Referentiality is not graded: a matter of degree. It derives from the likewise non-graded notion of natural language. It is related to Jerrold Katz’s nearly 50-year-old “effability” hypothesis (that any language can express any proposition).
Effability (also known as propositionality) cannot be proved, but it is easily refuted, with a single counter-example; yet no one has produced one so far (though there have been attempts, so far all unsuccessful).
It follows from this non-graded property of “effability” that there is really no such thing as a “protolanguage” – a “lesser-grade” language that can express some, but not all, of what can be expressed in any other language.
(A little background: a content-word, or “open class” word is a word that has a referent, whether simple and concrete, like “cat,” or complex and abstract, like “catharsis”. In contrast, a function-word, or “closed class” word is a word that performs a grammatical or logical function, like “the” or “not.”: It has a use, in forming a proposition, but it does not have a referent. Almost all the words in any language are content-words; the function words are few, and similar across languages.)
A little reflection will show that lacking a content-word in the current vocabulary of a language to refer to the referent of any content word in any other language is just a question about vocabulary – what has so far been lexicalized in a given language? It is not about differences in the language’s expressive power. If the content-word is missing today, tomorrow it is there. All you need do is to coin it, with an arbitrary new word plus a definition composed of already lexicalized content-words.
If what the new content-word refers to is important and useful, it will be adopted. It’s always easier to refer to something with a single referring content-word rather than a long verbal description (“chunking”). But for Katz’s effability hypothesis it makes no difference: (The hypothesis is not that every proposition can be expressed in every language using the same number of words!)
The connection with referentiality is that if every language can express every proposition, then the referent of any content-word can be defined (or described) in words, to as close an approximation as desired. (One can always extend a definition to cover [or exclude] more cases.) And a definition (or description) is a proposition (or a series of propositions).
Like referentiality, propositionality, too, looks like a simple property. But propositionality has profound consequences that can be shown to connect with referentiality. If someone can express – and understand – any proposition, then with propositionality, they can express and understand the definition of the referent of any content word.
A proposition is a declarative sentence with a subject, a predicate, and a truth value (True or False). There is nothing in between true and false (the law of the excluded middle): There is no truth-value between T and F; no gradation. Yes, what is true may be uncertain, or only a matter of probability. But it is only word-play to call this degrees of “truth.” (I won’t dwell on this here now.)
So propositionality inherits the all-or-none nature of statements about what is true (or not true).
Now, perhaps the most important point: propositionality and referentiality are related, but they are definitely not the same thing. Words have referents, but they do not have truth-values. “Cat” is neither T nor F, because it does not assert (propose, or predicate) anything. “That is a cat” (while pointing to a cat) does propose something, and it is either T or F. So does the proposition “a cat is a canid” (its truth value happens to be F).
So an agent that can recognize cats, and distinguish them from dogs, and can learn to approach a cat and not a dog, or can learn to look for a cat when someone says the word “cat”, or can even bring a cat toy when they want you to take them to the real cat, or can even learn to bark once if they want to see a dog or twice if they want to see a cat – none of those agents are making propositions, hence none of them are referring, not even if they are trained to identify cats and dogs by pressing successive buttons that make the sounds THIS IS A CAT or THIS IS A DOG.
If they ever could express, and mean, the proposition “This is a cat,” then they could learn to express any proposition, simply by recombinations of subjects and predicates. (Ask yourself: if not, why not? That’s Katz’s challenge in reverse!)
What this means is that to express and mean any proposition is much more than just the behavioral capacities I’ve described (recognizing, approaching, fetching, soliciting). How much more? Having the capacity to enter into this discourse with us. That’s what propositionality and reference make possible – for those who really have it.