Models, information and meaning

Highlights

• Models of signalling are widely employed to learn about semantic relations in the natural world.

• This success fails to lend support to an informational theory of meaning.

• Signalling models face the partition problem, and any plausible solution seems to presuppose (rather than provide) a theory of meaning.

• Current accounts fail to satisfactorily address the classical objecions to informational theories of meaning.

• In signaling models, information is used to track non-informational relations in the target system.

Abstract

There has recently been an explosion of formal models of signaling, which have been developed in order to learn about different aspects of meaning. This paper discusses whether that success can also be used to provide an original naturalistic theory of meaning in terms of information or some related notion. In particular, it argues that, although these models can teach us a lot about different aspects of content, at the moment they fail to support the idea that meaning just is some kind of information. As an alternative, I suggest a more modest approach to the relationship between the informational notions used in models and semantic properties in the natural world.

Introduction

The idea of explaining the origin and development of meaning in terms of some kind of evolutionary process has been popular for some time. Nonetheless, it has been difficult to spell this idea out in detail. Partly for this reason, many have welcomed the use of game-theoretic models, which have already provided more precise definitions, new arguments and suggestive ideas that have greatly enriched the debate and contributed to a better understanding of this phenomenon. Some striking results concern the emergence of meaning (Skyrms, 2010a; Hutteger, 2007a; 2007b), the evolution of perceptual categories (O'Connor, 2014), concepts (Barrett, 2014) and moral norms (Harms & Skyrms, 2008), among others.

A clear virtue of game-theoretic models of signaling is that they provide a simplified and indirect representation of much more complex real-world dynamics, which makes it possible for the theorist to give precise definitions and analyze specific questions. In particular, current research usually quantifies the information (or some other correlational measure) between variables in the model in order to learn about meaningful relations in the natural world. This fact has encouraged some people to argue that a naturalistic analysis of meaning can be provided in terms of information or related concepts. This is the main claim that I would like to assess in this paper. More precisely, I would like to discuss the relationship between meaning and the semantic notions defined in models.

The paper has three main sections. Section 2 frames the discussion, puts forward a key assumption (‘model-independence’), and briefly explains some notions of information that have been put forward in the literature. Section 3 discusses the main target of the paper: the ‘immodest view’, according to which the relative success of game-theoretic approaches can support an informational theory of meaning. Finally, in section 4, I present and defend a more plausible view of the relationship between information and meaning, which I label the ‘modest view’.