What Nobody Tells You About Compiling Dictionary Entries
Lexicography is not a straightforward job of looking up words and copying definitions. Anyone who has actually worked on a dictionary knows that the real work happens in the gaps between entries, in the decisions that never make it into the final product. I spent eight years on a commercial learner's dictionary project before moving to a descriptive corpus-based reference work. The difference in approach is the difference between writing a textbook and writing a manual for people who have to fix what breaks when real users actually pick up the thing. The dark side starts with scope creep, which lexicographers usually call "entry proliferation." A dictionary that once targeted three thousand headwords for intermediate learners slowly expands to eighteen thousand when someone in marketing decides the market has shifted. I watched a project balloon from a manageable six-month compilation cycle to a three-year grind because each new client request added a category we had originally excluded. The budget was fixed. The deadline was fixed. The word count was not. Entry proliferation is the term for this, and it is not just about adding more words. It is about the cascading rework. Every new entry needs pronunciation guides, usage labels, example sentences, cross-references, and sometimes grammatical information for verbs that did not exist in the original brief. A single entry in a well-formatted learner's dictionary averages four hundred to eight hundred words of editorial content. At six entries per hour for a mid-level lexicographer, you are looking at roughly twenty-four hundred to forty-eight hundred words of output per hour of actual writing time. That does not include the review cycles, the fact-checking passes, or the formatting adjustments that happen when the typesetter finally returns the galleys.
Here is something most people do not understand about the process. Definition writing is the part that takes the longest, and it is the part that sounds the easiest. You might think defining a word like "run" is simple. It is not. "Run" has forty-seven distinct senses in the Oxford English Dictionary. A learner's dictionary might trim that to twelve, but choosing which twelve to include and how to phrase them without circularity requires actual decisions. I spent three days on a single entry for "set" because every definition I wrote contained the word "set" or a synonym that required its own definition, creating an unavoidable circular reference that tripped up the editorial review system. The workaround I used was to build a dependency map first. Before writing a single definition, I listed every term that would appear in the definitions of that entry and checked whether those terms were themselves headwords or could be defined using a closed subset of high-frequency vocabulary. This reduced the average revision cycle from four passes to two passes and cut the definition-writing phase by approximately forty percent. The initial mapping took about an hour per entry, but it saved roughly two to three hours downstream. For a project with six thousand entries, that is a difference of six thousand hours versus eighteen thousand hours of editorial time. Another problem that does not make it into the trade journals is the issue of usage label inflation. When editors are unsure about how widely a word is used, they tend to add labels like "informal," "colloquial," "slang," "offensive," or "regional" as a hedge against criticism. Over time these labels become the dominant feature of the entry rather than the definition itself. I worked on a regional dictionary project where nearly sixty percent of entries carried at least one usage label, and half of those labels were based on editorial intuition rather than corpus evidence. The corpus we had access to was a fifty-million-word collection of transcripts from three provinces, which is not sufficient to make confident claims about national usage patterns.
The fix was to require a minimum corpus frequency threshold before any usage label could be applied. Words that appeared fewer than fifteen times per million tokens in the relevant register could not carry a label unless external linguistic fieldwork confirmed the pattern. This eliminated approximately thirty-five percent of the usage labels in the final draft and forced a harder conversation about what the dictionary actually claimed to represent. Some stakeholders pushed back. They wanted the labels because they made the product feel more comprehensive to buyers. I explained that a label without evidence is worse than no label, and the compromise was to move unsupported usage claims into the editorial notes rather than the main entry. Frequency data is another area where the practice diverges sharply from what publishing marketing materials suggest. Corpus frequency does not equal entry importance. A word might appear frequently in academic papers but zero times in spoken conversation, or vice versa. The decision of which corpus to weight more heavily is not a technical one. It is a decision about what the dictionary is for. I once saw a project leader insist that a technical glossary should be built from a general corpus because the technical corpus was "too narrow," which resulted in an entry list that reflected frequency in casual blog comments rather than frequency in the domain it claimed to serve. The book sold poorly to its target audience and was returned at a rate of roughly twenty-two percent, which is well above the industry average for reference works. The cross-referencing problem is worse than it looks. Every internal link in a dictionary entry is a potential point of failure. If entry A references entry B and entry B is later removed or renamed, the link breaks. In a print dictionary this means a dead reference that the reader encounters randomly. In a digital dictionary this means a broken hyperlink that triggers a support ticket. I worked on a transition from print to digital where we discovered that approximately eight percent of our cross-references were broken after a minor editorial revision cycle. The fix was to implement an automated reference validation script that ran after every revision and flagged any entry that contained a link to a non-existent or renamed headword. This took about twelve hours to build and reduced the reference error rate to below one percent in subsequent cycles.
Get the Full Details
Pronunciation notation is another area where the gap between theory and practice is significant. IPA transcription assumes a standard variety, but most dictionaries serve speakers of multiple varieties. A British English pronunciation guide is not useful to a learner whose exposure is primarily American English, and vice versa. I once compiled entries for a bilingual dictionary where the pronunciation section had to accommodate three distinct accent traditions for the same headword. The solution was to provide primary and secondary pronunciations with clear labels, but this tripled the typesetting complexity for that section and required a separate editorial pass for each variant. The additional cost was roughly eighteen percent of the total production budget for that volume. Etymology is perhaps the most unreliable section of a typical dictionary entry, and most readers never realize it. Historical linguistics is a speculative discipline at best. Claims about word origins often rest on fragmentary evidence and competitive hypotheses that scholars still dispute. I encountered an entry where the etymology section cited a tenth-century Old English source that may or may not have existed in the form described, and the medieval French cognate was drawn from a nineteenth-century source that modern scholarship has largely abandoned. The entry was accepted as-is because the etymology editor had moved on to a different project, and there was no one left to catch the error before publication. The practical consequence of this is that etymology sections should be treated as informed hypothesis rather than established fact. A well-written etymology acknowledges uncertainty where it exists. Most commercial dictionaries do not do this because uncertainty does not sell. The workaround in my current project is to append a brief "scholarly debate" footnote to entries where the etymology is contested, which adds roughly forty words per affected entry but increases the accuracy rating of that section from an estimated sixty percent to approximately eighty-five percent based on external peer review.
Sense ordering is another decision that appears mechanical but is actually deeply interpretive. Should the oldest meaning come first, or should the most common contemporary meaning come first? Historical dictionaries put the oldest sense first. Learner's dictionaries often put the most frequent sense first. Both approaches have legitimate rationales, and both alienate some segment of the readership. I once worked on a hybrid project that attempted to order senses by a combination of historical priority and corpus frequency, which resulted in a structure that satisfied no one and required three separate editorial reviews to resolve the internal contradictions. The final compromise was to use a two-tier system: primary senses ordered by frequency for the target audience, with a historical appendix that listed senses in chronological order for readers who wanted that information. This added roughly two thousand words to the front matter but reduced the confusion rate in reader feedback by an estimated sixty percent. The additional front matter increased the page count by approximately fourteen pages, which had a marginal impact on production cost but a measurable impact on perceived comprehensiveness among evaluators. Example sentence creation is a skill that most people underestimate. A good example illustrates the sense without introducing ambiguity, cultural bias, or grammatical complexity that distracts from the target construction. A bad example does all three. I spent two days on a single example sentence for the phrasal verb "give in" because every draft either implied a moral judgment about surrender, used a scenario that was culturally specific to a demographic that represented less than eight percent of the target market, or contained a subordinate clause that obscured the grammatical pattern the entry was supposed to illustrate.
The solution was to write a set of twenty candidate examples and run them through a readability filter that flagged complex syntax, culturally loaded vocabulary, and ambiguous referents. The remaining six candidates were then tested with a panel of fifteen target-language learners who selected the example they found clearest. This process took approximately forty-five minutes per entry but reduced the example-related revision cycle from an average of three passes to one pass. For a project with six thousand entries, that is a savings of roughly twenty-four thousand minutes of editorial time. The business side of lexicography is where most projects fail, and it is rarely discussed openly. Dictionary projects are expensive. A medium-quality learner's dictionary with three thousand entries and full editorial review typically costs between four hundred thousand and nine hundred thousand dollars to produce, depending on the scope of corpus work, the number of accent variants, and the complexity of the cross-referencing system. The return on investment is slow. Dictionary sales have declined by approximately thirty-four percent over the past decade as free online resources have replaced paid reference works for casual lookup needs. The publishers who continue to produce printed dictionaries are those who have secured institutional contracts, educational adoption deals, or government grants. A dictionary adopted by a national education ministry for use in public schools can generate revenue for fifteen to twenty years with minimal additional production cost. A dictionary sold through retail channels faces a much harder path. I have seen three major reference publishers exit the print dictionary market entirely in the past five years, citing unsustainable margin compression and declining institutional demand.
Digital transition is the obvious alternative, but it introduces problems that print did not have. Digital dictionaries require ongoing maintenance, server infrastructure, and a content management system that supports version control and collaborative editing. The upfront cost of building a proper digital reference platform is higher than printing a single edition, but the marginal cost of updating an entry digitally is near zero compared to the cost of a revised print edition. The break-even point for a project of moderate scale is typically three to five years, depending on subscription revenue and the frequency of required updates. One specific digital problem that deserves attention is the handling of sensitive content. Dictionary entries for words related to violence, sexuality, or controversial political topics can generate reader complaints, media coverage, and in some cases legal scrutiny. I worked on a project where an entry for a racial slur triggered a formal complaint from a civil rights organization within forty-eight hours of launch. The entry was accurate, properly contextualized, and necessary for a reference work of this scope, but the organization argued that publishing the headword itself caused harm regardless of context. The decision to retain or remove the entry was made by senior editorial staff after legal review, and it took approximately seventy-two hours from complaint to resolution. The lesson from that incident is that digital publication changes the risk profile of lexicography in ways that print did not. A printed dictionary sat on a shelf. A digital dictionary is searchable, citable, and instantly distributable. An entry that takes months to research and write can generate controversy in hours. The editorial team needs a clear content policy that distinguishes between descriptive accuracy and editorial endorsement, and that policy needs to be communicated to legal, marketing, and executive leadership before a project launches, not after a complaint arrives.
Here is a practical checklist for anyone considering a dictionary project, based on what I have learned from watching three projects succeed and two fail: Define the scope before you define the words. A dictionary is a tool with a purpose. If the purpose is unclear, the entry selection will be arbitrary, and the arbitrary entry selection will alienate the readership the project was designed to serve. I have seen projects abandon their original audience mid-development because the initial brief was too vague to sustain coherent decision-making. Invest in corpus infrastructure before investing in entry writing. A well-structured corpus with documented provenance and searchable metadata will save approximately two hundred to four hundred hours of editorial research time per thousand entries compared to relying on individual lexicographers to verify usage patterns from secondary sources. The upfront cost of corpus development is significant, but the downstream savings are real and measurable.
Hire domain-specific reviewers, not just general lexicographers. A general lexicographer can write a competent entry for everyday vocabulary. A general lexicographer cannot reliably write an entry for specialized technical terminology without input from a domain expert. I have seen projects lose credibility in specific communities because the terminology was accurate to general usage but inaccurate to professional usage within the target domain. Plan for the revision cycle before publication, not after. A dictionary that is not updated is a dictionary that becomes obsolete. The standard revision cycle for a living reference work is three to five years, depending on language change velocity and corpus update frequency. Budget for revision from the beginning. Projects that treat revision as an afterthought typically publish a product that is already outdated within eighteen months of release. Accept that some decisions will be wrong and build in the mechanism to correct them. Every dictionary contains errors. The question is not whether your dictionary will contain errors but whether your process allows those errors to be identified and corrected efficiently. A digital platform with version control and a public correction submission system is infinitely preferable to a print edition with an errata sheet that nobody reads.
The field is changing. Machine translation, large language models, and automated sense induction are already being used to draft entries at scale. The output is frequently adequate for high-frequency vocabulary and rarely adequate for low-frequency vocabulary or ambiguous polysemous entries. I have reviewed automated drafts that produced structurally correct but semantically inappropriate definitions for roughly twenty-two percent of entries in a test set of five hundred low-frequency technical terms. The error rate dropped to approximately nine percent when a human lexicographer reviewed and corrected the automated output, but the time savings were still significant: the automated draft reduced the average entry completion time from four hours to approximately one hour and twelve minutes. The practical reality is that AI-assisted lexicography is now a standard component of production workflows in most commercial projects, but it is a tool for drafting, not a substitute for editorial judgment. The entries that require the most human oversight are the ones that matter most to readers: the ambiguous entries, the entries with contested usage, the entries that sit at the intersection of multiple semantic domains. Those are the entries that automated systems struggle with most, and those are the entries that define the quality of the final product. If you are considering entering this field, the practical advice is straightforward. Learn corpus linguistics. Learn IPA and at least one alternative phonetic transcription system. Learn basic bibliographic management and reference checking. Learn to write definitions that are circularity-free, which is harder than it sounds. And learn to accept that the work you do will rarely be noticed by the people who use it, which is both the greatest professional satisfaction and the greatest professional frustration of a career in lexicography.
The dictionaries that survive are the ones that are useful. The dictionaries that are useful are the ones that are accurate. The dictionaries that are accurate are the ones whose editors understood that accuracy is not a destination but a process, and that the process never truly ends. Every revised edition is an admission that the previous edition was incomplete. The best lexicographers are the ones who can write with that admission at the back of their minds on every single entry they produce.
