Do you mean ‘face’ for πρόσωπον? This is also found in Sanskrit as prátīka ‘opposite, countenance’, going back to *próti-h₃kʷ-om or thereabouts. The Tocharian forms have strange vocalism and accentuation--shouldn’t we see pretsāk or prātsak? But the TB form is an ā-stem feminine, and I assume the TA form is too.
“Everything except Anatolian, Italic, Celtic, and Tocharian innovates a -y mediopassive”. Does it? Unless there’s something weird lurking in the dialect of three 90-year-old women in rural Latvia I’m not aware of any mediopassives in Balto-Slavic. Greek has a y-mediopassive--but Phrygian has an r-mediopassive, and Phrygian and Greek are close, with shared innovations like an oblique stem for ‘woman’ in -aik-. (Remember that Phrygian is just as full a member of the family as Greek is, just a poorly-attested one.)
I am inclined to ignore lexicostatistics entirely. Shared lexical isoglosses can still be found by hand, but determining innovations vs. retentions is something of an art. Thus in Algonquian, the word for ‘winter’ reflects a long second-syllable vowel *pepōnwi in most daughters--but not in Cree or Arapaho which reflect *pepŏnwi. Because Cree and Arapaho do not otherwise form a group and have a number of other archaisms, we conclude that an ‘inner core’ group irregularly lengthened the vowel in ‘winter’. But Algonquian is young, and transparent, and there were already lots of reasons to think Cree and Arapaho broke off early and that the languages to their south and east might form a group. In the absence of that evidence, we might put Cree and Arapaho together, because we wouldn’t be able to make the qualitative judgment calls putting *pepŏnwi in the ‘archaism’ bin rather than the ‘shared innovation’ bin.
Is the genitive dual a ‘rebuilding’? Let’s take a look at genitive duals across the family.
Greek’s ending is -Vιν; -αιν in the first declension and -οιν in the second and third. Additionally, some Homeric lines have to be scanned with -οιιν. (I expect the third declension borrowed from the second and can be discounted). Hilmarsson says there’s an Arcado-Cypriot second-declension form -οιυν, with the -οι- formation possibly comparable to the TB -ais- formative of the genitive dual -(n)aisäñ. (Duals in Tocharian are built on an n-stem). Tocharian A seems to have a genitive dual ending -e or -is, as in āmpe/āmpine ‘of both’, aśnis ‘of two eyes’.
Sanskrit has -ayos in thematics, or simply -os in athematics, and this looks like it has to reflect *-ews or *-ows. But Avestan has separate dual endings for the genitive and locative, with -å in the genitive and -ō in the locative--supposed to be from Pre-Avestan -āh and -aŭ, respectively. De Vaan and Martínez reckon the Sanskrit ending represents a ‘blend’ of distinct endings in Indo-Iranian--which suggests *-oHs or *-eHs. This could easily be secondary--take the nominative dual in *-e/oh₁, add the -s from the genitive, and hey presto, you have a genitive dual.
OCS has -u uniformly in the genitive dual, which looks like *-ow or--maybe-- *-ew with the intercalated yod analogized away. On the other hand, it also has the same genitive/locative syncretism in the dual that Sanskrit does, so perhaps this is the equivalent of the -o- of Sanskrit, which added an -s which was either never added in OCS or was lost when the open-syllable law hit. Samogitian and the Lithuanian pronominals seem not to distinguish the genitive dual from the genitive plural (both are -ų/-u).
Old Irish has lenition mutation, which Stifter reconstructs to Proto-Celtic -ow, from *-o(h₁)u or -ows or similar (but if there was an -s, why not spirantization mutation?). The “prepositional” dual is a *bʰ-oblique and isn’t relevant.
I think if we’re going to reconstruct anything, we want a locative dual ending *-o(h₁)u. Maybe this did double duty for the genitive at an early date as well, because the Avestan ending suggests *-Vh₁-s which is very easy to innovate after break-up, and then Sanskrit represents a merger of the two.
The nu of the Greek ending and the -ñ of the Tocharian ending don’t have to be cognate and probably aren’t (why not -aisäṃ?). The real trouble is the *i. IE *oi is supposed to go to Tocharian e, *ou to o or au. What we might expect is something along the lines of *-oh₁u-s-Vn(i) shared in Tocharian and Greek, but this would give, what, -οῦν and -ausäñ/-osäñ. Not what we see. Winter proposed *-oisun-, but Hilmarsson says this doesn’t explain Tocharian for Reasons. We could get Homeric from -oih₁i(n) with dissimilation in Arcadocypriot; Hilmarsson proposes competing genitive dual endings *-oiH₁ou and -oiH₁i.
Note that -oy-su is, uh...exactly what we think the Sanskrit o-stem locative plural ending -eṣu comes from. Avestan has -šu under RUKI, otherwise -hu. The Greek genitive dual, but not the Tocharian, in *-oysi looks a lot like the locative plural--we have -οισι in Homeric, probably formerly -οϊ, with the sigma reintroduced from consonant stems.
Perhaps we are looking at a grab-bag of postpositions or adverbs or something of the form *isu, *usu, *isi, etc. in IE which glommed on to existing forms? Homeric can easily reflect an *h₁, the dual marker. Intercalation of yod in Sanskrit -eṣu might be by analogy with e.g. -ebhyas; it’s also present in the dual in -ayos, which could be secondary after it at an early stage. There’s a Hittite preverb u- ‘hither’, which Kloekhorst takes to be from *h₂ou and cognate to Skt. ava-, Greek αὖ ‘again’,etc-- but then why not ḫu-? It has an allative meaning in verbs, too, not a stative one.
Pre-Greek vocabulary in Tocharian is difficult to dismiss if it exists and I would like to know more about this. The equation of wánax/wánassa with nātäk/nāśi...hmm. Why don’t we see wanāk, wanāśi in Tocharian? This nātäk word doesn’t show up in Adams’ dictionary of Tocharian B (even the new second edition) (is it TochA?). There’s a word ñakte ‘god’ which appears to be from *(H)néktos or thereabouts, with various discussion in the dictionary connecting it to *h₁neḱ- and *ǵʰew-. We could get nā- from *neh₁-- *eh₁ doesn’t produce palatalization and turns to ā in Tocharian; *(s)neh₁- meant ‘to weave’ but there’s also *neyh₁- ‘to lead’.
The “Northwest Group” seems sketchy to me, a hodgepodge of areal similarities and innovations that look areal. Much of the group is centum, but Balto-Slavic and Albanian aren’t, and it leaves out Greek. It mixes y- and r-mediopassives, along with Balto-Slavic which is silent on the question.
