Multilingual Conlang Database

It's very possible I got the connector types wrong, so I'll try to explain what the ER diagram is supposed to be.

The Lexeme table has just an ID, a lemma, and a pronunciation. It represents a single word in one of your conlangs. You could remove the pronunciation if the Romanization makes it clear how to pronounce the word. You could also add things like usage notes and inflection paradigms.

The Sememe table is a list of semantic units, ideas, or single senses of a word. It has the obligatory ID, the sense itself, and the sense's part of speech. Examples would be the concrete noun “chair”, or the action verb “run”, as well as more abstract fare like a noun meaning “the head of an organization” or the noun “an instance of running” as in “I went for a run today”. You could also add a semantic domain field to put the idea in a broader context. “Chair” could have the domain “furniture” for example.

A lexeme can have multiple senses, a one-to-many relationship, so a Senses junction table links a Lexeme to one or more sememes. The Senses table has the Lexeme ID, the Sememe ID, and a rank indicating how salient that particular sense of the word is, with lower numbers being more salient.

For example, say your language's word for “chair” is “kudi”. As in English, kudi can mean either the physical piece of furniture as well as the metanymous meaning of “head of an organization”. And like English, if a speaker of your conlang hears “kudi” out of context, they're likely to think of the item of furniture. So kudi shows up twice in the senses table, paired with “chair” it has a rank of 1, and with “head of an organization” it has a rank of 2.

To represent etymologies, cognates, and derivatives there is a Derivatives table. It has a source word column and a derived word column, both pointing to a lexeme id. There's also a column for an explanation of the relationship. For example, say your lang's word for mushroom, “kui”, derives from a dialectal variant of the standard “kudi” for chair, with the connection being similar to “toadstool” in English.

If you're only working on one language in isolation, you can stop there, but here's the secret sauce. While there's only one Sememe, Senses, and Derivatives table, you can have multiple Lexeme tables, one for each of your conlangs. Let's call the first language lang A, and the second lang B. Lang A has “kudi” for chair, and lang B, a closely related language, has huti. Both kudi and huti point to “chair” in the senses table, so you can translate kudi to huti when translating from lang A to lang B. Not only that, but kudi and huti are paired in the derivatives table, indicating the two words are cognates.