I hope I'll be forgiven for reviving an old thread! A fellow forum member linked me to this thread a little while ago, and after digging around a bit, I had some references and some thoughts that I felt may be marginally worth adding to the discussion.
In case people haven't seen it before, there's an old doctoral dissertation of Misra's titled "Historical Phonology of Modern Standard Hindi" (dated 1967). Certainly not everything discussed in this dissertation is original to this dissertation, but it does at least give a lot of references to other things in the literature, which is nice. In particular, there's substantial discussion in that dissertation about the development of the voiced retroflex stops [ɖ⁽ʱ⁾] in Indo-Aryan (which apparently did not exist in Proto-Indo-Iranian). There's also a slightly different theory regarding these stops proposed by S. K. Chatterji in "Indo-Aryan and Hindi" [p. 62].
Misra also discusses the retroflex flaps [ɽ⁽ʱ⁾]. It seems the intervocalic realization of the phonemes /ɖ⁽ʱ⁾/ as [ɽ⁽ʱ⁾] dates quite far back: there's seems to be some evidence (the finding of which is attributed to S. K. Chatterji's tome on the origin of Bengali) that Middle Indo-Aryan words involving an intervocalic /ɖ/ were transliterated into Greek using the letter rho, and some people seem to believe that this realization may even go back to the Old Indo-Aryan period [cf. Misra p. 165].
In any case, it seems that word-final (or maybe, more precisely, morpheme-final) short vowels were lost at some point in the transition from "Old Hindi" to modern times, and this resulted in the phoneme /ɖ⁽ʱ⁾/ being realized as [ɽ⁽ʱ⁾] not just intervocalically, but also between a vowel and the end of a word(/morpheme).
One thing that isn't really made clear in Misra's dissertation is what kind of a phonological rule the rule ɖ⁽ʱ⁾ → ɽ⁽ʱ⁾ was. In American English, there's a intervocalic t → ɾ phonological rule, but I think it's what the lexical phonologists would call "post-lexical," in that the rule pays no attention to the morpheme structure of an utterance (for example, the phoneme /t/ in "they eat" is not realized as a flap, but the /t/s in "they eat ice-cream" and "butter" are).
I kind of suspect that the intervocalic ɖ⁽ʱ⁾ → ɽ⁽ʱ⁾ phonological rule in Indo-Aryan may *not* have been post-lexical (ie, that this rule *would* have been sensitive to the morpheme structure of an utterance). If it had been post-lexical, it would mean that people at some point may have realized the phoneme /ɖ/ in phrases like
uskaa Dar as a [ɽ]. This feels fairly surprising given my personal experience with modern Hindi-Urdu phonology, but I'm not really sure how one might figure out what the historical situation was like... Since the written record isn't completely reliable when it comes to phonetic realizations, this may just be an unfalsifiable musing.
Apparently the words
niDar and
niDhaal are attested in writing as far back as the 1400s [Misra, p. 218]. (I found
niDar in poetry of Kabir, which explains the dating of that word. I don't know how
niDhaal was dated.) Another word that comes to mind along these lines is
suDaul, but I also don't know how to date this word.
Misra argues that words like
niDar and
niDhaal indicate that the retroflex flaps [ɽ⁽ʱ⁾] had become distinct phonemes as early as the 1400-1500s, well before loans from English had had any substantial effect on the language. I find this a slightly tenuous argument, since
niDar and
niDhaal evidently involve prefixation, as does the word
suDaul. If we just have the data of just these words, it could be that the ɖ⁽ʱ⁾ → ɽ⁽ʱ⁾ phonological rules were sensitive to the morpheme boundary, or that they applied prior (ie, at a "deeper stratum") than the morphological rules involving prefixation of the morphemes /nɪ/ and /sʊ/. (It's worth pointing out, in Misra's defense, that lexical-phonological models didn't exist at the time of his dissertation.)
I can't think of any Hindi-Urdu words outside of English loans that involves [ɖ⁽ʱ⁾] between a vowel and either another vowel or the end of a word, and where this situation isn't the result of prefixes.
Now the aspirated stop [ɖʱ] doesn't ever occur in English loans as far as I'm aware, so it seems to me that there *may* still be an argument to be made that there is only a single phoneme /ɖʱ/ and that [ɖʱ] and [ɽʱ] are two allophonic realizations of this phoneme. One could perhaps envision a rule that roughly says that, the phoneme /ɖʱ/ is realized as [ɽʱ] whenever it occurs in a sequence of the form VɖʱW, where V is a vowel, W is either a vowel or the end of a morpheme, and these is no morpheme boundary between V and ɖʱ...? (This is definitely a falsifiable hypothesis, since it's a completely synchronic assertion, and I'd love to hear about any counterexamples people are able to find to this!)
On the other hand, English loans have definitely eroded the complementary distribution of [ɖ] and [ɽ], so it's probably fair to say that these two sounds are distinct phonemes now, even if they weren't distinct phonemes 600 years ago. (That said, I do remember being half-asleep once on a drive from Delhi to Chandigarh and looking out of the car window and seeing a sign where "road" was transliterated as रोड़. Maybe I dreamt it, maybe it was just an unintentional spelling error, but maybe the phonemicization isn't quite complete yet for some speakers...)
Another thing worth pointing out regarding the retroflex flap [ɽ] is that there also seems to be some interplay between [ɽ] and [ɾ]. For example, I've at least seen some authors (eg, Nirmal Verma) write झरना (
jharnaa) for the verb that I'm used to hearing as [dʒʱəɽnaː], which suggests that there may be some kind of dialectical variation. And then there's perhaps a somewhat different kind of interplay suggested by the word
paiRii ("step," eg of a ladder), which apparently derives from
pair ("foot").
EDIT: *I am thinking of the word "akRan". Not sure whether it follows the original distribution, or whether it is in analogy to "akaRnaa".
I think that the "analogy to
akaRnaa" does explain things quite well! There's a reasonably productive derivational suffix with phonemic representation /ən/ that suffixes onto verbal roots and turns them into nouns [1]. This suggests that
akRan might be analyzed as having a morphophonemic representation /əkəɽ#ən/, where /əkəɽ/ is the verbal root and # denotes a morpheme boundary.
Now I think that there's some paper out there where the author proposes a schwa deletion rule where /ə/ gets deleted in sequences of the form VCəC#V (where V denotes vowels, C consonants, and # a morpheme boundary) [2]. I'm having trouble finding the paper now, but in any case this schwa deletion rule applies perfectly to /əkəɽ#ən/ and yields the expected surface representation [əkɽən]. Similar analyses also explain the surface representations of
ragRan and
pakRam-pakRaaii [3].
Probably analyzed in this way, one could propose some morphophonemic restrictions on the consonant clusters in which the flap /ɽ/ can be involved...
-----
[1]: There are lots of examples of this suffix in action:
dhaRkan,
phislan,
rapTan,
saRan, ...
[2]: This schwa deletion rule explains the schwa deletion that happens in a number of morphological processes aside from suffixation of /ən/, like the one in going from
saRak to the plural
saRkeN, or the one in going from
saraknaa to the causative
sarkaanaa, ...
[3]: My proposed morphological breakdown for
pakRam-pakRaaii (the game of "tag") would be /pəkəɽ#ən#pəkəɽ#ɑːiː/. The derivational suffixes /ən/ and /ɑːiː/ both turn verbal roots into nouns. We then apply the schwa deletion rule mentioned above, and a rule involving nasal assimilation, and we get the expected the surface representation [pəkɽəmpəkɽɑːiː].