When the maxim be relevant was formulated for the first time as part of a cooperation principle governing people's conversations, it was already clear that its terse formulation concealed a number of problems, such as questions about what different kinds of focuses of relevance there might be. Thus it was not much later that in an article on conversational relevancy a distinction was suggested between 'pragmatic' and 'semantic' relevancy, the former being goal-dependent, the latter referring to a symmetrical relationship. Pragmatic relevance is, then, defined as the relevance of speech acts to certain goals and must be considered as a 'specialization of the general notion of relevance of an action to a goal' (part of the theory of goal-directed behavior). Semantic relevance is said to be 'the relevance of certain linguistic, logical or cognitive entities (propositions) to other entities of the same type'. As the concepts which it involves have been mentioned reference, entailment and meaning relations.

The 'pragmatic' and 'semantic relevancy' of philosophy of language have been called "practical" and "semantic relatedness" in logics. So one logician speaks of "the practical relatedness of events and actions" and "the semantic relatedness of the subject-matters of propositions". (Another logician writes that relatedness is only chosen instead of relevance because of its different technical meaning in relevance logics.) There is accordingly no essential difference between pragmatic and semantic relevancy in philosophy of language on the one hand and practical and semantic relatedness on the other.

Authors who do not distinguish a symmetrical from an a- or non-symmetrical relevancy relation simply presuppose that all relevancy is of the goal-dependent kind like pragmatic relevancy and practical relatedness. The first philosopher to publish an article on relevancy wrote already that the relevance of a thing 'lies in its value for us and in our attitude towards it'. Whether something is relevant depends 'on the purpose of the moment', and one must not turn 'the usefulness of things for our purposes into an attribute of the things in themselves', into 'a quality residing in the thing thought of per se'. (Relevance implies a relation to a purpose 'by its very etymology' this author said.) The goal-dependent character of the relevancy notion is less clear, but certainly present too in the first author to write on relevancy from a phenomenological point of view. 'E speaks about everyone's 'system of relevances' being determined by one's 'interest at hand'. This interest 'motivates all one's thinking, projecting, acting, and therewith establishes the problems to be solved by one's thought and the goals to be attained by one's actions'. The concept of goal is closely related to that of value, and consequently it is not surprising that the same author has claimed that a theory of relevances (or of 'topical relevances', to be precise) will contribute to 'a classification of the concept of value', that is, to a classification of the values by which one wants to be guided in one's theoretical and practical life. When later theorists point out that it is useful to talk of utterances, and to see them 'as being relevant to previous utterances, a discourse, a question, an issue or a goal' it is clearly goal-dependent or pragmatic relevancy again they are thinking of.

Relevancy has not only been subdivided into a pragmatic and a semantic type in philosophy of language. A further subdivision in that discipline is that into topical, marginal and potential relevance. 'Topical relevance' is what is said to be at the center of the subject's field of attention at time t; 'marginal relevance' is everything that is still somewhere in the subject's field of attention, but in 'er horizon; and 'potential relevance' is said to concern the members of a domain of stored data which might be referred to as "the background for S at t". What is particularly interesting about this classification is its similarity with a phenomenological subdivision to be discussed in the next section. This will be another confirmation of the fact that philosophers in different fields of inquiry have not seldom categorized the types of relevancy in ways which highly resembled each other. And this, in turn, will be a clear indication that there is indeed a unity in the diversity of relevancy notions.

But before turning to phenomenology we should first have a brief look at a conception of relevancy of quite a different complexion which has been developed in philosophy of science. In the theory in question a factor is said to be statistically relevant to the occurrence of an event if it makes a difference to the probability of that occurrence. Facts can be statistically relevant on this view even in the absence of high probability, since both the probability with and the probability without a particular property may be low. If they are not equally low, a property is statistically relevant nevertheless, whereas it would be irrelevant if both probabilities were high, but equally high. This relation of statistical relevancy is said to be symmetrical. The rule is to choose the broadest homogeneous reference class to which a single event belongs. Yet, it has been admitted that this formulation may not remove 'all ambiguities about the selection of reference classes either in principle or in practise'. If the explanation is causal (instead of symptomatic), this provides a more homogeneous reference class, and --it has been argued-- 'each progressively better partitioning makes the preceding partitioning statistically irrelevant'.

The notion of statistical relevancy seems to be subjected to a rather stringent procedure, and to be little belief-dependent or not belief-dependent at all. In spite of this, the philosopher of science or the scientist cannot guarantee either that a partitioning at any time is the best one, nor that the selection of reference classes is in no way arbitrary or merely traditional. Without any change in the external world (or nonpropositional reality) something that is 'statistically relevant' today may be 'statistically irrelevant' tomorrow. But this means that the notion of statistical relevance is knowledge-dependent, and in that sense not objective either. At the most it is of some epistemic nature.