In happy moments, Proto-Indo-European can be internally compared to uncover lexical items of distinguished age. Of course, the pronominals are first on the list (Fortson once called them "Devonian rocks" for their phonological inertness). But there are many other bits and pieces that can be uncovered from of period that predates the traditional Proto-Indo-European era. Heck, Lehmann wrote an entire book on it (Pre-Proto-Indo-European), which was controversially received.
Inspired by a brief mention on Gąsiorowski's site, I thought it fun to a very brief look at the evidence behind a very archaic piece of Indo-European language: a word for "woman" that predates the traditional words (i.e. Pokorny's IEW: *meri̯o- "(young) woman" and *gu̯ē̆nā "woman," "queen"). The feminine noun class did not exist in the Proto-Indo-European, instead PIE had an animate/inanimate contrast which is reflected in the Anatolian languages. After the Anatolian branch leave the pack did late Proto-Indo-European split the animate class into masculine and feminine and redevelop the inanimate class into a neuter.
But without a feminine noun class how did the Indo-Europeans distinguish female things when a speaker needed to get particular? For example, in English (a language that has lost much of its gender inflections), we see a vibrant selection of "particulars" that go way beyond saying "female." There's folk > women-folk ; friends > girlfriends (in a platonic sense) ; dudes > dudettes. Speaking very broadly about the productive state of Modern English, either a word meaning (young) woman fronts the head noun or a feminizing suffix is attached.
It seems that Proto-Indo-European did the same. When the feminine gender suffix *-eh2 was created by the Indo-Europeans (I will ignore the debate over its mysterious origins), *-eh2 neatly replaced whatever suffix existed. But though the Indo-Europeans did a very good job of replacing the suffix, they did not do a perfect job: that suffix was likely *-sr- and probably came from an even older word *sōr- meaning "woman." Let's take a look at the evidence, starting with the smoking guns:
*su̯e-ḱur-"[parent]-in-law" ~ *su̯e-sr- "sister-in-law" ~ *su̯e-lo- "brother-in-law" The contrast is very stark. Evidently the first element is from *s(u̯)e- "one's own," one of the strongest attested morphemes among Indo-European languages. But *su̯e-ḱur- did not stand on its own, a masculine suffix was attached for father-in-law or a feminine suffix for mother-in-law. The etymon for a brother-in-law is not well attested, confined to Old Norse svili and Greek ἀέλιοι, but it existed and it's important to consider.
Hieroglyphic Luwian nānašri "of a sister" We see the suffix appear again, this time in an Anatolian language. So even though Anatolian split away from Proto-Indo-European before the creation of a feminine gender, we see that the branch did not depart prior to the creation of *-sr-. The first element nāna can be traced to a Cuneiform Luwian nāna- "brother" which I am guessing comes from Proto-Anatolian *ne(ǵ)o- "sibling" (of unknown origin, cf. Hittite neka "sister" nekna "brother"). But how do we know *-sr- did not mean sister and predated the Proto-Indo-European word for sister, *nepto- instead? The Luwians attached *-sr- as a way of clarifying the femininity of the person.
Hittite ašraḫiyaš- "of the womanhood" Right in the middle is an evident piece of evidence for *-sr-, but without cognates it is difficult to make a persuasive case. Interestingly, however, Neri and Schuhmann argue that Hittite a- is a feminine prefix from Proto-Indo-European *h2eh1-. This isn't entirely convincing to me unless we assume that *-sr- is a reflex of a separate suffix. The way we can safely assume this is to return to nānašri in Luwian and see that there was an evident suffixing pattern involving *-sr- that denoted woman-ness.
So even independent of Gąsiorowski's argument for *-sr- via the feminine form of the number four, *kʷete-sr-es, we can see there is plent of reason to believe a woman-suffix existed in a zero-root. As a noun, the Italic languages suggest *sōr- on the basis of its reflex in Latin soror and Greek ἔορ, though a more conservative rendering would be *ser-.
So there you have it. We just discovered a word that existed even before the era of the Proto-Indo-Europeans.