Why opossums should rule and other fantasies

Most languages have never been committed to writing, so that when linguists try to reconstruct their prehistory and the prehistory of the peoples who spoke them, they are as palaeontologists without a fossil record, as archaeologists without middens to excavate.

These four volumes, published over two years, comprise 58 chapters that attempt to explore how archaeology and linguistics - often uncomfortable bedfellows - can join forces to elucidate our past. Opening with the same preface and general introduction, each volume can be read independently. Each is subtitled as if papers had been grouped by topic, but the relationship between subtitle and contents is tenuous. José Garanger's contribution on archaeology and oral traditions, although a factual report, appears in volume one, Theoretical and Methodological Orientations . Some of the chapters are  papers originally presented at an archaeological congress in 1994; the remainder was commissioned for this publication and their quality varies immensely: three outstandingly good, three outstandingly bad, the rest a continuum in between.

Perhaps as a warning to the reader, the series opens with a very poor paper by Bernard Bichakjian. Divested of the jargon and of the irrelevant discussion of Chomskyan innateness, it claims that some features of language are more advantageous than others, and that languages always evolve towards those features. The survival of the glibbest, as it were.

What are those features? Mainly, fricatives are better than stops (French has more fricatives and fewer stops than Indo-European), vowels are good (French has more vowels than Latin), ergative is bad, prepositions are good, postpositions bad and modifiers should follow the modified, not precede it. All this correlates with human gestation: "It is pertinent to observe that early linguistic proficiency ties in with a remarkable feature of human gestation. The period is considerably shorter than what the mammalian norm would predict."

This last claim is fantasy: hippopotami have an eight-month gestation, lions 106 days, whereas humans, with nine months, have one of the longest gestation periods of mammals of comparable size. As for the claim that linguistic proficiency ties in with faster gestation, Bichakjian explains that the shorter the gestation, the sooner the young are exposed to the outside world, the sooner they learn, the

sooner their brains develop. No wonder then that Australian Aborigines remained so primitive: not only were they competing with marsupials, with gestations typically one tenth and less that of humans, but most of their languages entirely lack fricatives, most have only three vowels, most use postpositions instead of prepositions, most put modifier before modified - all disadvantageous

features. With their 12 days of gestation,

opossums are bound to rule the earth.

Mercifully, Bichakjian says nothing of Australian languages, and we are spared this sinister proof of his theories. But how does he justify such fantasies? By apparently conjuring them out of thin air. Thus: "The evolution of word order  provides a good opportunity to introduce an important qualification. The changes described above apply to all Indo-European languages in general." To claim that, you must dismiss English and all Germanic languages (modifier before modified), and ignore Hindi and Indic languages (modifiers first, postpositions, sentence-final verbs, lots of stops, few fricatives). Bichakjian is aware that Germanic languages do not fit his theory, so he fudges: "The rates  vary from language to language and from feature to feature, but the changes proceed in the same direction." Once again, you must ignore English, the syntax of which is fast evolving into the modifier-first direction, with entire sentences sometimes modifying a following noun, for example, this caption for a cartoon by Gary Larson: "At the I've Fallen and I Can't Get Up Building."

On the basis of this opening chapter, nothing in these books can be trusted, everything must be examined critically.

Fortunately, the second tome contains an outstanding contribution by Kathrin Krell, in the light of which many of the other articles can be read critically, even by the layman. The putative cradle of Proto-Indo-European (PIE) has been rocked east, west, north and south, on such evidence as the absence or presence of words for "salmon", "beech", "palm", and so on. Some have had it on the shores of the Baltic Sea, some in Anatolia. Here, Krell dissects an umpteenth attempt at relocating the homeland of PIE, this time by the late Marija Gimbutas.

Gimbutas identified the homeland of PIE with the Kurgan culture (from a Russian word meaning "barrow") characterised by the construction of earthen barrows for burials, and extending from the lower Volga basin to the upper Yenisei river. Gimbutas's argument is that the PIE vocabulary as reconstructed by comparative linguists matches archaeological findings about the Kurgan culture. For instance, PIE is reconstructed as having words for "horse", "sheep", "goat", "dog", and the excavations of Kurgan sites yield remains of those animals. In clearly laid-out tables, Krell shows where Gimbutas has erred - by focusing on positive evidence and ignoring negative evidence: Kurgan faunal remains also include donkeys, aurochs and badgers, but no PIE words can be reconstructed for them.

Gimbutas's second error was to assume that words keep their meanings unaltered. They do not, for example, "to table", which has diametrically opposed meanings in British and in American English, and "maize", which is "corn" in American English. To quote Krell: "We should not presume that in prehistoric times words corresponding to Ôhorse', Ôcow' or Ôsheep' carried the same semantic content as in historic times."

Her third error is failing to recognise that we have no idea what languages the Kurgan people may have spoken, because they have left no inscriptions, and Krell illustrates this point: "Written records notwithstanding, would archaeologists from Mars assume, 5,000 years from now, that 20th-century Hungarians spoke the same basic language as their neighbours?" How indeed could those Martian archaeologists of AD7000 figure out, from the ruins of Budapest, without written records, that today's Hungarians spoke a language nothing like their neighbours', but more like those of Finland and Estonia, 1,000 miles away?

Armed with Krell's dissection of Gimbutas's theses, one can move on to an abstruse paper by Robert Blust, "Early Austronesian terms for metals". Blust is perhaps the foremost expert in Austronesian, a family of languages sweeping through half the globe, from Madagascar to Easter Island, via the Philippines and Taiwan. From the evidence of today's Austronesian languages, Blust reconstructed Proto-Austronesian (PAN) terms for gold, iron and lead. These reconstructions have drawn sharp criticisms on the grounds that there is little or no archaeological evidence for the use of those metals at the time when Austronesian constituted a single language (PAN), before its diaspora, which Blust estimates to have occurred out of Taiwan, perhaps c.4000BC. Thus the linguist Terry Crowley writes about Blust's PAN word for "iron", barish : "There is no support for this kind of reconstruction... Archaeologists are fairly confident that metallurgy appeared suddenly in Southeast Asia only about 2,200 years ago."

Blust counters by arguing that use of iron does not imply knowledge of smelting, quoting the example of Greenlandic Eskimos, who used meteoritic iron to make knives before European contact. Yet it is sufficient to remember Gimbutas's mistake. Words do not necessarily keep their meanings unaltered and so the grounds for disagreement between Blust and his critics evaporate. We do not know what this word for "iron" precisely meant originally, anymore than Krell's Martians of AD7000 will be able to tell what "corn" meant - wheat, barley, or maize?

Teaming archaeology with language has in this case brought more confusion than clarification. But this is not always so. In "Oral traditions and archaeology: two cases from Vanuatu", JosŽ Garanger recounts how two myths of Vanuatu were confirmed by archaeology. The first myth, collected by a missionary who lived on Tongoa Island from 1879 to 1930, tells how an island, Kuwae, was "destroyed by a volcanic cataclysm deliberately provoked by a young man named Tumbuk some 20 generations previously" in revenge for having been tricked by its inhabitants. A study by a geologist, Jean-Jacques Espirat, found evidence for such a cataclysm in the Shepherd archipelago of which Tongoa is part. Research by Garanger later confirmed that a major volcanic eruption had taken place c. 1450, a date strikingly in agreement with the "20 generations" of oral tradition. The legend goes on to tell how a young man escaped the eruption and fled to Makura, an islet 30km south, to come back seven years later to what was left of Kuwae, and take control of the first island he reached, Tongoa, where he was eventually buried with two women to accompany him. Following indications in the legend, Garanger uncovered a burial site, the layout of which closely matched the descriptions of the oral tradition. Garanger was able to date it to 1475±85, again confirming the tradition.

The second myth concerns Roy Mata, a chieftain said to have lived on Efate Island (about 60km south of the Shepherd archipelago) long before the destruction of Kuwae. According to oral traditions, Roy Mata went to conquer central Vanuatu as far as Kuwae and was buried on Retoka, an islet off Efate. A collective grave has been discovered on Retoka, at the site indicated by tradition, and radiocarbon dating assigned it a date of 15±140, confirming the legends.

In an introduction to this paper, to show the worth of oral traditions, Garanger mentions how the New Zealand Maoris' legends of a giant flightless bird, the moa, had been dismissed as myth; how after the discovery of the Dinornis, it was argued that "these bird fossils should be dated to a geological era well before the presence of human beings"; how, as recently as 1959, the anthropologist G. P. Murdock wrote that native traditions were "virtually valueless". Yet archaeology has confirmed that the Dinornis was once contemporary with humans, and its eggs were found associated with Maori burial sites. Blust's PAN word for "iron" may turn out to be as much of a myth as the moa.

In the same volume, Margaret Sharpe and Dorothy Turnbridge recount how they examined and cross-correlated evidence from comparative linguistics, Australian Aboriginal oral traditions, geology, botany and zoology. This chapter is too rich to be summarised. Its interest lies both in the data presented and in the care taken to weigh the pros and cons. For instance, they discuss a legend, collected by Robert Dixon in 1964, about the origin of three crater lakes in North Queensland: two men incurred the wrath of the rainbow-serpent, who caused the earth to erupt, and several deep lakes were formed. The informant added that, at that time, the country around the lakes was not rainforest as now, but open scrub. The volcanic eruptions that formed the lakes are estimated by geologists to have occurred at least 13,000 years ago, but pollen counts give the rainforest as only 7,600 years old, corroborating the Aboriginal story. Was this tradition handed down from generation to generation for 13,000 years, as some suggested? Sharpe and Turnbridge note that the story was collected in 1964, shortly before the pollen count in 1968. They conclude that "the possibility of some influence from recent speculation cannot be... excluded". This paper combines scholarly rigour with the appeal of a whodunit.

Krell's, Garanger's, and Sharpe and Turnbridge's contributions really answer to the title of this tetralogy, Archaeology and Language . There are other worthwhile papers - Blust's for instance - but many can claim only the most tenuous links with archaeology, some none; some, mere rehashes of earlier publications, are clearly padding; a few are as worthless as Murdock claimed native traditions to be.

To give one more example of the latter, in a chapter entitled "Linguistic similarity measures using the minimum message length principle", Anand Raman and Jon Patrick open by summarising earlier attempts at measuring similarities between languages. The names mentioned there (Kroeber, Sapir, Swadesh, and so on) will be familiar to linguists who have dabbled in such exercises, and will perhaps lull them into a sense of security. Raman and Patrick then present their own metric. A mixture of finite-state machines and of Shannon and Weaver's mathematical theory of information, it is incomprehensible to the vast majority of linguists and archaeologists. Space precludes my showing that it is actually a complete mathematical nonsense. How could such a paper make it into print?

Such contributions pose the question of editorial responsibility. The difficulty for the editors is that the books' topics are so diverse, and many are so specialised, that it would take a polymath to vet everything. But the unreliability of some parts of this tetralogy makes it worth buying only for research libraries frequented by expert specialists.

Jacques B. M. Guy is an Austronesian comparative linguist and computer scientist.