A fun linguistics word is
murmured voice. The wiki page has some useful information (including an interesting factoid about Gujarati and it's murmured vowels that I didn't know!).
The wiki page asserts that Hindi makes a two-way distinction between voiced and murmured nasals, which means it's asserting that Hindi does have the phonemes /mʱ/ and /nʱ/. Wiki doesn't make any assertion about murmured liquids /lʱ/. All of that said, there's no academic reference cited on the Wiki page to reference this fact, and I'm not sure this is really an accurate assessment. Let me explain my thought process (perhaps a bit different than
@marrish and
@Dib's proposed prosodic considerations).
Wiki defines murmured voice as "a phonation in which the vocal folds vibrate but are adjusted to let more air escape." In words like
tumhaaraa, I would expect that, since the /m/ is so close to the /ɦ/, when speaking at a normal pace there would in fact be an "adjustment" of the vocal folds when pronouncing the [m]. Lots of sounds exhibit assimilation with nearby sounds like this, so it would be very surprising if this wasn't true. But I don't know if anyone's hooked up Hindustani speakers to phonological devices than can measure these kinds of things. (This is probably
not the kind of thing one can introspect even if you're a native Hindi-Urdu speaker. If you can understand the question being asked, your introspection may convince you of something that might not be what you actually do...)
In any case, even assuming that that is true, that would just be saying that sometimes, the phonetic realization of a Hindi-Urdu utterance may contain sounds like [mʱ]. In other words, it would be saying that
phones like [mʱ] are possible, but that is
not the same as asserting that Hindi-Urdu has a
phoneme /mʱ/. Whether or not Hindustani has a phoneme /mʱ/ is probably more a question of parsimonious modeling: does positing the existence of such a phoneme lead to an accurate description of the language that's as simple as possible?
Let me rephrase slightly. Deciding whether certain sounds are phonemes in that language is a part of building mathematical model of the language. When you're building a model (of anything, be it a language or a physical system or whatever), you want two things: you want the model to be
accurate in that it explains all of the data you've observed, and you want the model to be
simple. In other words, you're working under Occam's razor. Which means one possible rough strategy towards model-building is: start with something that's as simple as possible, and then add complexity until you've explained all of your observations.
Of course this is overly simplistic and you may find that you have to scratch everything at some point and set up the model in an entirely different way, but... let's stick with the oversimplification for the moment. Since adding phonemes to the inventory makes a model more complicated, the question becomes: Are there linguistic observations that
force us to add murmured nasal and liquid phonemes to the phonemic inventory of Hindustani?
I suspect that the answer to this question is "no."
1. Here is a
paper by Esposito, Khan, and Hurst (EKS) that showed up when I was looking around on the internet (which discusses not only Hindi-Urdu but also Bengali). It's very related to the following comment:
The difference between an aspirate and a cluster of Cons+h is probably the length of the breathy part and also its quality I suppose, apart from the fact that no schwa should be inserted in between (like pahaaR ≠ phaaR).
Here is a brief summary of the EKS findings with respect to Hindi-Urdu. One aspect of Hindi-Urdu is that speakers often add a schwa between clusters they find problematic (a phenomenon called "schwa epenthesis"). This is why we get things like
garam in place of
garm and
dharam in place of
dharm. The decision about whether or not a particular cluster is problematic is a bit idiosyncratic (some people have no problem saying
garm), but you never get people pronouncing
bhaarii as [bəɦɑːri
ː] or
ghar as [gəɦər]. This is one line of reasoning you might use to decide that /bʱ/ and /gʱ/ are in fact single phonemes in Hindi-Urdu rather than clusters.
The EKS paper collects some data and finds that some Hindi-Urdu speakers do in fact sometimes insert a schwa between sequences consisting of a nasal plus /ɦ/ (it seems like they were specifically interested in the word
kumhaar), which suggests that speakers parse nasals plus /ɦ/ as a cluster that can be broken up by schwa epenthesis (rather than as a inseparable unit like /bʱ/).
[Interestingly, EKS finds that the time taken to pronounce an murmured stop like /bʱ/ is more than the time taken for the unmurmured /b/! I would not have expected that.]
2. We could also try to think along the lines of "Aspiration and Nasalization in the Generative Phonology of Hindi-Urdu" by Narang and Becker (NB). NB don't explicitly discuss murmured nasals at all. What they do is suggest the following line of reasoning for positing the existence of a single phoneme like /d͡ʒʱ/ in Hindi-Urdu rather than the cluster /d͡ʒɦ/.
The idea is to start with the verbal root
samajh- ("understand"), and then notice that, in forming its participle
samjhaa, we've undergone a schwa deletion. They formulate a precise and relatively simple schwa deletion rule for Hindi-Urdu that explains these morphophonemic changes, and then notice that if the
jh in
samajh- was regarded as a cluster /d͡ʒɦ/, then we would not expect the schwa deletion in passing to the participle
samjhaa. But, since we do observe a schwa deletion here, we have two options: either (a) regard the "cluster" as a single phoneme /d͡ʒʱ/, or (b) to rewrite the schwa deletion rule to deal with this issue. If we try going down route (b), we quickly find that the schwa deletion rule would become significantly more complicated because this is not just an issue with
samajh-, it's also an issue with
pighal- ("melt") and many many other words. So, their line of reasoning goes, Occam's razor forces us to veer towards (a), ie, that these murmured stops are in fact single inseparable phonemes rather than clusters.
We could try applying this NB style of reasoning to murmured nasals. In other words, the question is this: Are there any words which do undergo schwa deletion under standard morphological changes (participles, plurals, obliques, etc), but where if we analyzed them as having a phonemic representation involving a cluster /mɦ/ or /nɦ/ or /lɦ/, the NB schwa deletion rule would not operate?
I'm not able to come up with any, which of course doesn't mean that there are in fact none. But, if there are in fact none, then NB's style of reasoning doesn't force us to add murmured nasals and liquids to our phonemic inventory, so the more parsimonious solution seems to be not having them.
3. Here's another line of reasoning one might use. I don't know of a reference for this, but I'd be shocked if someone hadn't thought along these lines before... The basic starting observation is that Hindi-Urdu words rarely begin with consonant clusters. I think that most of the exceptions to this are English or
tatsam Sanskrit loans (...?), but in any case, this is particularly true of
tadbhav and
deshaj words (ie, of "Indic" origin, but not
tatsam Sanskrit loans --- which form most of the core stock words of Hindi-Urdu).
If we analyzed the first sound in a word like
ghar as a consonant cluster, we'd have a violation of this constraint. We could rewrite the rule and make it more complicated to deal with these kinds of exceptions, but these exceptions are quite numerous:
khaanaa,
bhaarii,
chhor,
jhanDaa,
ThanDaa, so, by a similar train of thought as in the NB argument above, it seems like we're led to the conclusion that maybe it's most parsimonious to add aspirated and murmured stops to the phonemic inventory.
This leads us to the following question: are there any
tadbhav or
deshaj words that begin with mh-, nh-, or lh-? If so, maybe it's parsimonious to have the murmured nasals and liquids /mʱ/, /nʱ/, and /lʱ/ as a part of our phonemic inventory. But I doubt such examples exist.
