In 1755, Samuel Johnson defined network as 'any thing reticulated or decussated, at equal distances, with interstices between the intersections.' The definition is accurate. It is also nearly useless to anyone who does not already know what a network is. A reader must now look up reticulated, decussated and interstices, and each of those leads to yet more words.

Johnson's sentence is usually quoted as a joke, but it exposes something serious. A dictionary appears to be a simple inventory, with words on the left and meanings on the right. In practice, it is an attempt to represent one of the least understood objects in cognitive science: the mental lexicon, the store of words and meanings that every speaker carries and almost none can articulate.

Lexicographers confront the problem of meaning daily, and under deadline. Their solutions, compromises and outright failures form a kind of accidental experiment. They show which aspects of meaning can be captured in words and which cannot. They also show why the authority we grant dictionaries rests on a foundation far stranger than most readers assume.

The easiest words to define are concrete terms with sharp boundaries. A hexagon is a six-sided polygon, and the definition works because the category has necessary and sufficient conditions. This is the classical model of meaning, inherited from Aristotle: name the general class, add the distinguishing features, and the concept is fixed. Most words, however, do not cooperate.

Consider the, of or if. These function words refer to nothing in the world. Instead, they encode grammatical relations. Because dictionaries cannot say what of means, they describe what it does: 'used to indicate belonging,' 'used to indicate material.' That retreat from definition to description of use is revealing. In generative terms, such words belong to the functional architecture of the sentence rather than to its conceptual vocabulary, and their meaning cannot be separated from syntax.

Content words resist in different ways. Wittgenstein observed that no single feature unites everything called a game. Instead, members share overlapping 'family resemblances.' In the 1970s, Eleanor Rosch's experiments showed that categories have internal structure. People judge a robin a better example of bird than a penguin, and they verify facts about typical members faster. William Labov found that whether speakers called a vessel a cup or a bowl shifted with its proportions. It also shifted with whether they imagined it holding coffee or mashed potatoes.

Sensory words pose yet another problem. No sequence of words can convey red to someone who has never seen it, which is why dictionaries fall back on comparison: 'the colour of blood.' Each class of resistant words points to a different kind of meaning, whether grammatical, prototypical or experiential. None of them fits the tidy equation that a definition implies.

Takeaway

A definition is a model of a meaning, not the meaning itself. The words that resist definition show where meaning actually lives: in grammar, in typical cases, and in direct experience rather than in lists of features.

Every dictionary is a closed system. Each word is defined using other words in the same book, so following definitions far enough must eventually lead back to where you started. Look up big, then look up a word in its definition, and you will soon find large being defined by big. Circularity is not a flaw of careless editors. It is an unavoidable property of any finite, self-contained vocabulary.

The cognitive scientist Stevan Harnad called this the symbol grounding problem. He noted that trying to learn Chinese from a Chinese–Chinese dictionary would be an endless merry-go-round of symbols pointing to other symbols. Dictionaries work for humans only because readers arrive already holding meanings anchored in perception, action and social experience. The page does not create meaning. It relies on the meaning the reader brings.

Lexicographers manage the circle rather than escape it. The Longman Dictionary of Contemporary English pioneered a controlled defining vocabulary of roughly 2,000 common words, so learners need only a small base to decode every entry. The COBUILD dictionaries wrote definitions as full sentences, such as 'If you describe something as...', which model how a word behaves in use. The linguist Anna Wierzbicka proposed a more radical solution. She identified a small set of semantic primes, such as someone, want, know and good, which she argues exist in all languages and from which other meanings can be built.

Computational analyses of dictionary networks support the intuition behind these strategies. When definitions are mapped as a graph, a relatively small core of words turns out to be enough, in principle, to define all the rest. The circle cannot be broken, but it can be made very small. That small core may hint at what the mind must ground directly in experience.

Takeaway

Language cannot pull itself up by its own bootstraps. Every chain of definitions eventually has to touch something outside language, such as a perception, an action or a shared experience.

Many readers treat the dictionary as a legislature. If a word is 'in the dictionary', it is legitimate; if not, it is suspect. Modern lexicographers see their role differently. Their task is descriptive: to record how a language is actually used by its speakers rather than to dictate how it ought to be used. This stance follows from a basic finding of linguistics. The rules of a language reside in the minds of its speakers, not in books about it.

The method is empirical. James Murray's Oxford English Dictionary was built from millions of quotation slips sent in by volunteer readers, each documenting a word in context. Since the 1980s, when John Sinclair's COBUILD project analysed a large electronic corpus, lexicography has rested on collections of hundreds of millions, and now billions, of words. Editors examine collocations, frequency and change over time. A new sense earns an entry when evidence shows sustained, widespread use.

This approach can provoke fury. When Webster's Third New International Dictionary appeared in 1961, critics attacked it as permissive and singled out its treatment of ain't. Similar outrage greets the recognition of literally as an intensifier, even though that use has been documented since the eighteenth century. In each case, the dictionary was reporting evidence, but readers heard an endorsement.

Yet descriptive dictionaries are not neutral mirrors. Editors choose which texts to sample, which senses to list first and which labels to apply, such as informal, nonstandard or offensive. Those labels describe how usage is perceived socially, but users read them as verdicts. A dictionary's authority is therefore real but borrowed. It derives from the collective behaviour of speakers, and once published, it feeds back into that behaviour as a standard.

Takeaway

A dictionary is less a rulebook than a carefully drawn map of collective behaviour. Its authority comes from speakers, and it then returns to shape them.

Defining a word turns out to mean proposing a theory of that word, and theories can be incomplete. Function words, prototypes and sensory terms each expose a different layer of meaning that no single format can capture.

The circularity of definitions shows that language is grounded outside itself, in perception and shared experience. The descriptive turn shows that meaning is a social fact, sustained by millions of speakers rather than decreed by any authority.

The dictionary, then, is an extraordinary artefact: a public, partial model of a private mental system. Its gaps are not failures of scholarship. They are evidence about the structure of human meaning itself, separating what we can say about words from what we can only know by using them.