Does any serious scholar really believe that some languages have no distinction between verbs and nouns?
Wikipedia pages suggest this. I studied physics, so linguistics is not my field at all.
Question What's the global difference between nouns and verbs? raises a similar topic, but the final conclusions are missing in almost each answer.
One has to be careful how the words Noun and Verb are understood, if one wants a good answer. Semanticists talk about Entities and Events, and leave Noun and Verb as formal categories, dependent on criteria of usage (can it be a subject? can it take a definite article?) instead of meaning.
Frawley (Linguistic Semantics, 1992 Ch.3 "Entities", p.62) notes that,
This is because nouns normally refer to entities, and verbs normally refer to events (including states). In other words, those definitions of noun and verb from grade school are completely backwards.
There's all kinds of linguistic evidence that most languages do distinguish between noun and verb.
But that's not evidence that categories like Noun and Verb can be attached to individual words.
At least not permanently; lexical categorization is a separate issue. Pretty much any English word, for instance, can be used as just about any open-category part of speech: noun, verb, adjective, adverb. But we have good tests for what a given word is being used as -- in a given sentence. Not generally.
However, the joker in the question is that it presupposes that there are words in a language.
Most languages do have words -- i.e, they can put word-level constituents together into larger ones.
But there are polysynthetic languages, like Eskimoan languages and Salishan languages. In a polysynthetic language there is usually a very simple root system with dozens of derivations and inflections that get added on to form even the simplest utterance. And these roots are very general, and have lots of metaphoric and idiomatic associations -- like any other language -- so their possible forms get big very fast. Both in the sense of getting longer and longer, and in the sense of there being almost infinite extensibility of words.
It says something about Skagit (Puget Salish; Northern Lushootseed) that over 3/4 of the nouns in the language start with the overt nominalizer s- --even the words for 'man' and 'woman'. They're built on CVC roots, as is most of the rest of the language. So in Skagit, while there are nouny things and verby things and you can tell the difference in a given sentence, they don't split up the descriptive labor the way we're used to, and the roots are normally neither verby nor nouny, but can be made to refer to either. Plus there isn't a really well-defined boundary between words and sentences.
So serious scholars will question the presuppositions.
Yes, some serious scholars do. (Not me -- I have no opinion.) The earlier thread here which you referred to and the references given there surely show that. Should a reasonable person believe such a thing? What's the problem? You can probably think of English sentences without any nouns, so why shouldn't all sentences of some language be like that?
Here's an example: "To err is human; to forgive, divine." No nouns.
It is hard to imagine how a human language could do without predicates and arguments, but predicates don't need to have verbs, and arguments don't need to have nouns.
Bearing in mind that noun, verb (and adjective) are formal categories, not substantive (semantic) ones, a necessarily-universal presence of the distinction would mean either (a) category membership can be deduced from the meaning of the word or (b) there are formal markers which identify category membership, and all words must have one of these formal markers. Since "eating" (as in "His eating bothers me") is formally a noun but refers to an action, not a person, place or thing, then semantically-based inference of category membership will give you the wrong results.
Possibility (b) can also be dispensed with. Criteria like "inflects for tense" or "can be immediately preceded by a definite article" don't work -- many languages don't inflect for tense, many languages don't have definite articles, and even in English, the definite article can immediately precede adverbs, adjectives, or whatever else an NP can have before a noun. Nouns vs. verbs do not differ in an identifiable way in all languages languages. The only marginally-plausible potentially-universal formal marker that could signal the noun / verb difference is the dominating syntactic category N vs. V. But there is no independent way to identify whether the lexical item [č'ƛ'aʔ] is dominated by N or V. In other words, words do not wear their formal category membership on their sleeves.
It is possible to learn in a specific language that there are nouns, verbs and adjectives. In this paper, Chung gives formal properties of Chamorro identifying nouns, versus verbs, versus adjectives. These properties, such as incorporability, cannot universally identify nouns, since not all languages have incorporation. So, it is possible that all existing human languages accidentally happen to make a learnable noun / verb distinction. But there is no reason to hold that the distinction is necessarily (by design) present in all languages, and as Chung notes is not theoretically required under Minimalist / DM assumptions.
In my philosophy of language, the one model of human language is the process. Therefore every human language has grammatical categories for nouns (objects), verbs (processes), and adjectives (results of processes i.e. states).
