Piotr Gasiorowski wrote:
>> [Sean:] I think this is an innovation in Gk, IA.
>
>
> No, it's a well-known PIE pattern of compound stress, but I'm short of
> time now and can't develop the argument. I'll be back on Wednesday, and
> then we can discuss this further.
Abbreviation:
M1/2 -- first/second member (of a compound)
Here's a bird's eye view of PIE compound stress (ignoring a number of
minor problems which may be tackled later):
PIE exocentric compounds were usually accented on M1. So were
endocentric compounds if M1 was a noun, adjective, adverb or
preposition, and M2 was a basic noun, an adjective in *-to-/*-no- (the
type of Ved. sú-kr.ta-, devá-kr.ta-), or a verbal noun in *-ti-. This
includes compounds (of either type of centricity) with M1 = *n.-, such
as *n.'-p&2to:r 'fatherless' or *n.'-g^Huto- 'not offered'. The pattern
is supported not only by Greek and Vedic, but also by Germanic, where
the accentual difference is sometimes visible in forms related via
Verner's Law, such as free-standing *le:þa- 'land (as property)' vs.
compounded *unle:da- 'poor' < *léh1to- vs. *n.'-leh1to- 'poor'.
However, if a thematic adjective was based on a compound treated as a
unit, the adjectivising thematic vowel was perhaps accented already in
PIE: Skt. dvi:pá- 'island' < *[dwi- + h2p-]-ó-.
If M2 was a thematic agent noun, and M1 was a noun governed by the
corresponding verb, the accent was on M2: *dru-tomh1ó-s 'wood-cutter'.
Vedic has this pattern also when M2 is a root noun, but the evidence of
Greek partly contradicts that of Vedic, and since the vocalism of M2 is
usually nil, the Vedic pattern is probably an innovation (on the analogy
of the highly productive *-tomh1ó- type).
Jens Rasmussen ("Betonte Schwundstufe im Indogermanischen...". In: W.
Smoczyn'ski (ed.), Kuryl/owicz Memorial Volume, Kraków 1995 [1996],
93-100) argues that the reduction of M1 is _not_ due to absence of
stress in (pre-)PIE but to a universal tendency of compounds: they lose
their phonetic substance because they contain two lexical roots squeezed
into the time-frame of a single word.
One additional argument in favour of the ancientness of compounds with a
reduced but accented M1 is the vocalism of M2, typical of syllables
unaccented already in the protolanguage:
*n'-p&2to:r vs. *p&2té:r
*h1sú-h2no:r (Gk. eué:no:r 'joy of men') vs. *h2ne:r
*trí-dk^omt- 'thirty'
etc. (cf. *népo:ts, *tétk^o:n, etc.)
This pattern is simply too archaic to have developed after
branch-specific accent shifts.
Piotr