Bringing Substrata back to PIE

One undiscussed commonality of suspected European substrata is the disagreement over velar *k and *g in daughter languages. You could probably write a well-received paper by merely attempting to resolve the problem. Let's take a look at the embitterned disagreement between Celtic and Germanic forms respectively:

Here's the common course of development in Proto-Indo-European (PIE) to Proto-Celtic (PC)

  1. PIE to Late PIE: Plain and palatalvelars merge. *ḱ > *k ; *ǵ > *g ; *ǵh > *gh
  2. Late PIE to Early PC: But velars followed by *w merge as labiovelars. *kw > *kw ; *gw > *gw ; *ghw > *ghw
  3. Early PC: ...but then labiovelar *gw becomes *b.
  4. Early PC: Deaspiration of stops. *gh > *g ; *gwh > *gw

The development of PIE into Proto-Germanic (PG) is as follows:

  1. PIE to Late PIE: Plain and palatalvelars merge. *ḱ > k ; *ǵ > *g ; *ǵh > *gh
  2. Pre-PG: *h3 > *g between sonorant and *w.
  3. Late PIE to Early PG: Grimm's Law. Fricativization of voiceless plosives. *k > *h ; *kw > *hw. But *sk and *skw survive.
  4. Late PIE to Early PG: Spiranization of velar + (*w) + *t clusters. *gt , *gwt , *ght , *gwht , *kt , *kwt > *ht.
  5. Late PIE to Early PG: Devoicing. *g > *k ; *gw > *kw.
  6. Late PIE to Early PG: Deaspiration of voice plosives. *gh > *g ; *gwh > *gw
  7. Late PIE to Early PG: Verner's Law. *h > *g ; *hh > *gw when preceded by an unaccented syllable.
  8. Late PIE to Early PG: Laviovelar *gw-initial becomes *b-initial, similar to Celtic development #3.
  9. Late PG: *gw > *w.

So what we can expect to see is something like this:

PIE PC PG
*k , *ḱ *k *h
*kw *kw *hw
*g , *ǵ *g *k
*gh , *ǵh *g *g
*gw *b *kw
*gw-initial *b *b
*gwh *gh *g
*gwh-initial *gh *b
*gt , *gwt , *ght , *gwht , *kt , *kwt *ht
Exceptions
sonorant + *h3w sonorant + *ow sonorant + *kw
( *f , *s , *h ) + *k , *ḱ ( *f , *s , *h ) + *k (*f , *s , *h ) + *k
*-k- stop *-k- *-g-

There is remarkable regularity in these sound shifts, but not perfect regularity. As I said before, there are some words that are unresolved and difficult:

1. 2.3.
PC Meaning < PIE PG Meaning < PIE PGMeaning< PIE
*gabro- "billy goat" *g- , *ǵ- , *gh- , *ǵh- *hafra- "billy goat" *k- , *ḱ- *habran- "oats," "goat" *k- , *ḱ-
*seski- "sedge" *k , *ḱ *sahaza- "sedge" *k , *ḱ *sagja- "sedge" *ǵ , *ǵh , *k-stop
Both from Guus Kroonen, Etymological Dictionary of Proto-Germanic. Indo-European Dictionaries Online. Ed. by Alexander Lubotsky. Brill. 2014.

The first example is thoroughly obscure. There is no obvious resolution to the problem without invoking bizarre or ad hoc solutions. But perhaps the second example can be fixed.

As Kroonen (2014) pointed out, the Proto-Celtic form looks to be a metathesis of a pre-form *seks-(k)i-(?). That's a good start but let's simplify the metathesis to just *seks-i. Maybe you can see where this is going now.

PG *sahaza- should be **sagaza- if the inner consonant was a *-k- voiceless stop. This is further confirmed by PG *sagja- which does contain the expected *g as the result of an older voiceless stop. But if *sahaza came from a pre-form of *sak-sa- and the suffix was later fronted by an a-vowel, we get a very safe answer.

What I propose is that we re-analyze the solution. The Germanic and Celtic forms points to a verbal root *sek- with three unknonwn suffies *-sa- (< PG *sahaza-); *-si- (< PC *sesk- < Pre-PC *sek-si-); and *-ja- (< PG *sagja-). This would not be the first time that Proto-Germanic unexpectedly grew a *-sV- suffix into *-VsV-, cf. the names for alder trees in Proto-Balto-Slavic *a/el(i)snio- ~ PBS *a/elisaH- PG *alis/zo ~ PG *al-s- ~ Proto-Italic *alsno- (Derkson 2008).