A
hypothetical language spoken (some believe) by the
ancestors of all
human beings at some point, and from which all modern
languages are
descended. The
ancestral group presumably spoke it while still in
Africa between 50 000 and 100 000 years ago, before any modern
Homo sapiens had branched out into the rest of the world.
That all languages are ultimately genetically related is a reasonable idea, because of many similarities and psychological studies suggesting the deep structure of language is encoded in the brain from birth. Noam Chomsky is most associated with one form of this thesis; Steven Pinker and others write popular expositions of competing theories along broadly similar lines. This ultimate genetic relationship is called the monogenesis hypothesis. Not all linguists would agree with it; and many would be non-committal.
What is not anywhere near as reasonable is to expect any evidence still to remain of what it actually sounded like. The conventional technique of comparative reconstruction can produce solid results for a certain depth back in time: 5000 years certainly, maybe a bit more, conceivably in some extraordinary cases as much as 10 000 years. Very few linguists believe it is possible to talk about 'Proto-World' as a real language or ascertain any facts about it, other than the common human inheritance of Universal Grammar, which it would share with all its descendants.
The ancestors of the modern Indo-European, Sino-Tibetan, Semitic, and Austronesian families, for example, can be reconstructed with high confidence from abundant evidence. But because all languages change continually, they diverge from one another, and items common at one period either get replaced by unrelated words or diverge so far in sound that the kinship can no longer be recognized. Entropy continually destroys evidence of relationship.
After five or ten thousand years any resemblance between two stocks is equally likely to have arisen by chance as to represent a genuine common ancestor. The technique of glottochronology can only take us as far back as this point where chance is as salient as cognateness.
It may be possible to use other techniques to delve a bit further and get wider affinities -- as with the debatable superfamilies known as Nostratic, Dene-Caucasian, Austric, Amerind, and Indo-Pacific --, but the Proto-World idea is that we can find traces from fifty to a hundred thousand years ago. Almost no linguist accepts this. The position is put forward most strongly by Merritt Ruhlen, though it was supported by the very respected Joseph Greenberg, who favoured a technique called mass comparison to show up possible matches across groups. However, the great majority think Greenberg overstepped himself here.
Items that have been put forward as candidates, with instances in various half-recognizable forms across the world, include roots vaguely like HIT for 'eat', AKW for 'water' (aqua), TIK for 'toe, finger', and MALK for 'suck; throat; milk'.
The claims of the monogeneticists like Ruhlen are to some (disputable) degree bolstered by biological genetics. Some scientists, of whom Luca Cavalli-Sforza is the best known, have in recent years used genetic markers to trace back lines of human ancestry. The main division is between Africans and non-Africans, then there is the branch of Australians and New Guineans, then within the Eurasian branch there are (very roughly) Western/Caucasian/Central types, North Asian types, American types, and South-East Asian types. There is a significant though not perfect match with the largest language phyla that can be discerned.
Before this? Before Homo sapiens sapiens emerged from Africa? Who knows. Evidence of any kind is virtually unobtainable: what kind of communicative ability did the culture of pre-humans require? The biological evidence from palaeontology is too weak to decide: did Neanderthals have a Broca's area in their braincase? Do their hyoid bones show a descended larynx suitable for speech? We can't tell.