How Are Dictionaries Made? Recording a Language That Will Not Hold Still

The central resource is a corpus, a very large collection of text and transcribed speech assembled to be representative of the language, running into billions of words and drawn from newspapers, fiction, academic writing, correspondence, broadcast and the web. Software then produces concordances, listing every occurrence of a word with its surrounding context, so that a lexicographer looking at a screen of several hundred real uses can see which senses exist, how common each is, what words habitually appear alongside it, which grammatical patterns it enters and whether it is formal, technical, regional or rude. Before computers this was done by hand on paper slips, and the Oxford English Dictionary's first edition was built from around five million such slips contributed by volunteer readers over decades, who were asked to copy out sentences illustrating words and post them in. That project's most prolific contributor, William Chester Minor, sent in thousands of quotations from a secure psychiatric hospital where he was detained after killing a man.

The parts of an entry each require a separate set of decisions:

• •The headword and its variant spellings, which requires deciding whether two forms are one word or two
• •Pronunciation, in a phonetic alphabet, and the choice of which accent to represent as the default is itself contested
• •Etymology, tracing the word back as far as the evidence supports and no further, with honest admissions of uncertainty where the trail stops
• •Sense division, which is the hardest part, since meaning is continuous and an entry must cut it into discrete numbered senses that no speaker actually has in their head
• •Order of senses, either historical, as in the Oxford dictionary, where the oldest meaning comes first, or by frequency, as in most learner dictionaries, and the two produce very different-looking entries
• •Citations showing the word in use, with the earliest known example dating its entry into the language
• •Labels for register, region, currency and offensiveness, each of which is a judgement that will be argued with

Samuel Johnson began his dictionary of 1755 intending to fix the language and concluded in his preface that the attempt was futile, comparing a lexicographer trying to arrest change to someone lashing the wind. The modern profession is descriptive by default, which regularly produces public anger, since a dictionary that records a disputed usage is accused of endorsing it. The clearest case is Webster's Third New International Dictionary of 1961, which dropped most usage labels, included words that critics considered illiterate, and defined ain't without condemnation; the reaction was ferocious, with reviewers accusing it of abandoning standards and one magazine calling it a scandal. The dictionary was right about how English was being used and it also demonstrated that readers want guidance as well as description, which is why many dictionaries since have added usage notes explaining that a form is disputed without refusing to list it.

The criteria for inclusion are evidential rather than aesthetic. A word needs sustained use over a period, typically several years, across a range of independent sources, in contexts where its meaning is not being explained, which shows that readers are expected to know it. That is why a word coined by a newspaper and repeated within one industry does not qualify, and why slang that survives a decade does. Words are rarely removed from the largest historical dictionaries, which aim to record everything the language has ever contained and mark obsolete terms as such, while smaller print dictionaries must cut to make room and have been criticised for dropping nature words in children's editions in favour of technology terms. The other direction is more active than ever, since the web supplies evidence in quantities that make it possible to track a word's spread within months of its appearance, and major dictionaries now publish updates quarterly rather than waiting for a new edition.

Historical dictionaries operate on timescales that are hard to credit. The Oxford English Dictionary was proposed in 1857, began publication in fascicles in 1884, and was completed in 1928, taking seventy years and outliving several editors including James Murray, who worked in a corrugated iron shed in his garden and died with the letters past T unfinished. Its second edition appeared in 1989, and a full revision begun in 1993 is still in progress, published online and expected to continue for many years. The Grimm brothers began a German dictionary in 1838 that was finished in 1961, by which time the early entries needed revising. A dictionary of ancient Egyptian begun in Berlin in 1897 and a Sanskrit project in Pune have run for over a century each. The lesson these projects illustrate is that a complete description of a living language is a moving target: by the time the last entry is written, the first is out of date, which is why the online form, continuously revised and never finished, suits the work better than a book ever did.

Lexicographers work from a corpus of billions of words of real usage, reading concordances that show every occurrence in context, and they describe what words are being used to mean rather than granting permission. An entry requires decisions about variants, pronunciation, etymology, how to divide continuous meaning into discrete senses, and what to label as informal or offensive. Inclusion depends on sustained independent use over years, the descriptive approach caused a public scandal in 1961, and the great historical dictionaries take so long that their earliest entries need revising before the last are written.