Departmental Papers (ASC)

Document Type

Conference Paper

Date of this Version


Publication Source

National Conference on Content Analysis


This paper identifies the problem of analyses of message content as one of making specific inferences from recorded text to characteristics of a source that are not directly observable. The problem is common to a variety of analytical situations in the humanities and in the social sciences; and the way adequate solutions are found therefore deserves systematic attention.

Choices among investigative techniques always imply assumptions regarding the structure of a source. In message analysis, investigative methods crucially affect the conceivable relations between the recorded text and the content if presumable conveys to the analyst. While there is no appropriate theory of these (basically semantic) relations to which the notion of "message" refers, it seems that three classes of analytical constructs of basic models of messages account for much of current investigative efforts. This paper examines the nature and limitations of these models which are as follows:

Association models of message are identified by their use of statistical correlations as a logical basis for content inferences from text. Whether correlations are demonstrated or postulated, such models assume that content indicators permeate throughout a text, that the text is not purposively intended and that syntactic constructions and their possible referentiality can be ignored. While preferred by many content analysts, association models provide the weakest basis for content inferences.

Discourse models take linguistic references as the primary basis for inferences from text. Requisite analytical procedures are not statistical but essentially algebaical and incoporate psychological or social constructs of the semantic domain of a discourse. Discourse models are incapable of handling instrumental uses of language.

"Communication models of messages" refers to a class of analytical procedures that go beyond linguistic references and/or associations by considering the behavioral constraints that the exchange of messages may impose on a system of interacting communicators. Recorded texts then take the form of chronologies of exchanges and communication models render such chronologies informative about the parameters of an interaction system including the relations among the communicators and their mutual control. While communication models are the most interesting, least is known about them. This paper therefore elaborates only on some of their formal prerequisites.

Association models empoly familiar behavioral science methods and therefore provide so serious obstacles for their possible computerization. Algorithms for discourse models presuppose a considerable thoretical work, particularly in linguistics and semantics, and it is already evident that no "general discourse analyser" can be constructed. So far attempts to computerize communication models of messages have been limited to the most reduced situations. Further, although such models are potentially most powerful very little can be expected from current computational technology.

The paper finally suggests that efforts should be directed toward formalizations of content inference processes if analytical success is to be improved.


Preliminary Draft: Prepared for Presentation at the National Conference on Content Analysis, November 16-18, 1967.

Included in

Communication Commons



Date Posted: 09 March 2012