We can at least start to narrow down the possibilities, by examining the oldest written samples of Indo-European languages. For a long time, the oldest samples that scholars could identify were Iranian texts of around 1000-800 BC, and Sanskrit texts probably composed around 1200–1000 BC but written down later. Texts of a Mesopotamian kingdom called Mitanni, written in a non-Indo-European language but containing some words obviously borrowed from a language related to Sanskrit, push the proven existence of Sanskrit-like languages back to nearly 1500 BC. The next breakthrough was the late-nineteenth-century discovery of a mass of ancient Egyptian diplomatic correspondence. Most of it was written in a Semitic language, but two letters in an unknown language remained a mystery until excavations in Turkey uncovered thousands of tablets in the same tongue. The tablets proved to be the archives of a kingdom that thrived between 1650 and 1200 BC and that we now refer to by the biblical name 'Hittite'. In 1917 scholars were astonished by the announcement that the Hittite language proved on deciphering to belong to a previously unknown, very distinctive and archaic, now-vanished branch of the Indo-European family, termed Anatolian. Some obviously Hittite-like names mentioned in earlier letters of Assyrian merchants at a trading post near the Hittite capital's future site push the detective trail back to nearly 1900 BC. This remains our first direct evidence for the existence of any Indo-European language.

Thus, as of 1917, two Indo-European branches—Anatolian and Indo-Iranian—had been shown to exist by around 1900 and 1500 BC, respectively. A third early branch was established in 1952, when the young British cryptographer, Michael Ventris, showed that the so-called Linear B writing of Crete and Greece, which had resisted deciphering since its discovery around 1900, was an early form of the Greek language. Those Linear B tablets date to around 1300 BC. But Hittite, Sanskrit, and early Greek are very different from each other, certainly more so than are modern French and Spanish, which diverged over a thousand years ago. That suggests that the Hittite, Sanskrit, and Greek branches must have split off from PIE by 2500 BC or earlier. How much earlier do the differences between those branches imply? How can we obtain a calibration factor that converts 'percentage difference between languages' into 'time since the languages began to diverge'? Some linguists use the rate of word change in historically documented, written languages, like the changes from Anglo-Saxon to Chaucer's English to Modern English. These calculations, which belong to a science called glottochronology (or chronology of languages), yield the rule of thumb that languages replace about twenty per cent of their basic vocabulary every thousand years.

Most scholars reject glottochronology calculations, on the grounds that word replacement rates must vary with social circumstances and with the particular words themselves. However, the same scholars are generally still willing to make a seat-of-the-pants estimate. The usual conclusion from either glottochronology or pants' seats is that PIE may have started to break up by 3000 BC, surely by 2500 BC, and not before 5000 BC.

There is still another, completely independent approach to the dating problem—the science termed linguistic paleontology. Just as paleontologists try to discover what the past was like by looking for relics buried in the ground, linguistic paleontologists do it by looking for relics buried in languages.

To understand how this works, recall that linguists have reconstructed nearly 2,000 words of PIE vocabulary. It is not surprising that these include words like 'brother' and 'sky', which must have existed and been named since the dawn of human language. But PIE should not have had a word for 'gun', since guns were not invented until about 1300 AD, long after PIE-speakers had already scattered to speak distinct languages in Turkey and India. In fact, the word for 'gun' uses different roots in different Indo-European languages: 'gun' in English, fusil in French, ruzhyo in Russian, and so on. The reason is obvious: different languages could not possibly have inherited the same root for 'gun' from PIE, and they each had to invent or borrow their own word when guns were invented.

The gun example suggests that we should take a series of inventions whose dates we know, and see which of those do and which do not have reconstructed names in PIE. Anything—like gun—that was invented after PIE began to break up should not have a reconstructed name. Anything—like brother—that was invented or known before the break-up might have a name. (It does not have to have a name, because plenty of PIE words have surely become lost. We know the PIE words for 'eye' and 'eyebrow' but not'eyelid', although PIE speakers must have had eyelids.) Perhaps the earliest major developments without PIE names are battle chariots, which became widespread between 2000 and 1500 BC, and iron, whose use became important between 1200 and 1000 BC. The lack of PIE terms for these relatively late inventions does not surprise us, since the distinctness of Hittite had already convinced us that PIE broke up long before 2000 BC. Among earlier developments that do have PIE names, there are words for 'sheep' and 'goat', first domesticated by around 8000 BC; cattle (including separate words for cow, steer, and ox), domesticated by 6400 BC; horses, domesticated by around 4000 BC; and ploughs, invented around the time that horses were domesticated. The latest datable invention with a PIE name is the wheel, invented around 3300 BC.

Therefore, linguistic paleontology, even in the absence of any other evidence, would date the break-up of PIE as before 2000 BC but after 3300 BC. This conclusion agrees well with the one reached by extrapolating the differences between Hittite, Greek, and Sanskrit backwards in time. Hence if we wish to find traces of the first Indo-Europeans, we should be safe concentrating on the archaeological record between 2500 and 5000 BC, and perhaps slightly before 3000 BC. Having reached fair agreement about the 'when' question, let's now ask: where was PIE spoken? Linguists have disagreed about the PIE homeland ever since they first began to appreciate its significance. Almost every possible answer has been proposed, from the North Pole to India, and from the Atlantic to the Pacific shores of Eurasia. As the archaeologist J.P. Mallory puts it, the question is not, 'Where do scholars locate the Indo-European homeland? , but 'Where do they put it nowT

To understand why this problem has proved so difficult, let's first try to solve it quickly by looking at a map (see page 228). As of 1492, most surviving Indo-European branches were virtually confined to Western Europe, and only Indo-Iranian extended east of the Caspian Sea. Western Europe would be the most parsimonious solution to the search for the PIE homeland, the solution that required the fewest people to move long distances. Unfortunately for that solution, in 1900 a 'new' but long-extinct Indo-European language was discovered in a triply unlikely location. Firstly, the language (Tocharian, as it is now known) turned up in a secret chamber behind a wall in a Buddhist cave monastery. The chamber contained a library of ancient documents in the strange language, written around 600–800 AD by Buddhist missionaries and traders. Secondly, the monastery lay in Chinese Turkestan, east of all extant Indo-European speakers and about a thousand miles removed from the nearest ones. Finally, Tocharian was not related to Indo-Iranian, the geographically closest branch of Indo-European, but possibly instead to branches used in Europe itself, thousands of miles to the west. It is as if we suddenly discovered that the early medieval inhabitants of Scotland spoke a language related to Chinese.