I have been looking for what I believe would be a very valuable resource in language construction. Ideally, this would be a full-size dictionary, where every description is a short list of words that approximate the meaning.
Ex:
Sleep: Still, Human, State, Noun.
Or something like this? Seems like this could be a good guide to creating more complex portmanteau words from a smaller base.
You are probably thinking about WordNet, where a synset is a set of words with exactly the same (not just approximate) meaning. But WordNet is much more than this, it also contains informations about relations of hypernym, hyponymy and more...
The original Princeton WordNet was created for English, but there are WordNets (of different sizes, quallity and level of usefulness) for many different languages.
Modern take on the idea is the approximation of semantic closeness in word embeddings models, which has the advantage that it needs "just" a big corpus and no human annotation (and a disadvantage that lacking human proofreading, it contains a lot of errors). See e.g. here (disclaimer: I am the author of that webpage and models).
Looking for a controlled vocabulary to bootstrap the semantics of a conlang brings Natural Semantic Metalanguage (NSM) by Anna Wierzbicka and other authors to my mind. I'm not sure how far any language description in NSM is really worked out and written down, but it may give you a starter for thought.
