How a Bilingual Dictionary Is Actually Made

A dictionary entry looks like a simple statement of fact: this word means that. It is the output of a long chain of decisions, most of which are invisible in the finished product and any of which could reasonably have gone another way.

Deciding what a word is

Before anything can be listed, the boundaries have to be drawn.

QuestionWhy it is hard
Is a phrasal verb one entry or several?give up, give in, give away — separate meanings
Are compounds separate words?German and Finnish build them freely
Do inflected forms get entries?Turkish or Finnish would need millions
Is this one word with two senses, or two words?bank as riverbank and bank as institution

For agglutinative languages the last two questions are decisive. A Turkish verb generates an enormous number of surface forms, so dictionaries list stems and expect the user to strip suffixes — which means the dictionary is only usable by someone who already knows how.

Sense division is judgement, not measurement. How many meanings does run have? Major English dictionaries give figures ranging from dozens to several hundred, and they are all describing the same word. The number depends entirely on how finely you slice, and there is no fact of the matter that settles it.

The corpus

Modern lexicography for well-resourced languages runs on corpora — large collections of real text and transcribed speech, searchable for frequency and context.

What a corpus showsWhy it matters
How often a word occursdecides inclusion and sense order
What words it appears withcollocations belong in the entry
Which registers use itformal, slang, technical labels
Which senses dominatethe first sense should be the common one
Whether it is currentcatches words that have fallen out of use

Corpus evidence routinely overturns intuition. Lexicographers frequently find that the sense they would have listed first is not the one people actually use most.

Bilingual dictionaries have an extra problem

A monolingual dictionary defines. A bilingual dictionary must find an equivalent, and often none exists exactly.

SituationApproach
Clean one-to-one matchgive the equivalent
Source word is broaderlist the several target words and when each applies
Source word is narrowergive the general term plus a qualifier
No equivalent at alla short explanation, not a fake single word

The last row is where cheap dictionaries fail. Supplying a single misleading word is worse than admitting the gap, because the user has no way to know the equivalence is false.

Under-resourced languages change the method

Well-resourcedUnder-resourced
Large corpus availableLittle or no digital text
Frequency data drives decisionsSpeaker consultation and fieldwork
Established spelling standardCompeting or unsettled orthographies
Prior dictionaries to build onSometimes a first attempt
Tone and grammar encoded in textOften absent, must be supplied

For a tonal language whose standard orthography omits tone, the dictionary is the only place a learner will encounter the tone at all. The same applies to verb stem alternations, noun animacy and classifier assignment — grammatical information that ordinary text does not carry and that a dictionary must therefore include per entry.

Why this argues for offline

Dictionaries for small languages are painstaking to build and, once built, are often the single best resource available — better than automatic translation, which has no data to work from.

That makes their availability a practical matter. A resource that requires a connection is unavailable in exactly the settings where these languages are most used: rural areas, community classrooms, and among speakers who do not maintain a data plan. A database on the device is not a convenience feature in that context; it is the difference between the work being usable and not.

45 dictionaries, built to be used offline

NDT Studio publishes free offline dictionaries across 45 language pairs, including languages automatic translation serves poorly.

Open the dictionary app

Frequently asked questions

How do lexicographers decide which words to include?

Mostly by frequency in a corpus — a large body of real text — balanced against coverage of everyday domains. A general dictionary aims to cover what an ordinary user will actually look up, which is not the same as the most frequent words, since the most frequent are usually already known.

Why do dictionary entries order senses differently?

Some order by frequency, putting the most common sense first, which suits learners. Others order historically, tracing how the meanings developed, which suits scholarly use. Neither is wrong, but the choice substantially changes how useful the entry is for a quick lookup.

Why are dictionaries for smaller languages harder to make?

Because the corpora that modern lexicography depends on may not exist. Entries must be built from fieldwork, existing wordlists and speaker consultation rather than from frequency data, which is slower and requires judgement that automatic methods cannot supply.