The Horse, the Wheel, and Language
David W. Anthony
Published by Princeton University Press
Anthony, W..
The Horse, the Wheel, and Language: How Bronze-Age Riders from the Eurasian Steppes Shaped the Modern World.
Princeton: Princeton University Press, 2010.
Project MUSE., https://muse.jhu.edu/.
For additional information about this book
Access provided by Harvard University (15 Oct 2018 21:57 GMT)
https://muse.jhu.edu/book/36661
21
Chapter Two
How to Reconstruct a Dead Language
Proto-Indo-European has been dead as a spoken language for at least
forty-five hundred years. The people who spoke it were nonliterate, so
there are no inscriptions. Yet, in 1868, August Schleicher was able to tell a
story in reconstructed Proto-Indo-European, called “The Sheep and the
Horses,” or Avis akvasas ka. A rewrite in 1939 by Herman Hirt incorporated
new interpretations of Proto-Indo-European phonology, and the title
became Owis ek’woses-kw
e. In 1979 Winfred Lehmann and Ladislav
Zgusta suggested only minor new changes in their version, Owis ekwoskw
e.
While linguists debate increasingly minute details of pronunciation in
exercises like these, most people are amazed that anything can be said
about a language that died without written records. Amazement, of course,
is a close cousin of suspicion. Might the linguists be arguing over a fantasy?
In the absence of corroborative evidence from documents, how can
linguists be sure about the accuracy of reconstructed Proto-Indo-
European?1
Many archaeologists, accustomed to digging up real things, have a low
opinion of those who merely reconstruct hypothetical phonemes—what is
called “linguistic prehistory.” There are reasons for this skepticism. Both
linguists and archaeologists have made communication across the disciplines
almost impossible by speaking in dense jargons that are virtually
impenetrable to anyone but themselves. Neither discipline is at all simple,
and both are riddled with factions on many key questions of interpretation.
Healthy disagreement can resemble confusion to an outsider, and
most archaeologists, including this author, are outsiders in linguistics.
Historical linguistics is not taught regularly in graduate archaeology programs,
so most archaeologists know very little about the subject. Sometimes
we make this quite clear to linguists. Nor is archaeology taught to
graduate students in linguistics. Linguists’ occasional remarks about archaeology
can sound simplistic and naïve to archaeologists, making some
of us suspect that the entire field of historical linguistics may be riddled
with simplistic and naïve assumptions.
The purpose of these first few chapters is to clear a path across the noman’s
land that separates archaeology and historical linguistics. I do this
with considerable uncertainty—I have no more formal training in linguistics
than most archaeologists. I am fortunate that a partial way has already
been charted by Jim Mallory, perhaps the only doubly qualified linguistarchaeologist
in Indo-European studies. The questions surrounding IndoEuropean
origins are, at their core, about linguistic evidence. The most
basic linguistic problem is to understand how language changes with
time.2
Language Change and Time
Imagine that you had a time machine. If you are like me, there would be
many times and places that you would like to visit. In most of them, however,
no one spoke English. If you could not afford the Six-MonthImmersion
Trip to, say, ancient Egypt, you would have to limit yourself to
a time and place where you could speak the language. Consider, perhaps,
a trip to England. How far back in time could you go and still be understood?
Say we go to London in the year 1400 ce.
As you emerge from the time machine, a good first line to speak, something
reassuring and recognizable, might be the opening line of the Lord’s
Prayer. The first line in a conservative, old-fashioned version of Modern
Standard English would be, “Our Father, who is in heaven, blessed be your
name.” In the English of 1400, as spoken by Chaucer, you would say, “Oure
fadir that art in heuenes, halwid be thy name.” Now turn the dial back
another four hundred years to 1000 CE, and in Old English, or AngloSaxon,
you would say, “Fæader ure thu the eart on heofonum, si thin nama
gehalgod.” A chat with Alfred the Great would be out of the question.
Most normal spoken languages over the course of a thousand years undergo
enough change that speakers at either end of the millennium, attempting
a conversation, would have difficulty understanding each other.
Languages like Church Latin or Old Indic (the oldest form of Sanskrit),
frozen in ritual, would be your only hope for effective communication
with people who lived more than a thousand years ago. Icelandic is a frequently
cited example of a spoken language that has changed little in a
thousand years, but it is spoken on an island isolated in the North Atlantic
by people whose attitude to their old sagas and poetry has been one approaching
religious reverence. Most languages undergo significantly more
22 Chapter 2
Reconstructing a Dead Language 23
changes than Icelandic over far fewer than a thousand years for two reasons:
first, no two people speak the same language exactly alike; and, second,
most people meet a lot more people who speak differently than do the
Icelanders. A language that borrows many words and phrases from another
language changes more rapidly than one with a low borrowing rate.
Icelandic has one of the lowest borrowing rates in the world.3
If we are
exposed to a number of different ways of speaking, our own way of speaking
is likely to change more rapidly. Fortunately, however, although the
speed of language change is quite variable, the structure and sequence of
language change is not.
Language change is not random; it flows in the direction of accents and
phrases admired and emulated by large numbers of people. Once a target
accent is selected, the structure of the sound changes that moves the speaker
away from his own speech to the target is governed by rules. The same rules
apparently exist in all our minds, mouths, and ears. Linguists just noticed
them first. If rules define how a given innovation in pronunciation affects
the old speech system—if sound shifts are predictable—then we should be
able to play them backward, in effect, to hear earlier language states. That
is more or less how Proto-Indo-European was reconstructed.
Most surprising about sound change is its regularity, its conformation to
rules no one knows consciously. In early Medieval French there probably
was a time when tsent’m ‘hundred’ was heard as just a dialectical pronunciation
of the Latin word kentum ‘hundred’. The differences in sound between
the two were allophones, or different sounds that did not create
different meanings. But because of other changes in how Latin was spoken,
[ts-] began to be heard as a different sound, a phoneme distinct from [k-]
that could change the meaning of a word. At that point people had to decide
whether kentum was pronounced with a [k-] or a [ts-]. When French
speakers decided to use [ts-], they did so not just for the word kentum but in
every word where Latin had the sound k- before a front vowel like -e-. And
once this happened, ts- became confused with initial s-, and people had to
decide again whether tsentum was pronounced with a [ts-] or [s-]. They
chose [s-]. This sequence of shifts dropped below the level of consciousness
and spread like a virus through all pre-French words with analogous sequences
of sounds. Latin cera ‘wax’, pronounced [kera], became French cire,
pronounced [seer]; and Latin civitas ‘community’, pronounced [kivitas],
became French cité, pronounced [seetay]. Other sound changes happened,
too, but they all followed the same unspoken and unconscious rules—the
sound shifts were not idiosyncratic or confined to certain words; rather,
they spread systematically to all similar sounds in the language. Peoples’
ears were very discriminating in identifying words that fit or did not fit the
analogy. In words where the Latin k- was followed by a back vowel like -o it
remained a k-, as in Latin costa>French côte.
Sound changes are rule-governed probably because all humans instinctively
search for order in language. This must be a hard-wired part of all
human brains. We do it without committee meetings, dictionaries, or even
literacy, and we are not conscious of what we are doing (unless we are linguists).
Human language is defined by its rules. Rules govern sentence
construction (syntax), and the relationship between the sounds of words
(phonology and morphology) and their meaning. Learning these rules
changes our awareness from that of an infant to a functioning member of
the human tribe. Because language is central to human evolution, culture,
and social identity, each member of the tribe is biologically equipped to
cooperate in converting novel changes into regular parts of the language
system.4
Historical linguistics was created as a discipline in the nineteenth century,
when scholars first exposed and analyzed the rules we follow when
speaking and listening. I do not pretend to know these rules adequately,
and if I did I would not try to explain them all. What I hope to do is indicate,
in a general way, how some of them work so that we can use the “reconstructed
vocabulary” of Proto-Indo-European with some awareness of
its possibilities and limitations.
We begin with phonology. Any language can be separated into several
interlocking systems, each with its own set of rules. The vocabulary, or
lexicon, composes one system; syntax, or word order, and sentence construction
compose another; morphology, or word form, including much of what
is called “grammar” is the third; and phonology, or the rules about which
sounds are acceptable and meaningful, is the fourth. Each system has its
own peculiar tendencies, although a change in one (say, phonology) can
bring about changes in another (say, morphology).5
We will look most
closely at phonology and the lexicon, as these are the most important in
understanding how the Proto-Indo-European vocabulary has been recon-
structed.
Phonology: How to Reconstruct a Dead Sound
Phonology, or the study of linguistic sounds, is one of the principal tools
of the historical linguist. Phonology is useful as a historical tool, because
the sounds people utter tend to change over time in certain directions and
not in others.
24 Chapter 2
Reconstructing a Dead Language 25
The direction of phonetic change is governed by two kinds of constraints:
those that are generally applicable across most languages, and
those specific to a single language or a related group of languages. General
constraints are imposed by the mechanical limits of the human vocal
anatomy, the need to issue sounds that can be distinguished and understood
by listeners, and the tendency to simplify sound combinations that
are difficult to pronounce. Constraints within languages are imposed by
the limited range of sounds that are acceptable and meaningful for that
language. Often these language-specific sounds are very recognizable.
Comedians can make us laugh by speaking nonsense if they do it in the
characteristic phonology of French or Italian, for example. Armed with a
knowledge of both the general tendencies in the direction of phonetic
change and the specific phonetic conventions within a given language
group, a linguist can arrive at reliable conclusions about which phonetic
variants are early pronunciations and which come later. This is the first
step in reconstructing the phonological history of a language.
We know that French developed historically from the dialects of Latin
spoken in the Roman province of Gaul (modern France) during the waning
centuries of the Roman Empire around 300–400 CE. As late as the
1500s vernacular French suffered from low prestige among scholars, as it
was considered nothing more than a corrupt form of Latin. Even if we
knew nothing about that history, we could examine the Latin centum (pronounced
[kentum]), and the French cent (pronounced [sohnt]), both meaning
“hundred,” and we could say that the sound of the Latin word makes
it the older form, that the Modern French form could have developed
from it according to known rules of sound change, and that an intermediate
pronunciation, [tsohnt], probably existed before the modern form
appeared—and we would be right.
Some Basic Rules of Language Change: Phonology and Analogy
Two general phonetic rules help us make these decisions. One is that initial
hard consonants like k and hard g tend to change toward soft sounds
like s and sh if they change at all, whereas a change from s to k would generally
be unusual. Another is that a consonant pronounced as a stop in the
back of the mouth (k) is particularly likely to shift toward the front of the
mouth (t or s) in a word where it is followed by a vowel that is pronounced
in the front of the mouth (e). Pronounce [ke-] and [se-], and note the position
of your tongue. The k is pronounced by using the back of the tongue
and both e and s are formed with the middle or the tip of the tongue,
which makes it easier to pronounce the segment se- than the segment ke-.
Before a front vowel like -e we might expect the k- to shift forward to [ts-]
and then to [s-] but not the other way around.
This is an example of a general phonetic tendency called assimilation:
one sound tends to assimilate to a nearby sound in the same word, simplifying
the needed movements. The specific type of assimilation seen here is
called palatalization—a back consonant (k) followed by a front vowel (e)
was assimilated in French toward the front of the palate, changing the [k]
to [s]. Between the Latin [k] (pronounced with the back of the tongue at
the back of the palate) and the Modern French [s] (tip of the tongue at the
front of the palate) there should have been an intermediate pronunciation
ts (middle of the tongue at the middle of the palate). Such sequences permit
historical linguists to reconstruct undocumented intermediate stages
in the evolution of a language. Palatalization has been systematic in the
development of French from Latin. It is responsible for much of the distinctive
phonology of the French language.
Assimilation usually changes the quality of a sound, or sometimes removes
sounds from words by slurring two sounds together. The opposite
process is the addition of new sounds to a word. A good example of an
innovation of this kind is provided by the variable pronunciations of the
word athlete in English. Many English speakers insert [-uh] in the middle
of the word, saying [ath-uh-lete], but most are not aware they are doing
so. The inserted syllable always is pronounced precisely the same way, as
[-uh], because it assimilates to the tongue position required to pronounce
the following -l. Linguists could have predicted that some speakers would
insert a vowel in a difficult cluster of consonants like -thl (a phenomenon
called epenthesis) and that the vowel inserted in athlete always would be
pronounced [-uh] because of the rule of assimilation.
Another kind of change is analogical change, which tends to affect grammar
quite directly. For example, the -s or -es ending for the plural of English
nouns was originally limited to one class of Old English nouns: stān
for stone (nominative singular), stānas for stones (nominative plural). But
when a series of sound changes (see note 5) resulted in the loss of the phonemes
that had once distinguished nouns of different classes, the -s ending
began to be reinterpreted as a general plural indicator and was attached to
all nouns. Plurals formed with -n (oxen), with a zero change (sheep), and
with a vowel change in the stem (women) remain as relics of Old English,
but the shift to -s is driving out such “irregular” forms and has been doing
so for eight hundred years. Similar analogical changes have affected verbs:
help/helped has replaced Old English help/holp as the -ed ending has been
26 Chapter 2
Reconstructing a Dead Language 27
reinterpreted as a general ending for the past tense, reducing the once large
number of strong verbs that formed their past with a vowel change. Analogical
changes can also create new words or forms by analogy with old
ones. Words formed with -able and -scape exist in such great numbers in
English because these endings, which were originally bound to specific
words (measurable, landscape), were reinterpreted as suffixes that could be
removed and reattached to any stem (touchable, moonscape).
Phonological and analogical change are the internal mechanisms
through which novel forms are incorporated into a language. By examining
a sequence of documents within one language lineage from several
different points in the past—inscriptions in, say, classical Latin, late vulgar
Latin, early Medieval French, later Medieval French, and modern
French—linguists have defined virtually all the sound changes and analogical
shifts in the evolution of French from Latin. Regular, systematic
rules, applicable also to other cases of language change in other languages,
explain most of these shifts. But how do linguists replay these shifts
“backward” to discover the origins of modern languages? How can we
reconstruct the sounds of a language like Proto-Indo-European, for which
there are no documents, a language spoken before writing was invented?
“Hundred”: An Example of Phonetic Reconstruction
Proto-Indo-European words were not reconstructed to create a dictionary
of Proto-Indo-European vocabulary, although they are extraordinarily
useful in this way. The real aim in reconstruction is to prove that a list of
daughter terms are cognates, descended from the same mother term. The
reconstruction of the mother term is a by-product of the comparison, the
proof that every sound in every daughter word can be derived from a
sound in the common parent. The first step is to gather up the suspected
daughters: you must make a list of all the variants of the word you can find
in the Indo-European languages (table 2.1). You have to know the rules of
phonological change to do even this successfully, as some variants of the
word might have changed radically in sound. Just recognizing the candidates
and making up a good list can be a challenge. We will try this with
the Proto-Indo-European word for “hundred.” The Indo-European roots
for numbers, especially 1 to 10, 100, and 1,000, have been retained in
almost all the Indo-European daughters.
Our list includes Latin centum, Avestan satem, Lithuanian šimtas, and
Old Gothic hunda- (a root much like hunda- evolved into the English
word hundred). Similar-looking words meaning “hundred” in other
Indo-European languages should be added, and I have already referred to
the French word cent, but I will use only four for simplicity’s sake. The four
words I have chosen come from four Indo-European branches: Italic,
Indo-Iranian, Baltic, and Germanic.
The question we must answer is this: Are these words phonetically
transformed daughters of a single parent word? If the answer is yes, they
are cognates. To prove they are cognates, we must be able to reconstruct
an ancestral sequence of phonemes that could have developed into all the
documented daughter sounds through known rules. We start with the
first sound in the word.
The initial [k] phoneme in Latin centum could be explained if the parent
term began with a [k] sound as well. The initial soft consonants ([s] [sh])
Table 2.1
Indo-European Cognates for the Root “Hundred”
Branch Language Term Meaning
Celtic Welsh cant hundred
Old Irish ce¯t hundred
Italic Latin centum hundred
Tocharian TochA känt hundred
TochB kante hundred
Greek Greek ε′κατο′ν hundred
Germanic Old English hund hundred
OldHighGerm. hunt hundred
Gothic hunda 100, 120
OldSaxon hunderod (long) hundred
Baltic Lithuanian s˘imtas hundred
Latvian simts hundred
Slavic OldChurchSlav. su˘to hundred
Bulgarian sto hundred
Anatolian Lycian sñta unit of 10 or 100
Indo-Iranian Avestan satem hundred
OldIndic s´atám hundred
28 Chapter 2
Reconstructing a Dead Language 29
in Avestan satem and Lithuanian šimtas could have developed from a
Proto-Indo-European word that began with a hard consonant [k], like
Latin centum, since hard sounds generally tend to shift toward soft sounds
if they change at all. The reverse development ([s] or [sh] to [k]) would be
very unlikely. Also, palatalization and sibilation (shifting to a ‘s’ or ‘sh’
sound) of initial hard consonants is expected in both the Indic branch, of
which Vedic Sanskrit is a member; and the Baltic branch, of which Lithuanian
is a member. The general direction of sound change and the specific
conventions in each branch permit us to say that the Proto-Indo-European
word from which all three of these developed could have begun with ‘k’.
What about hunda? It looks quite different but, in fact, the h is expected—
it follows a rule that affected all initial [k] sounds in the Germanic branch.
This shift involved not just k but also eight other consonants in Pre-
Germanic.6
The consonant shift spread throughout the prehistoric PreGermanic
language community, giving rise to a new Proto-Germanic
phonology that would be retained in all the later Germanic languages, including,
ultimately, English. This consonant shift was described by and
named after Jakob Grimm (the same Grimm who collected fairy tales) and
so is called Grimm’s Law. One of the changes described in Grimm’s Law
was that the archaic Indo-European sound [k] shifted in most phonetic
environments to Germanic [h]. The Indo-European k preserved in Latin
centum shifted to h in Old Gothic hunda-; the initial k seen in Latin caput
‘head’ shifted to h in Old English hafud ‘head’; and so on throughout the
vocabulary. (Caput>hafud shows that p also changed to f, as in pater>fater).
So, although it looks very different, hunda- conforms: its first consonant
can be derived from k by Grimm’s Law.
The first sound in the Proto-Indo-European word for “hundred” probably
was k. (An initial [k] sound satisfies the other Indo-European cognates
for “hundred” as well.)7
The second sound should have been a vowel,
but which vowel?
The second sound was a vowel that does not exist in English. In ProtoIndo-European
resonants could act as vowels, similar to the resonant n in
the colloquial pronunciation of fish’n’ (as in Bob’s gone fish’n’). The second
sound was a resonant, either *m. or *n. , both of which occur among the
daughter terms being compared. (An asterisk is used before a reconstructed
form for which there is no direct evidence.) M is attested in the
Lithuanian cognate šimtas. An m in the Proto-Indo-European parent
could account for the m in Lithuanian. It could have changed to n in Old
Indic, Germanic, and other lineages by assimilating to the following t or
d, as both n and t are articulated on the teeth. (Old Spanish semda ‘path’
changed to modern Spanish senda for the same reason.) A shift from an
original m to an n before a t is explicable, but a shift from an original n to
an m is much less likely. Therefore, the original second sound probably was
m. . This consonant could have been lost entirely in Sanskrit satam by yet
another assimilative tendency called total assimilation: after the m changed
to n, giving *santam, the n was completely assimilated to the following t,
giving satam. The same process was responsible for the loss of the [k]
sound in the shift from Latin octo to modern Italian otto ‘eight’.
I will stop here, with an ancestral *k’m. -, in my discussion of the ProtoIndo-European
ancestor of centum. The analysis should continue through
the phonemes that are attested in all the surviving cognates to reconstruct
an acceptable ancestral root. By applying such rules to all the cognates,
linguists have been able to reconstruct a Proto-Indo-European sequence
of phonemes, *k’m. tom, that could have developed into all the attested phonemes
in all the attested daughter forms. The Proto-Indo-European root
*k’m. tom is the residue of a successful comparison—it is the proof that the
daughter terms being compared are indeed cognates. It is also likely to be
a pretty good approximation of the way this word was pronounced in at
least some dialects of Proto-Indo-European.
The Limitations and Strengths of Reconstruction
The comparative method will produce the sound of the ancestral root and
confirm a genetic relationship only with a group of cognates that has evolved
regularly according to the rules of sound change. The result of a comparative
analysis is either a demonstration of a genetic connection, if every
phoneme in every cognate can be derived from a mutually acceptable parental
phoneme; or no demonstrable connection. In many cases sounds may
have been borrowed into a language from a neighboring language, and
those sounds might replace the predicted shifts. The comparative method
cannot force a regular reconstruction on an irregular set of sounds. Much
of the Proto-Indo-European vocabulary, perhaps most of it, never will be
reconstructed. Regular groups of cognates permit us to reconstruct a ProtoIndo-European
root for the word door but not for wall; for rain but not for
river; for foot but not for leg. Proto-Indo-European certainly had words for
these things, but we cannot safely reconstruct how they sounded.
The comparative method cannot prove that two words are not related,
but it can fail to produce proof that they are. For example, the Greek god
Ouranos and the Indic deity Varuna had strikingly similar mythological
attributes, and their names sound somewhat alike. Could Ouranos and
30 Chapter 2
Reconstructing a Dead Language 31
Varuna be reflexes of the name of some earlier Proto-Indo-European god?
Possibly—but the two names cannot be derived from a common parent by
the rules of sound change known to have operated in Greek and Old
Indic. Similarly Latin deus (god) and Greek théos (god) look like obvious
cognates, but the comparative method reveals that Latin deus, in fact,
shares a common origin with Greek Zéus.8
If Greek théos were to have a
Latin cognate it should begin with an [f] sound (festus ‘festive’ has been
suggested, but some of the other sounds in this comparison are problematic).
It is still possible that deus and théos were historically related in some
irregular way, but we cannot prove it.
In the end, how can we be sure that the comparative method accurately
reconstructs undocumented stages in the phonological history of a language?
Linguists themselves are divided on the question of the “reality” of
reconstructed terms.9
A reconstruction based on cognates from eight IndoEuropean
branches, like *k’m. tom-, is much more reliable and probably
more “true” than one based on cognates in just two branches. Cognates in
at least three branches, including an ancient branch (Anatolian, Greek,
Avestan Iranian, Old Indic, Latin, some aspects of Celtic) should produce
a reliable reconstruction. But how reliable? One test was conceived by Robert
A. Hall, who reconstructed the shared parent of the Romance languages
using just the rules of sound change, and then compared his
reconstruction to Latin. Making allowances for the fact that the actual
parents of the Romance languages were several provincial Vulgar Latin
dialects, and the Latin used for the test was the classical Latin of Cicero
and Caesar, the result was reassuring. Hall was even able to reconstruct a
contrast between two sets of vowels although none of the modern daughters
had retained it. He was unable to identify the feature that distinguished
the two vowel sets as length—Latin had long vowels and short vowels, a
distinction lost in all its Romance daughters—but he was able to rebuild a
system with two contrasting sets of vowels and many of the other, more
obvious aspects of Latin morphology, syntax, and vocabulary. Such clever
exercises aside, the best proof of the realism of reconstruction lies in several
cases where linguists have suggested a reconstruction and archaeologists
have subsequently found inscriptions that proved it correct.10
For example, the oldest recorded Germanic cognates for the word guest
(Gothic gasts, Old Norse gestr, Old High German gast) are thought to be
derived from a reconstructed late Proto-Indo-European *ghos-ti- (which
probably meant both “host” and “guest” and thus referred to a relationship
of hospitality between strangers rather than to one of its roles)
through a Proto-Germanic form reconstructed as *gastiz. None of the
known forms of the word in the later Germanic languages contained the
i before the final consonant, but rules of sound change predicted that the
i should theoretically have been there in Proto-Germanic. Then an archaic
Germanic inscription was found on a gold horn dug from a grave in
Denmark. The inscription ek hlewagastiz holitijaz (or holtingaz) horna tawido
is translated “I, Hlewagasti of Holt (or Holting) made the horn.” It
contained the personal name Hlewagastiz, made up of two stems, Hlewa‘fame’
and gastiz ‘guest’. Linguists were excited not because the horn was
a beautiful golden artifact but because the stem contained the predicted i,
verifying the accuracy of both the reconstructed Proto-Germanic form
and its late Proto-Indo-European ancestor. Linguistic reconstruction had
passed a real-world test.
Similarly linguists working on the development of the Greek language
had proposed a Proto-Indo-European labiovelar *kw
(pronounced [kw-])
as the ancestral phoneme that developed into Greek t (before a front
vowel) or p (before a back vowel). The reconstruction of *kw
was a reasonable
but complex solution for the problem of how the Classical Greek
consonants were related to their Proto-Indo-European ancestors. It remained
entirely theoretical until the discovery and decipherment of the
Mycenaean Linear B tablets, which revealed that the earliest form of
Greek, Mycenaean, had the predicted kw
where later Greek had t or p before
front and back vowels.11
Examples like these confirm that the reconstructions
of historical linguistics are more than just abstractions.
A reconstructed term is, of course, a phonetic idealization. Reconstructed
Proto-Indo-European cannot capture the variety of dialectical
pronunciations that must have existed more than perhaps one thousand
years when the language was living in the mouths of people. Nevertheless,
it is a remarkable victory that we can now pronounce, however stiffly,
thousands of words in a language spoken by nonliterate people before
2500 BCE.
The Lexicon: How to Reconstruct Dead Meanings
Once we have reconstructed the sound of a word in Proto-Indo-European,
how do we know what it meant? Some archaeologists have doubted the
reliability of reconstructed Proto-Indo-European, as they felt that the
original meanings of reconstructed terms could never be known confi-
dently.12
But we can assign reliable meanings to many reconstructed
Proto-Indo-European terms. And it is in the meanings of their words that
we find the best evidence for the material culture, ecological environment,
32 Chapter 2
Reconstructing a Dead Language 33
social relations, and spiritual beliefs of the speakers of Proto-IndoEuropean.
Every meaning is worth the struggle.
Three general rules guide the assignment of meaning. First, look for the
most ancient meanings that can be found. If the goal is to retrieve the
meaning of the original Proto–Indo–European word, modern meanings
should be checked against meanings that are recorded for ancient cognates.
Second, if one meaning is consistently attached to a cognate in all language
branches, like hundred in the example I have used, that is clearly the
least problematic meaning we can assign to the original Proto-IndoEuropean
root. It is difficult to imagine how that meaning could have
become attached to all the cognates unless it were the meaning attached to
the ancestral root.
Third, if the word can be broken down into roots that point to the same
meaning as the one proposed, then that meaning is doubly likely. For example,
Proto-Indo-European *k’m. tom probably was a shortened version of
*dek’m. tom, a word that included the Proto-Indo-European root *dek’m. ‘ten’.
The sequence of sounds in *dek’m. was reconstructed independently using
the cognates for the word ten, so the fact that the reconstructed roots for
ten and hundred are linked in both meaning and sound tends to verify the
reliability of both reconstructions. The root *k’m. tom turns out to be not
just an arbitrary string of Proto-Indo-European phonemes but a meaningful
compound: “(a unit) of tens.” This also tells us that the speakers of
Proto-Indo-European had a decimal numbering system and counted to
one hundred by tens, as we do.
In most cases the meaning of a Proto-Indo-European word changed
and drifted as the various speech communities using it became separated,
centuries passed, and daughter languages evolved. Because the association
between word and meaning is arbitrary, there is less regular directionality
to change in meaning than there is in sound change (although some semantic
shifts are more probable than others). Nevertheless, general meanings
can be retrieved. A good example is the word for “wheel.”
“Wheel”: An Example of Semantic Reconstruction
The word wheel is the modern English descendant of a PIE root that had
a sound like *kw
ékw
los or *kw
ekw
lós. But what, exactly, did *kw
ékw
los mean in
Proto-Indo-European? The sequence of phonemes in the root *kw
ékw
los
was pieced together by comparing cognates from eight old Indo-European
languages, representing five branches. Reflexes of this word survived in
Old Indic and Avestan (from the Indo-Iranian branch), Old Norse and
Old English (from the Germanic branch), Greek, Phrygian, and Tocharian
A and B. The meaning “wheel” is attested for the cognates in Sanskrit,
Avestan, Old Norse, and Old English. The meaning of the Greek cognate
had shifted to “circle” in the singular but in the plural still meant “wheels.”
In Tocharian and Phrygian the cognates meant “wagon” or “vehicle.”
What was the original meaning? (table 2.2).
Five of the eight *kw
ékw
los cognates have “wheel” or “wheels” as an attested
meaning, and in those languages (Phrygian, Greek, Tocharian A &
B) where the meaning drifted away from “wheel(s),” it had not drifted far
(“circle,” “wagon,” or “vehicle”). Moreover, the cognates that preserve the
meaning “wheel” are found in languages that are geographically isolated
from one another (Old Indic and Avestan in Iran were neighbors, but neither
had any known contact with Old Norse or Old English). The meaning
“wheel” is unlikely to have been borrowed into Old Norse from Old
Indic, or vice versa.
Some shifts in meaning are unlikely, and others are common. It is common
to name a whole (“vehicle,” “wagon”) after one of its most characteristic
parts (“wheels”), as seems to have happened in Phrygian and
Tocharian. We do the same in modern English slang when we speak of
someone’s car as their “wheels,” or clothing as their “threads.” A shift in
meaning in the other direction, using a word that originally referred to the
whole to refer to one of its parts (using wagon to refer to wheel), is much
less probable.
The meaning of wheel is given additional support by the fact that it has
an Indo-European etymology, like the root for *k’m. tom. It was a word created
from another Indo-European root. That root was *kw
el-, a verb that
meant “to turn.” So *kw
ékw
los is not just a random string of phonemes reconstructed
from the cognates for wheel; it meant “the thing that turns.”
This not only tends to confirm the meaning “wheel” rather than “circle” or
“vehicle” but it also indicates that the speakers of Proto-Indo-European
made up their own words for wheels. If they learned about the invention
of the wheel from others they did not adopt the foreign name for it, so the
social setting in which the transfer took place probably was brief, between
people who remained socially distant. The alternative, that wheels were
invented within the Proto-Indo-European language community, seems
unlikely for archaeological and historical reasons, though it remains possible
(see chapter 4).
One more rule helps to confirm the reconstructed meaning. If it fits
within a semantic field consisting of other roots with closely related reconstructed
meanings, we can at least be relatively confident that such a word
34 Chapter 2
Reconstructing a Dead Language 35
Table 2.2
Proto-Indo-European Roots for Words Referring to Parts of a Wagon
PIE Root Word Wagon Part Daughter Languages
*kw
ekw
los (wheel) Old Norse hvēl ‘wheel’; Old English hweohl ‘wheel’;
Middle Dutch wiel ‘wheel’; Avestan Iranian axtra‘wheel’;
Old Indic cakrá ‘wheel, Sun disc’; Greek kuklos
‘circle’ and kukla (plural) ‘wheels’; Tocharian A kukal
‘wagon’; Tocharian B kokale ‘wagon’
*rot-eh2
- (wheel) Old Irish roth ‘wheel’; Welsh rhod ‘wheel’; Latin rota
‘wheel’; Old High German rad ‘wheel’; Lithuanian rātas
‘wheel’; Latvian rats ‘wheel’ and rati (plural) ‘wagon’;
Albanian rreth ‘ring, hoop, carriage tire’; Avestan
Iranian ratha ‘chariot, wagon’; Old Indic rátha ‘chariot,
wagon’
*ak*s-, or (axle) Latin axis ‘axle, axis’; Old English eax ‘axle’; Old High
German *ha
ek*s- ahsa ‘axle’; Old Prussian assis ‘axle’;
Lithuanian ašís ‘axle’; Old Church Slavonic osiˇ ‘axle’;
Mycenaean Greek a-ko-so-ne ‘axle’; Old Indic áks*a
‘axle’
*ei-/*oi-, or (thill) Old English ār- ‘oar’; Russian vojë ‘shaft’; Slovenian oje
‘shaft’; Hittite h2
ih3
s or hišša- ‘pole, harnessing shaft’;
Greek oisioi* ‘tiller, rudderpost’; Avestan Iranian aēša
‘pair of shafts, plow-pole’; Old Indic i-s*a ‘pole, shaft’
*wégˆheti- (ride) Welsh amwain ‘drive about’; Latin veho- ‘bear, convey’;
Old Norse vega ‘bring, move’; Old High German wegan
‘move, weigh’; Lithuanian vežù ‘drive’; Old Church
Slavonic vezo
˙
‘drive’; Avestan Iranian vazaiti ‘transports,
leads’; Old Indic váhati ‘transports, carries,
conveys’. Derivative nouns have the meaning “wagon”
in Greek, Old Irish, Welsh, Old High German, and Old
Norse.
could have existed in Proto-Indo-European. “Wheel” is part of a semantic
field consisting of words for the parts of a wagon or cart (table 2.2). Happily,
at least four other such words can be reconstructed for Proto-IndoEuropean.
These are:
1. *rot-eh2
-, a second term for “wheel,” with cognates in Old Indic
and Avestan that meant “chariot,” and cognates that meant “wheel”
in Latin, Old Irish, Welsh, Old High German, and Lithuanian.
2. *aks- (or perhaps *h2
eks-) ‘axle’ attested by cognates that had not
varied in meaning over thousands of years, and still meant “axle”
in Old Indic, Greek, Latin, Old Norse, Old English, Old High
German, Lithuanian, and Old Church Slavonic.
3. *h2
ih3
s- ‘thill’ (the harness pole) attested by cognates that meant
“thill” in Hittite and Old Indic.
4. *wégheti, a verb meaning “to convey or go in a vehicle,” attested by
cognates carrying this meaning in Old Indic, Avestan, Latin, Old
English, and Old Church Slavonic and by cognate-derived nouns
ending in *-no- meaning “wagon” in Old Irish, Old English, Old
High German, and Old Norse.
These four additional terms constitute a well-documented semantic field
(wheel, axle, thill, and wagon or convey in a vehicle) that increases our confidence
in reconstructing the meaning “wheel” for *kw
ékw
los. Of the five
terms assigned to this semantic field, all but thill have clear Indo-European
etymologies in independently reconstructed roots. The speakers of ProtoIndo-European
were familiar with wheels and wagons, and used words of
their own creation to talk about them.
Fine distinctions, shades of meaning, and the word associations that
enriched Proto-Indo-European poetry may be forever lost, but gross
meanings are recoverable for at least fifteen hundred Proto-Indo-European
roots such as *dekm- ‘ten’, and for additional thousands of other words derived
from them, such as *km. tom- ‘hundred’. Those meanings provide a
window into the lives and thoughts of the speakers of Proto-Indo-
European.
Syntax and Morphology: The Shape of a Dead Language
I will not try to describe in any detail the grammatical connections between
the Indo-European languages. The reconstructed vocabulary is
most important for our purposes. But grammar, the bedrock of language
36 Chapter 2
Reconstructing a Dead Language 37
classification, provides the primary evidence for classifying languages and
determining relationships between them. Grammar has two aspects: syntax,
or the rules governing the order of words in sentences; and morphology,
or the rules governing the forms words must take when used in
particular ways.
Proto-Indo-European grammar has left its mark on all the IndoEuropean
languages to one degree or another. In all the Indo-European
language branches, nouns are declined; that is, the noun changes form
depending on how it is used in a sentence. English lost most of these declinations
during its evolution from Anglo-Saxon, but all the other languages
in the Germanic branch retain them, and we have kept some
use-dependent pronouns (masculine: he, his, him/feminine: she, hers, her).
Moreover, most Indo-European nouns are declined in similar ways, with
endings that are genetically cognate, and with the same formal system of
cases (nominative, genitive, accusative, etc.) that intersect in the same way
with the same three gender classes (masculine, feminine, neuter); and
with similar formal classes, or declensions, of nouns that are declined in
distinctive ways. Indo-European verbs also share similar conjugation classes
(first person, second person or familiar, third person or formal, singular,
plural, past tense, present tense, etc.), similar stem alterations (run-ran,
give-gave), and similar endings. This particular constellation of formal
categories, structures, transformations, and endings is not at all necessary
or universal in human language. It is unique, as a system, and is found
only in the Indo-European languages. The languages that share this grammatical
system certainly are daughters of a single language from which
that system was inherited.
One example shows how unlikely it would be for the Indo-European
languages to share these grammatical structures by random chance. The
verb to be has one form in the first-person singular ([I] am) and another in
the third-person singular ([he/she/it] is). Our English verbs are descended
from the archaic Germanic forms im and ist. The Germanic forms have
exact, proven cognates in Old Indic ásmi and ásti; in Greek eimí and estí;
and in Old Church Slavonic jesmıˇ and jestuˇ. All these words are derived
from a reconstructable Proto-Indo-European pair, *h1
e’smi and *h1
e’sti. That
all these languages share the same system of verb classes (first person,
second person or familiar, and third person), and that they use the same
basic roots and endings to identify those classes, confirms that they are
genetically related languages.
Conclusion: Raising a Language from the Dead
It will always be difficult to work with Proto-Indo-European. The version
we have is uncertain in many morphological details, phonetically idealized,
and fragmentary, and can be difficult to decipher. The meanings of
some terms will never be fully understood, and for others only an approximate
definition is possible. Yet reconstructed Proto-Indo-European captures
key parts of a language that actually existed.
Some dismiss reconstructed Proto-Indo-European as nothing more
than a hypothesis. But the limitations of Proto-Indo-European apply
equally to the written languages of ancient Egypt and Mesopotamia,
which are universally counted among the great treasures of antiquity. No
curator of Assyrian records would suggest that we should discard the palace
archives of Nineveh because they are incomplete, or because we cannot
know the exact sound and meaning of many terms, or because we are uncertain
about how the written court language related to the ‘real’ language
spoken by the people in the street. Yet these same problems have convinced
many archaeologists that the study of Proto-Indo-European is too
speculative to yield any real historical value.
Reconstructed Proto-Indo-European is a long, fragmentary list of words
used in daily speech by people who created no other texts. That is why it is
important. The list becomes useful, however, only if we can determine
where it came from. To do that we must locate the Proto-Indo-European
homeland. But we cannot locate the Proto-Indo-European homeland until
we first locate Proto-Indo-European in time. We have to know when it
was spoken. Then it becomes possible to say where.
38 Chapter 2