bridge the differences, is a more realistic programme than insisting on a universal standard.

Answer: B

One ontology's employee may exclude contractors that another's includes, so the best available mapping is a subset or overlap relation rather than an equivalence. Recording the imprecision honestly is better than asserting a false equivalence. Software capability, speed and rewriting are secondary. Representing the degree of correspondence, rather than forcing a binary match, is what keeps the resulting inferences trustworthy.

Answer: C

Since expressiveness costs tractability, the discipline is to identify what genuinely must be stated and to choose the weakest formalism that states it. Maximal expressiveness buys undecidability, abandoning formality loses auditability, and tool availability is a constraint rather than a criterion. Selecting the least powerful tool that suffices is a principle from language design that applies here with unusual force, because the cost of excess power is measurable.

Answer: D

The required inferences determine the expressiveness needed, and that determines which formalism will serve at acceptable cost. Popularity, volume and familiarity are practical considerations that follow rather than lead. Asking what conclusions must be drawn, before asking how knowledge should be written, is the same discipline the agent framework applies when it insists the performance measure be settled before the architecture.

Answer: A

Semantic networks, frames and description logics were each shown to correspond to fragments of first order logic, which is what gave their inferences a precise meaning. The variety lies in notation, in what is made easy and in what is deliberately excluded. Logic was not replaced, and formal semantics is what the correspondence supplies rather than a precondition. Recognising a family of notations as one underlying formalism differently packaged is what lets results transfer between them.

Answer: B

Restricting to a fragment buys termination guarantees and often polynomial reasoning, which unrestricted first order logic cannot offer. Nothing outside logic becomes expressible, semantics is required rather than avoided, and compatibility must still be arranged. That the whole family exists because the general case is intractable, rather than because logic is inadequate, is the summary worth carrying away from the topic.