The issue I see with the "can't express at all" view, is that, if it can't be expressed, then how do newborns learn to speak the language?
If from some sequence of sense perceptions, a child can learn to associate some word(s) with some concept, why couldn't one describe that sequence of sense perceptions in another language, and have the listener, by imagining those sense perceptions, grasp the concept?
Now, I don't want to be absolutist about that. Maybe some concepts get attached to some words through ways other than what sense perceptions pick out, somehow? Like, maybe when discussing theology or whatever, God intervenes and influences what meanings people learn for different words? (like, in a way that can't exactly be formalized and expressed in terms of math, sense perceptions, and any ideas that might be built-in to the human mind which one might intuitively associated with some combination of the previous two?)
But, outside of things like that, I would expect that meanings for words that are shared among an identifiable collection of people, can be explained in any of the most common natural languages.
(Though, maybe not so much for the meanings or aspects of meanings that are specific to one person.)
Unless there is some mechanism by which a meaning could be communicated from one person to another child-person, which can't be replicated with another language.
Now, that's all just for concepts between humans. For the Alpha-Centurion, perhaps they could have some innate ideas which they could learn words for, but which we would not learn to associate the idea with the word if we were given analogous sense perceptions, because we don't have those ideas built in to us? This also doesn't seem likely to me, but I seem to have less argument against it than I do for the same thing for the analogous thing between different human languages.
We should still be able to describe the statistics of how they use certain words together though, and how this correlates to the world, or at least, the world as described through those concepts that we can comprehend. And, perhaps we could also describe the statistics of what words they would use to describe the ways in which our description of how they use the words (including correlation with aspects of the world that we comprehend), falls short of the true meaning of the words.
There's an idea of "semantic primes", supposedly semantically irreducible concepts, that can't be defined except in terms of words that would be defined in terms of these (though, one might ask, "couldn't one pick some other collection of concepts as the base case instead?" and idk what the counterargument is), and which supposedly every natural human language has a word for each of these (though the word might not only mean one of these semantic primes/primitives, possibly having other meanings as well).
The idea goes that every word in any natural human language can ultimately be expressed in terms of these primitive concepts (of which there are supposedly like 65).
If this is true, then no idea in any natural human language would be entirely untranslatable to any other human natural language.
But, it does raise of course raise the question, "what if there was something else beyond these 65 or so, that we (humans) lack the concept of?" (which is I think similar to the question you were raising)