I: The hunter-gatherer lexicon

Our hunter-gatherer lexicon avoids terms characteristic of both hunter and farmer cultures (e.g., “to collect” may refer to collecting produce from the wild or an agricultural field). We avoid most plant and animal names that could have been introduced either before or after their domestication. This leaves mainly terms related to fire and hunting, whose inclusion in the lexicon is justified by archeological data discussed below. A priori one might expect a similar 3c/2c ratio in all semantic fields. But, as we show below, this is not the case.

II: The farmer's lexicon

Evidently, more farming than hunting terms survived in PS, and nearly all have 3c morphology. Table 3 lists 27 agricultural terms that have been archeologically dated. Verbs like “collect”, “grind” and “bake”, characterizing both agricultural and pre-agricultural communities, and animal or plant names that could have originated either before or after domestication, are not listed. The discussion below provides archeological evidence that the entries in Table 3 originate within the Neolithic or Chalcolithic societies (ca. 11,000–6,000 BP).

As of the Pre-Pottery Neolithic B (PPNB), ca. 10,500 BP, the farmer (#3.1) lived in a large village (#3.15), constructed of square houses [3], often made of straw (#3.26) reinforced [40] sun-baked mud bricks (#3.17). Indeed, straw became readily available after the Pre-Pottery Neolithic A (PPNA) wheat domestication [1], and hence its identification as an agricultural commodity.

The farmer would work in an agricultural field (#3.10, 3.11), which he would plow (#3.12), sow (#3.27) or plant (#3.25). Furrow tracts from W. Europe date to 5,500 BP [41] and must have appeared earlier in W. Asia. Cattle were domesticated in the upper Euphrates valley by 10,000 BP, spreading to Central Anatolia, Mesopotamia and the S. Levant around 8,500 BP [42]. This may mark the onset of ox-traction and hence the use of the scratch-plow (ard) for plowing. The ard might have also been instrumental in installing the first irrigation systems. An early irrigation canal (#3.19), over 7 kyr old, was discovered in Choga Mami, 110 km E. of Baghdad: "It is conceivable, indeed probable, that plough cultivation accompanies irrigation agriculture in the earlier Samarra period" [48].

Tilled fields can be sown only if grain from the previous year is stored under adequate conditions. PPNA granaries (#3.2 and 3.9), about 11,300 years old, were unearthed in the Dead-Sea region near Dhraʽ, Jordan [43]. These round structures, with suspended floors for air circulation and protection from rodents, were located between residential structures that contain plant-processing installations.

The first attested wells (#3.4) were dug by Neolithic farmers on the coast of Cyprus ca. 9,200 BP [44]. The oldest well found in Israel (8700–8400 BP) is in the undersea site of Atlit-Yam [45,46]. A Pottery Neolithic (PN) well, dated to ca. 8,300 BP, was found at Sha‘ar Hagolan in the Jordan Valley [47]. Thus wells were yet another important innovation of the Neolithic. The II-ˀ morphology of *biˀr “well” (#3.4) is also attested in the PS/3c verbal root for drawing water (šˀb, #3.22), possibly because they have originated in the same period.

The earliest mineral-tempered ceramics from Tell Sabi Abyad (N. Syria) was likely introduced for cooking (#3.5 and 3.24), leading to a "culinary revolution" nearly 9,000 years ago [49], when (PS/2c) “baking” [8] and “roasting” (#1.11) were supplemented by cooking. Only later was pottery utilized for storing liquids.

Herding began after goat and sheep domestication, either in the Neolithic or as late as the Chalcolithic [50]. Livestock (#3.8) was often composed of mixed sheep and goat herds that optimize vegetation exploitation. This contrasts with their non-overlapping habitats in Nature [2], suggesting that *ḍaˀn “livestock” (#3.8) is a post-agricultural innovation. The herd was lead by a stockbreeder (#3.18) to a drinking trough (#3.21). Due to lactose intolerance, milk utilization has begun rather late, at the end of the Chalcolithic or the Early Bronze [41]. However, recent fatty acid analysis of pottery sherds suggests that processed milk was used as early as 8,500 BP [51]. In agreement with this, there is no PS name for milk but there is one for butter (#3.14), a low lactose milk product.

Fermenting wine (#3.13) was made from grapes (#3.3) already in the Neolithic: jars from Georgia (in the Caucasus), dating to ca. 8,000 BP, were shown to contain resinated wine deposits, as have 7,300 BP sherds from the Zagros Mountains in Iran ([52], Chap. 4). The popular resin was from the terebinth tree (#3.6), Pistacia atlantica [52]. The earliest known winery (6100 BP) was recently found in an Armenian cave site [53]. The prominence of viticulture in the Fertile Crescent is echoed in toponyms derived from *karm, *karān “vineyard” (#3.16): Mt. Karmel in N. Israel and Karānā in Upper Mesopotamia (perhaps Tell ar-Rimāh, 60 km W. of Nineveh). Although a dry wasteland today, the high concentration of archeological mounds suggests it has once been fertile land ([52] p. 173).

Beer (#3.23a) was the most popular intoxicating (#3.23b) drink in Mesopotamia. Until recently, the earliest evidence for beer (from ca. 5,500 BP) was found in the Sumerian trading post of Godin Tepe in Iran [54]. But recent evidence from Göbekli Tepe (S.E. Turkey) suggests that beer was brewed already in the PPNB [55].

Millet (#3.7) was domesticated in N.E. China about 10,000 years ago [56]. It made its way to the Black-Sea region around 7,000 BP [57], just in time to be included in the PS lexicon. Because it came from outside W. Asia, its PS name depicts the domesticated plant and not its wild progenitor.

Exceptions to the regularity demonstrated in Table 3 namely, PS agricultural terms with 2c morphology, are hard to find. We have found two such examples (as compared with 27 entries in Table 3), and even these are not clearly exceptions. (i) It is suggested that PS/PAA/2c *marr “a hoe”, derived from the 2c root *mrr “to hoe”, originates within a PAA farming lexicon [16]. The noun is either Nostratic, ND #1482 [20], or a "wandering-word" borrowed into many languages from Sumerian [58]. If the verb *mrr has itself been borrowed by Sumerian from PAA [16], then its original meaning was “to dig” [59], an activity practiced by hunter-gatherers much before the agricultural era. (ii) The PS verb for herding, *rˁy, is 2c although herding postdates ungulate domestication that occurred after the transition to agriculture. However, in some Chadic dialects it means “to chase, follow”, DAE #663 [18]. This may go back to gazelle chases, involving gathering herds by "effective utilization of drives and surrounds" [60], including the utilization of huge traps known as "desert kites" [61]. Thus if gathering domesticated herds is the behavioral continuation of gathering herds of gazelles, the continued use of the same verb for depicting it could be understandable.

III: Word Frequency Analysis

The study thus far focused on statistics of culturally specific terms (hunting vs. farming) that could be correlated with archeology. These are mostly low frequency lexemes, hence not belonging to the "kernel lexicon". We now analyze the kernel lexicon of an Ancient Semitic text, the Hebrew Bible, bisecting it into its 2c vs. 3c components. We consider verbs, because their 2c vs. 3c origin can be determined rather mechanically (see Methods), allowing processing a large number of verbs. Yet they constitute the only part of speech whose Zipf plot is similar to that of the whole corpus [25]. The black circles in Figure 1 depict the frequency-rank dependence, f(r), for the BH (non-Aramaic) verbal roots with f ≥ 10 [31]. It indeed appears that there are two regimes here [26], with α₀=1.07and α₁≈2 (dashed lines). The switchover occurs around f = 20, so that the kernel of BH is characterized by f ≥ 20.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Frequency-rank plot for Hebrew verbal roots appearing more the 10 times in the Bible (black circles)

[31]. Black dashed lines: fits of the total frequency,f₀(r) , to Eq. (1) with A₀=16,000 and α₀=1.07or A₁=1.3×10⁶ andα₁=2. Blue triangles represent 2c/BH verbal roots in their broadest definition (see Methods). They were extracted from Lester's list [31] and collected in Table S5 of the Supporting Information. The non-2c verbs there were defined as 3c, and their frequencies are depicted by the red triangles. Blue line is a fit to Eq. (2) with t = 7.8 kyr andA_2c=3. The rate function k(f₀) from Eq. (3b) has B = 0.55 kyr^-1 and β = 0.13, as deduced from Figure 3a of Ref [29].. Red line is a fit to Eq. (4) with the same parameters, except forA_3c=0.09.

https://doi.org/10.1371/journal.pone.0083780.g001

Table 4 lists the 20 most frequent BH verbs. These are indeed very generic, not related to any specific culture or occupation, and likely used with high frequency in any natural language. Of these, 13 are 2c, far exceeding the fraction of 2c verbs in the Bible. This agrees with the observation that the most frequently used words in English tend to be short [24] (and also of Old English origin). According to Zipf's "principle of least effort" long words got shortened for ease of use. We have no evidence that 2c/PS verbs were shortened from 3c verbs, and thus suggest another mechanism leading to the prevalence of 2c verbs in Table 4.

As recently shown [29], frequently used words (actually, meanings) are replaced (by other words of the same meaning) less often than the less frequent ones. Thus if the 2c stratum indeed predated the 3c one, the frequently used 2c lexemes may have simply survived replacement during the subsequent 3c era. This is supported by their frequency-rank dependence in Figure 1. As opposed to the total BH verbs with α ≈ 1, the 2c/BH verbs (collected in Table S5) exhibit an observably larger Zipf exponent (α_2c=1.28), whereas the high frequency 3c verbs have a smaller α_3c=0.82(linear fits not shown). This might be explainable by the 2c > 3c transition: while the 2c language was alive, 2c words of a given meaning were depleted at the same rate as alternate 2c lexemes were generated, and the language maintained its steady-state with the usual exponent α ≈ 1. After the 2c era has ended, 2c roots were no longer created only eliminated. Because less frequently used words decay faster, α_2c increased with time.

One may turn this into a quantitative method for dating the 2c > 3c transition. Suppose that once there were only 2c words, and at some time (t = 0) they started to be replaced with new 3c words. Assume that (up to a constant) the frequency of use of a certain verbal meaning (at least in the kernel lexicon),f₀(r) , is an inherent property of human language and hence not strongly time-dependent. We thus equate it with the frequency of the total verb distribution (black circles in Figure 1). Therefore, at t = 0 the 2c frequency-rank relation wasf_2c(r,0)=A_2cf₀(r), where A_2c is some constant. We expect A_2c>1 if the 2c corpus was once used more frequently than today (or: with a smaller vocabulary each word is used more frequently).

Subsequently, the frequency of 2c verb utilization decayed exponentially with time:

The rate coefficient,k(f₀) , is a unique function of the (time-independent) verb meaning frequency,f₀(r). Eventually, after some time t that we opt to determine,f_2c(r,t) reached the values observed in the Biblical lexicon (blue triangles in Figure 1).

A similar equation was suggested by Leiberman et al. [28], see their supporting Eq. (3). It can be interpreted in two ways. Firstly, like in radioactive decay: the decay of any particle is instantaneous, and one counts the number of particles surviving by time t. This is useful when texts from different epochs are available, as in [28], but not for the analysis of a single text. However, words need not disappear instantaneously from the lexicon. Their use may gradually decrease over time until they eventually become obsolete, and this allows applying the above equation even when text(s) from just a single period are available.

To proceed, a functional form for k(f) is required. We adopt Pagel et al. [29] power-law rate coefficient for lexical replacement. It depends on the part of speech, but otherwise is rather universal for the IE family, and possibly for all languages [22]:

From the correlation line for English verbs in their Figure 3a, one estimates B = 0.55 kyr^-1 and β = 0.13. We do not vary these parameters in fitting our data. However, in Ref. [22] the frequencies are per million words of text, whereas in the Bible there are about 305,500 Hebrew words (a ratio of 3.27), hence what we insert into Eq. (2) is:

Adjusting t and A_2c to fit the 2c data (blue triangles), we obtain the blue line agreeing with the data over the whole frequency range, even where it deviates from Zipf's law, Eq. (1). This gives t = 7.8 kyr. Adding the presumed age of BH, ca. 3 kyr, gives 10.8 kyr for the 2c > 3c transition, agreeing nicely with the onset of agriculture.

An analogous model may describe the 3c verbs, which experience exponential growth rather than decay:

Of course, such growth cannot go on indefinitely, but we assume the time-depth is not large enough to observe saturation. With exactly the same parameters as above (excepting A_3c) we obtain the red line in Figure 1, which fits the 3c data at high frequencies. Thus 10.8 kyr BP marks both the end of the 2c era and the onset of 3c morphology.

As a check for the robustness of this analysis, we return to the "burning verbs" discussed in Subsec. I(b) above. We find 10 such verbs in BH (some of these are PS, and thus appear in Table 1). Their frequency-rank relation is shown in Figure 2 (circles). The deviation from Zipf's law, dashed line, is even larger and its exponent α = 2.5. Of these verbs, 6 are 2c (triangles). Although a rather small collection, we can repeat our analysis. Remarkably, when Eqs. (2) and (3b) are applied to the data, with exactly the same parameters as in Figure 1, we obtain either the dashed-dotted line (when the dashed line is used asf₀), or the full line (when the circles are used asf₀). Thus the "burning verbs" behave like the entire BH verb population, both yielding the same date for the 2c > 3c transition.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Frequency-rank plot for BH verbal roots that are near synonyms of “to burn” (circles), and their 2c subgroup (triangles).

Dashed line represents Eq. (1) with A = 900 and α = 2.5. The application of Eqs. (2) and (3b) to it gives the dash-dotted line, whereas their application to the data itself (circles) gives the full line. Parameters are identical to those in Figure 1. The frequencies of the 10 roots were taken from a Biblical Concordance, and are as follows: śrp 117 (3c), ḥry 94 (2c), bˁr 61 (3c), yṣt 30 (2c), kby 24 (2c), lhṭ 11 (3c), yqd 9 (2c), dlq 4 (3c), qly 3 (2c), kwy 2 (2c).

https://doi.org/10.1371/journal.pone.0083780.g002