On Fri, 6 Jun 2014 01:14:03 +0100, Jeff Daniel Rollin-Jones <[log in to unmask]> wrote:

>Whilst I'm on the subject, whilst there are certainly languages with uncountable nouns, and arguably some in which no nouns are countable (I'm thinking of those in which nouns require numeral classifiers to be counted, such as Chinese and Japanese), does anyone know of a language in which all (or almost all) nouns are countable? There are marginal cases of pluralised uncountable nouns in English (waters, breads [types of bread]), but you can't say "two waters" or "two breads").

WALS cited an example like this in chapter 34, though without using the countability terminology:

| Finally, in many languages not all inanimates are treated alike: nouns
| denoting nondiscrete masses such as 'sand', 'salt', 'water', 'milk' often
| lack plurals even in languages that otherwise obligatorily mark the plural
| of inanimates (e.g. in English). It is possible that this case is somewhat
| different from the non-occurrence of plurals in discrete nouns, because
| English speakers find it difficult to imagine what the semantic difference
| between _milk_ and _milks_ would be, especially if we leave aside the sort
| reading ('different sorts of milk').
| However, there are quite a few languages in which speakers find it easy to
| form and interpret plurals of nondiscrete nouns, for instance in Evenki
| (Tungusic; eastern Siberia), where such plurals are said to denote a large
| quantity or several pieces (e.g. _ulle_ 'meat', _ulle-l_ 'a lot of meat,
| pieces of meat'; Nedjalkov 1997: 190).

As for how many languages have a dual, I was hoping Corbett's book _Number_ from the Cambridge Textbook in Linguistics series might say, but I didn't turn it up in the segment Google Books gave me.  Corbett does at least give plenty of fine examples of languages with elaborated number systems, as many as five values, in Section 2.2.  

There're even natlang attestations of a special greater plural with the sense 'all Ns', like the South Omotic language Hamer, which until today was one of those features I'd only ever seen in conlangs.  Indeed, Corbett actually cites Quenya for this feature in a footnote -- though the forms he gives can only be Sindarin!  However, the "nullar number" remains un-ANADEWed in his survey.