You, hear me! Give this fire to that old man. Pull the black worm off the bark and give it to the mother. And no spitting in the ashes!

It's an odd little speech. But if it were spoken clearly to a band of hunter-gatherers in the Caucasus 15,000 years ago, there's a good chance the listeners would know what you were saying.

That's because all of the nouns, verbs, adjectives, and adverbs in the four sentences are words that have descended largely unchanged from a language that died out as the glaciers were retreating at the end of the last Ice Age.

The traditional view is that words can't survive for more than 8,000 to 9,000 years. Evolution, linguistic "weathering," and the adoption of replacements from other languages eventually drive ancient words to extinction, just like the dinosaurs of the Jurassic.

New research, however, suggests a few words survive twice as long.

Their existence, in turn, suggests there was a "proto-Eurasiatic" language that was the common ancestor of about 700 languages used today, and many others that have died out over the centuries.

"We've never heard this language, and it's not written down anywhere," said Mark Pagel, an evolutionary theorist at the University of Reading in England who headed the study that appears in the Proceedings of the National Academy of Sciences. "But this ancestral language was spoken and heard. People sitting around campfires used it to talk to each other."

Pagel and his collaborators came up with a list of two dozen "ultraconserved words." It contains both predictable and surprising members.

The most conserved word is thou, which is the singular form of you. I, not, what, mother, and man are also on the list. So are the verbs to hear, to flow, and to spit and the nouns bark, ashes, and worm. Together, they hint at what has been important to people over the last 15 millennia.

Pagel and three collaborators studied "cognates" - words that have the same meaning and a similar sound in different languages. Father (English), padre (Italian), pere (French), pater (Latin), and pitar (Sanskrit) are cognates. Those words, however, are from languages in one family, the Indo-European. The researchers looked much farther afield, in seven language families in all.

In addition to Indo-European, the language families include the Altaic (whose modern members include Turkish, Uzbek, and Mongolian); Chukchee-Kamchatkan (languages of far northeastern Siberia); Dravidian (languages of south India); Inuit-Yupik (Arctic languages); Kartvelian (Georgian and three related languages); and Uralic (Finnish, Hungarian, and a few others).

They are a diverse group. Some don't use the Roman alphabet. Some had no written form until modern times. They sound different to the untrained ear. Their speakers live tens of thousands of miles apart. In short, they seem unlikely candidates to share cognates.

Pagel's team used as its starting material 200 words that linguists know to be the core vocabulary of all languages.

Other researchers had looked for cognates of those words in members of each of the seven Eurasiatic language families. They looked, for example, for similar-sounding words for fish or to drink in the Altaic family of languages or in the Indo-European languages.

When they found cognates, they constructed what they imagined were the cognates' ancestral words - a task that requires knowing how sounds change between languages, such as "F" in Germanic languages becoming "P" in Romance languages.