Petri pointed out that if _nai elen siluva lyenna_ was a case of elision of
_elyenna_ to _lyenna_, then _*nai elen siluv' elyenna_ was likely too. Pavel
wrote exactly the same: "Besides, if elision were to happen at all, it would
happen on the first vowel - _*siluv' elyenna_."
My guess it that they both are right and that it suggests that _lyenna_ is
not *_elyenna_ after elision of the first vowel.
To support this point of view I will quote the well known greeting: _Elen
síla lúmenn' omentielvo._ (LR:79) As we see, it's the final A that was left
out, not the initial O. Morevover, as in the case of _*siluv' elyenna_,
neither A nor O/E are stressed. It seems to me that in such cases it is
generally the final vowel and not the initial one that undergoes elision.
Maybe _'n_ is the exception (regardless of its full form, _in_ or _en_ or
Tolkien knows what...).
Pavel also noted: "It is not unlikely that elision can happen in the case of
non-identical vowels if they are, say, both front or both back and so the
elision is in effect a case of assimilation." Well, the sentences above seem
to support theory, don't they? As a matter of fact, we have A/O elisions (or
assimilation, as Pavel put it) attested in LR. Depending on the origin of
_'n_ we may also have an E/I elision, if this one comes from *_in_.
In light of this, I would tend to think that:
1. _lyenna_ lacks no initial vowel;
2. it is alright to have words starting with LY, as Petri already noted in
his original message;
3. elision sometimes happens when similar (both front or back) or identical
unstressed vowels are in contact.