Relevant

Relevance is a relationship between entities signifynig a significant connection, suggesting that when dealing with one entity, it may be helpful to consider the other.

Discussion

Relevance is difficult to pin down philosophically and theoretically due to its highly contextual and broad nature. A broad probabilistic definition could state that “A is relevant to B if A increases the likelihood of B.”¹Hjørland & Sejer Christensen specifically define this in the context of goals in Work tasks and socio-cognitive relevance: a specific example, but I think it applies more broadly. E.g., “my mother’s existence is relevant to my existence because her existence increases the likelihood of my existence.”²The application of this definition, ironically, requires massaging the formulation of relevance statements into a relevant form. E.g., “A pedestrian near a (self-driving) car is relevant to the safe driving of the car.” Unearthing hidden assumptions is normal in logic and real-world reasoning.³Premise selection in mathematical theorom proving is essentially about determining the relevance of various axioms to a conjecture to be proven. Relevance is world model-dependent. Relevance theory‘s core principle is that “every utterance conveys a presumption of its own optimal relevance” (to the situation). I postulate that relevance could perhaps be grounded in meaning: something is relevant to an agent if it means something to the agent, which by Thórisson and Talevi’s theory of foundational meaning, implies that it affects the agents world model, goals, or plans.

I decided to keep the definition of relevant very vague, to be defined primarily in-use by considering relevant to be a binary predicate among two entities. This is inspired by (refers ?E1 ?E2), which is already in the SUMO KB and denotes that entity ?E1 refers to entity ?E2 in some manner. Philosophically, I see an argument that parameters for the agent and the situation should both be included. It may be good to have both subjective terms (e.g., similarity) and vaguely objective terms so that the differences in their use can be studied⁴Given resource constraints, I cannot formalize everything perfectly. Therefore, please look at this as an exploration as to how relevance may be formalized.. I sketched out a few introduction rules: for example, any object playing a case role in a process is (probably) relevant to the process.

SUMO

(documentation relevant EnglishLanguage "The predicate relevant attempts to ontologically represent the notion of 
an entity ?E1 being relevant to ?E2: (relevant ?E1 ?E2). Relevant: having a bearing on or connection with the 
subject at issue; 'the scientist corresponds with colleagues in order to learn about matters relevant to her 
own research'.")
(instance relevant BinaryPredicate)  
(domain relevant 1 Entity)
(domain relevant 2 Entity)

(=>
  (and
    (instance ?E2 Process)
    (patient ?E2 ?E1))
  (relevant ?E1 ?E2))

(=> 
  (and
    (instance ?E1 Object)
    (instance ?E2 Process)
    (exists (?ROLE)
      (playsRoleInEvent ?E1 ?ROLE ?E2)))
  (relevant ?E1 ?E2))

relevant is a binary predicate of two entities⁵Note that this generality allows for claims about relevance among abstract obects, not just physical ones.. If entity E1 is a patient of entity E2, then E1 is relevant to E2. If an object E1 plays some role in a process E2, then E1 is relevant to E2⁶The limited scope for “playsRoleInEvent” vs “patient” is a bit weird..

If a process (partly) takes place in a location, then the location is probably relevant to the process⁷Counterexapmle: a conversation on how to formally define relevance in a fast-food chain is not in a relevance relationship with the fast-food chain..

(=> 
  (and 
    (instance ?E1 Physical)
    (instance ?E2 Process)
    (partlyLocated ?E2 ?E1))
  (modalAttribute
    (relevant ?E1 ?E2) Likely))

A generalization of Hjørland & Sejer Christensen’s goal-centric definition can be codified: if formula F1 (being true) increases the likelihood of formula F2 (being true), then ?F1 is relevant to F2.

(=>
  (increasesLikelihood ?F1 ?F2)
  (relevant ?F1 ?F2))

The cognitive principle of relevance in relevance theory claims that “every utterance conveys the information that it is ‘relevant enough for it to be worth the addressee’s effort to process it.'” A weak form of this prinicple can be easily stated: for every instance of communication, the agent believes that the message is relevant to the situation (which includes the recipient).

(=>
  (and
    (instance ?COMM Communication)
    (agent ?COMM ?AGENT)
    (patient ?COMM ?MESSAGE)
    (situationOf ?COMM ?SITUATION))
  (holdsDuring (WhenFn ?COMM)
    (believes ?AGENT
      (relevant ?MESSAGE ?SITUATION))))