Andreas wrote:

<< I was looking at a chart of reflexes of PIE and Proto-Semitic consonants 
other day, and noted that while, say, the ancestral velar stops are mangled
wantonly in many daughter languages, *m and *n are perserved in every 
listed (in initial position at least).

Now, is this just a quirk of these particular families, or do different sound
have differing "intrinsic" probabilities for changing?>>

Well, let's think about the possible sound change that could happen to 
initial segments:

1.) The sound could be lost entirely.   I'd say that sonorants are less 
likely to be lost in initial position, and n and m are sonorants.

2.) They could devoice, if voiced.   Nasals don't have a salient voiceless 
counterpart, so they have nothing to devoice to.   While it's entirely possible, 
it's highly unlikely in initial environment, because what would cause such a 
devoicing?   If a vowel follows it couldn't be assimilation--it'd have to be a 
very odd case of dissimilation.   The only way I could think of them 
devoicing is if they were followed by a voiceless fricative--particularly, /h/--and 
this sound change *is* attested.

3.) They could voice, if unvoiced.   They're not unvoiced.

4.) They could aspirate/deaspirate.   This ties into (2), I think.   While 
it's not impossible (I don't think), I'd say it's highly unlikely, though could 
arise with a following /h/.

5.) They could become fricatives.   This happens with mutations in Gaelic 
(/m/ -> [v]), but generally not independently.   In order for it to happen, a 
nasal, which is [-continuant], would have to become [+continuant].   This is easy 
enough for oral stops, because it only involves one feature.   However, nasals
 are *necessarily* [-continuant], which means that you'd have to drop the 
feature [+nasal] as well.   In order for this to happen, there'd really have to 
be some sort of a plausible conditioning environment, and initially is not 
likely--it's not even that likely for oral stops (though it becomes more likeley 
if preceded by a V-final word in speech).

6.) They could become oral stops (i.e., /m/ -> [b/p], /n/ -> [d/t]).   This 
isn't likely, but not impossible (I'll discuss why below).   Why is not likely? 
  Word-initially is not a conditioning environment that would give rise to a 
spontaneous dropping of nasality.   I actually can't think of one that is...

7.) They could become some other sonorant (i.e., /m/ -> [w], /n/ -> [l]).   
Now I'm not sure about /m/ -> [w], but /n/ -> [l] is *widely* attested.   Take 
French/English, for example.   Our word "level" comes from Old French *nivel 
which came from an old n- initial Latin word.   Currently, I believe French has 
niveau but we   have level.   And, in fact, it's very common for both /n/ -> 
[l] and /l/ -> [n].   This can't really be explained phonetically, but there   
have been acoustic explanations offered--namely, that the difference between 
[l] and [n] to the ear is negligible, because both are [+coronal], 
[-continuant], [+sonorant].   The only difference between the two is that one is [+nasal] 
and the other is [+lateral].   Also, taking this a step further, there are 
attested sound changes in Bantu whereby /l/ -> [d] or [r] in certain 
environments, so /n/ could in a circuitous sort of way become [d] or [r], and from there, 
the sky's the limit.

8.) They could metathesize (e.g. /ma-/ -> [am-]).   I don't have an 
explanation for why this would be salient or why not.   The cases of metathesis I know 
of never took place word-initially.   It might be interesting to look this up.

9.) They could neutralize (e.g. /m/ -> [n]).   This, in part, happened in the 
Polynesian languages with *N (cf. *N -> n in Hawaiian everywhere), but never 
with [m] and [n].   I'd say there's an acoustic reason for this, and, indeed, 
an acoustic reason why labial segments, in general, tend not to pattern with 
coronal segments.   The audible difference between coronal and labial segments 
is too great for the two ever to be confused.   This is why, for example, the 
pull chain in Hawaiian didn't affect /p/ (that chain was: (1) Glottals 
Deletion: *? -> , *h -> ; (2) Fricative Neutralization: *s -> h, *f -> h; (3) Oral 
Stop Pull Chain: *k -> ?, *t -> k *but not* *p -> t).

10.) They could assimilate to/dissimilate from the following segment.   
Again, if followed by a vowel, this is a negligible change, though it should be 
noted that *n -> J / _[i] is very common.   This becomes more interesting, 
though, if you look at languages that allow other things to follow onset nasals.   
So, for example, what about *mw, *mj, *nw, *nj?   The first two generally 
produce no affect (the former in Bantu languages, the former in Russian), but the 
second two certainly do.   *nw -> Nw is quite common (though it could quite 
possibly become mw), and I think that *nj -> J or ->Jj goes without saying.   
Now, depending on how you count initial nasals in Bantu languages, they change 
all the time--or at least /n/ does: *n -> n (before alveolars) -> J (before 
palatals) -> m (before labials) -> N (before velars) -> n_0 (before /h/--I hope 
that notation means "voiceless").   Anyway, for dissimilation, well, the nasal 
would have to lose voice, if followed by a vowel, or become [-continuant], or 
something.   Highly strange.

Anyway, there are other possible sound changes, but these are the most 
pertinent ones I can think of at the time.   A lot of it has to do with /n/ and 
/m/'s status as nasals.   However, a question to ponder might be why *N so easily 
becomes [n].