Sakai Mika 坂井美日’s “Historical Development of the Nominalization Construction in the Kamigata Dialect of Japanese” (上方語における準体の歴史的変化) was selected by the Society of Japanese Linguistics as one of the two best papers they published that year. Here’s the abstract, with year ranges added by me:
This paper examines two types of nominalization (zero type and no-type henceforth) found in the Kamigata variety of Classical Japanese, with an exclusive focus on argument positions. It aims to give an account of the historical development of the no-type nominalization based on the following two facts. First, the nominalizer no started to be employed as a regular means to nominalize (i.e. head a noun phrase carrying) the adnominal clause two hundred years or so after it first began to attach to adnominal clauses in the Middle Japanese. During this period, the nominalization construction was used both for the referential and event uses with no statistically significant difference in frequency. Second, the no-type nominalization started to replace the zero type first in the referential use (during the Meiwa-An’ei era [1764–1781] to the Kansei-Bunka era [1789–1818]) then in the event use (during the Bunsei-Tenpo era [1818–1844] to Taisho era [1912–1926]). These facts indicate that there is no reason to believe that no was originally a pronoun designating a person or thing. Rather, it is reasonable to assume that the morpheme was a cognate of the genitive no, which does not have any specific referent. Furthermore, it is argued that the development of the no-type nominalization over the zero type is more naturally explained by addressing the structural reanalysis that occurred in the referential use of nominalization than the loss of the distinction between the conclusive and adnominal forms as often argued in the literature. This hypothesis is supported by the data from other dialects.
In Classical Japanese, the adnominal form (rentaikei 連体形) of a verb can function as a noun all on its own. Sakai gives two examples, both from the Tale of Genji (translations below added by me):
(1) a. 花の下に歩きて散りたるを多く拾ひて [“Walking below the flowers and picking up many of those that had fallen…”]
b.(風が吹き花が) 乱れ落つるがいと口惜しうあたらしければ [“The scattering [of the flowers in the wind] being most pathetic…”
(Incidentally, (1)a is what the abstract calls “referential use” [形状タイプ], while (1)b is “event use” [事柄タイプ]. In the former case, the verb is to be understood as a modifier—“[the Z that] Ys”—while in the latter, it refers to the act itself: “Y-ing”.)
The contemporary Japanese equivalents of these sentences need a no (or similar) to nominalize the verb:
(2) a. 散ったのを多く拾って
When did this change, and why? As Sakai explains, neither question has been satisfactorily answered yet. There is a vague idea that it was because when the conclusive form (shūshikei 終止形) merged with the adnominal, the no might have been added to help distinguish the two cases, but (again relying on Sakai) that is hard to believe because there is a multiple-century time lag involved.
Sakai’s argument is that the no started being added to these verb forms by analogy with “genitive + no” constructions (e.g. Ise ga no), where the no is basically redundant. Adnominal + no then survived for a couple of centuries as a rare variant of plain adnominal until, for some reason (Sakai offers a couple of reasonable-sounding theories but no firm conclusion), people decided that plain referential-use adnominals weren’t sufficient, upon which adding a no became more common and eventually required.. Event-use adnominals lagged behind, but caught up by the early 20th century.
From this, as mentioned in the abstract, it follows that the no in contemporary uses like chitta no “those that fell” was not descended from a word that means “thing” or similar, as is sometimes theorized (based on the similarity to constructions like chitta mono “things that fell,” where mono is unambiguously a noun). Instead, it comes directly from good old genitive no, and doesn’t really mean anything.
That is, when we say chitta no, the no might feel like a (formal) noun, but etymologically speaking we’re still using a very nearly plain adnominal with only the sparsest of syntactic decoration.