There are various theories about the relation between language and thought. One prominent version in contemporary philosophy is called the language of thought hypothesis. It states that thinking happens in the medium of a mental language. This language, often referred to as Mentalese, is similar to regular languages in various respects: it is composed of words that are connected to each other in syntactic ways to form sentences. This claim does not merely rest on an intuitive analogy between language and thought. Instead, it provides a clear definition of the features a representational system has to embody in order to have a linguistic structure. On the level of syntax, the representational system has to possess two types of representations: atomic and compound representations. Atomic representations are basic whereas compound representations are constituted either by other compound representations or by atomic representations. On the level of semantics, the semantic content or the meaning of the compound representations should depend on the semantic contents of its constituents. A representational system is linguistically structured if it fulfills these two requirements.

The language of thought hypothesis states that the same is true for thinking in general. This would mean that thought is composed of certain atomic representational constituents that can be combined as described above. Apart from this abstract characterization, no further concrete claims are made about how human thought is implemented by the brain or which other similarities to natural language it has. The language of thought hypothesis was first introduced by Jerry Fodor. He argues in favor of this claim by holding that it constitutes the best explanation of the characteristic features of thinking. One of these features is productivity: a system of representations is productive if it can generate an infinite number of unique representations based on a low number of atomic representations. This applies to thought since human beings are capable of entertaining an infinite number of distinct thoughts even though their mental capacities are quite limited. Other characteristic features of thinking include systematicity and inferential coherence. Fodor argues that the language of thought hypothesis is true as it explains how thought can have these features and because there is no good alternative explanation. Some arguments against the language of thought hypothesis are based on neural networks, which are able to produce intelligent behavior without depending on representational systems. Other objections focus on the idea that some mental representations happen non-linguistically, for example, in the form of maps or images.

Computationalists have been especially interested in the language of thought hypothesis since it provides ways to close the gap between thought in the human brain and computational processes implemented by computers. The reason for this is that processes over representations that respect syntax and semantics, like inferences according to the modus ponens, can be implemented by physical systems using causal relations. The same linguistic systems may be implemented through different material systems, like brains or computers. In this way, computers can think.

The alphabet of human thought (Latin: alphabetum cogitationum humanarum) is a concept originally proposed by Gottfried Wilhelm Leibniz that provides a universal way to represent and analyze ideas and relationships by breaking down their component pieces. All ideas are compounded from a very small number of simple ideas which can be represented by a unique character.