Classical versus Late Biblical Hebrew: Two Statistical Case Studies

When the Israelites arrive at Mount Sinai, God declares to them:

^{שמות יט:ו} וְאַתֶּם תִּהְיוּ לִי מַמְלֶכֶת כֹּהֲנִים וְגוֹי קָדוֹשׁ.

^{Exod 19:6} But you shall be to Me a kingdom of priests and a holy nation.

The word mamlakha (מַמְלָכָה—here in the construct form) is a standard term for “kingdom” in the Bible, appearing in various forms 117 times.^[1] A synonym from the same Hebrew root is malkhut (מַלְכוּת), which appears 91 times in the biblical corpus. For example, the term appears in Psalm 145, describing God’s kingdom:

^{תהלים קמה:יג} מַלְכוּתְךָ מַלְכוּת כׇּל עֹלָמִים וּמֶמְשַׁלְתְּךָ בְּכׇל דּוֹר וָדֹר.

^{Ps 145:13} Your kingdom is an eternal kingdom; your dominion is for all generations.

It may be initially tempting to find some fine distinction of meaning between these two words. For example, we might posit that mamlakha is associated with physical kingdoms, whether Israel or its neighbors, while malkhut can refer to the concept of kingship—this may be substantiated by pointing to its וּת (-út) ending, which is often associated with something abstract.^[2]

But this distinction does not hold. For example, in Deuteronomy the king of Israel sits on כִּסֵּא מַמְלַכְתּוֹ “the throne of his kingdom,” using mamlakha:

^{דברים יז:יח} וְהָיָה כְשִׁבְתּוֹ עַל כִּסֵּא מַמְלַכְתּוֹ וְכָתַב לוֹ אֶת מִשְׁנֵה הַתּוֹרָה הַזֹּאת עַל סֵפֶר מִלִּפְנֵי הַכֹּהֲנִים הַלְוִיִּם.

^{Deut 17:18} When he is seated on the throne of his kingdom, he shall have a copy of this Torah written for him on a scroll by the levitical priests.

The same expression is used in the late biblical book of Esther about Ahasuerus, king of Persia, but with malkhut:

^{אסתר א:ב} בַּיָּמִים הָהֵם כְּשֶׁבֶת הַמֶּלֶךְ אֲחַשְׁוֵרוֹשׁ עַל כִּסֵּא מַלְכוּתוֹ אֲשֶׁר בְּשׁוּשַׁן הַבִּירָה.

^{Esth 1:2} In those days, when King Ahasuerus occupied the throne of his kingdom in the fortress Shushan.

Similarly, when the ghost or spirit of Samuel isbrought up by the necromancer from Endor, she tells Saul that his rulership (mamlakha) is at an end:

^{שמואל א כח:יז} ... וַיִּקְרַע יְ־הֹוָה אֶת הַמַּמְלָכָה מִיָּדֶךָ וַיִּתְּנָהּ לְרֵעֲךָ לְדָוִד.

^{1 Sam 28:17} …YHWH has torn the rulership out of your hands and has given it to your fellow, to David.

Dealing with the same idea, replacing a king, in the book of Esther, Ahasuerus’ advisors tell him to replace Vashti with a new queen, but uses the word malkhut:

^{אסתר א:יט} ...וּמַלְכוּתָהּ יִתֵּן הַמֶּלֶךְ לִרְעוּתָהּ הַטּוֹבָה מִמֶּנָּה.

^{Esth 1:19} … And let Your Majesty bestow her rulership upon her fellow who is more worthy than she.

The verse in Esther is likely echoing the verse in Samuel, and yet, it chooses a different term.

Linguistic scholars, such as Avi Hurvitz of Hebrew University, have explained the difference between these terms as part of biblical Hebrew’s historical development.^[3] In Classical Biblical Hebrew (pre-exilic period), mamlakha was the term for kingdom; the term malkhut came to replace this older term in the post-exilic period, becoming the standard in Late Biblical Hebrew (LBH).

Classical Versus Late Biblical Hebrew

In their 2-volume Linguistic Dating of Biblical Texts, however, Ian Young and Robert Rezetko question whether CBH and LBH are really two Hebrew dialects that can be distinguished diachronically, that is over time.^[4] In this case, they note that the term malkhut does appear in some texts that are considered to have been written in CBH, and thus cannot be considered only a late term.

For example, in the book of Samuel (CBH), when King Saul admonishes his son Jonathan against friendship with his rival, David, he says:

^{שמואל א כ:לא} כִּי כׇל הַיָּמִים אֲשֶׁר בֶּן יִשַׁי חַי עַל הָאֲדָמָה לֹא תִכּוֹן אַתָּה וּמַלְכוּתֶךָ... וְעַתָּה שְׁלַח וְקַח אֹתוֹ אֵלַי כִּי בֶן מָוֶת הוּא.

^{1 Sam 20:31} For as long as the son of Jesse lives on earth, neither you nor your kingship will be secure….

If the term מַלְכוּת only developed in a later period, they argue, how can it appear in a book supposedly from an earlier period? To this, B. Elan Dresher of the University of Toronto responded:

It is a well-attested fact in many languages that competing forms may coexist over a period of time, and thus a late form may occur sporadically in early texts, and an early form may survive in late texts.^[5]

In other words, the issue is not whether a word appears at all in an early text, but its distribution, i.e., frequency. When we look at the distribution for each of these words, we find that in the Pentateuch and Prophets (primarily CBH texts), mamlakha appears a total of 86 times while malkhut appears a total of 6 times. In contrast, in Esther, Daniel, Ecclesiastes, and Ezra-Nehemiah, books that all scholars agree are from the post-exilic period, and are written in LBH, mamlakha appears only twice while malkhut appears a total of 51 times.

Dresher’s argument pushed Young and Rezetko to modify their view:

On the face of it, it seems likely that in the history of ancient Hebrew מלכות [malḵūṯ] gradually became more predominant, and probably due to Aramaic influence.^[6]

Nevertheless, Young and Rezetko note that the term mamlakha was available to the authors of LBH texts, such that they could archaize or “age” their work by using the older word. Admittedly, we do not know whether or not scribes were cognizant of one word being/sounding older than the other, and there likely are cases of purposeful archaizing language.^[7] Nevertheless, as we are speaking of a corpus of texts all written in LBH, archaizing seems to be an unlikely explanation, especially given the statistical significance of the distribution in this case.^[8]

The Statistical Significance of Observed Differences

For the sake of simplicity, let’s imagine that the CBH and LBH corpora are two ancient books that tell a single story using either “kingdom” or “monarchy” whenever they want to say “a state that is ruled by a king or a queen.” Altogether, the books use either of the words 150 times, with “monarchy” appearing 81 times, 54%, and “kingdom” appearing 69 times, 46%. Since both books tell a continuous story, one might first assume that they are both written by the same author who likes “monarchy” just a little more than “kingdom.” Statisticians would call this “the null assumption.”

When we look at each book separately, however, “monarchy” is chosen in the CBH book only 3 out of 48 times it references a state, that is 6.25%; statisticians would call this fB1 (i.e., the frequency of the chosen word being “Monarchy” in Book 1) = 0.0625. In the LBH book, it appears 78 out of 102 times, that is 76.47%; fB2 = 0.7647.

At this point, literary scholars and mathematicians would use these frequencies to question the correctness of the null assumption—that both books were written by the same author in the same time. Statisticians, however, would distinguish these frequencies from probabilities, considering them as data, and would then calculate the p-value.

The P-Value

To calculate the P-value, the statistician begins with an assumption that the choice between “monarchy” and “kingdom” is made at random, and that the probability of choosing either word is the same for both books, like a coin flipping experiment with the probability of heads ("monarchy") being 0.54.

Let’s consider an extreme case and make all 48 occurrences in the CBH Book “kingdom.” The p-value would be 0. This doesn’t mean that it is impossible that one author wrote both books, only that under the null assumption, a more radical result (in math language “sample”) is impossible to find.

Let’s consider an extreme case and make all 48 occurrences in the CBH Book “kingdom.” The p-value would be 0 to more than 16 decimal places. It is impossible to find a more extreme case, and the probability of getting exactly 0 under the null assumption is 0 (to computer machine accuracy). This doesn’t mean that it is impossible that one author wrote both books, only that under the null assumption, the probability of getting a result (in stat language, "sample") that is as radical or more radical is vanishingly small.

Going back to the actual numbers for CBH and LBH, if the word malkhut is chosen at 54% altogether in the corpora, and the distribution is random, then in CBH, malkhut should appear between 25 and 26 times of the 48 appearances of either malkhut or mamlakha. It would be very rare to see it only 3 times, a little over 6%. (Try flipping a coin 48 times and counting in how many sets of 48 flips it falls on “tail” less than 4 times: Rarely.)

Typically, a p-value less that .05 is considered as indicating a "significant" result, and one would say that the null assumption is seriously challenged. The p-value for the data 3/48 and 78/102, given the null assumption that CBH and LBH choose randomly between malkhut and mamlakha, turns out to be much smaller than .0001, and thus very highly significant.^[10]

The Weakness of “Random” as an Explanation

Now, this does not answer the biblical scholar's question, but it does mean that scholars claiming that the choice between malkhut and mamlakha stays the same throughout the Hebrew Bible must find a cogent explanation of why malkhut is used so much more often in LBH than in CBH. Specifically, it is not sufficient to note that biblical authors did not choose words at random. The question is whether the assumption of randomness is a reasonable approximation.

Furthermore, even if most choices can be explained, say by textual analysis, as long as no good reasons for the choice can be found in many of the instances, it may be possible to accept the random model as an approximation for these instances. In this case, we may ask how few such instances are needed that the difference between usage in CBH and LBH remains statistically significant. For example, how many of the instances of mamlakha in LBH would need to be explained (that is, non-random) in order for the difference in usage to remain statistically significant when applied only to the unexplained data?

Assuming all other instances are unexplained, then if 68 or fewer of the 78 mamlakhas in LBH are explained (that is, 10 or more remain unexplained) then the significance level remains less than .01 (approximately). This only requires that when we can find no explanation for the outcome, we may take the choice as a personal preference that can be reasonably approximated by a random model.

“Don’t” in LBH

Another example of a CBH to LBH development, noted by Avi Hurwitz, is the way the Bible expresses “don’t,” “you may not,” “it is not permitted/possible,” etc.^[11] In CBH, this is expressed with the word לא (lo) followed by an imperfect/future tense. For example, when commanding Israelites not to change any of the laws, Deuteronomy writes:

^{דברים יג:א} אֵת כָּל הַדָּבָר אֲשֶׁר אָנֹכִי מְצַוֶּה אֶתְכֶם אֹתוֹ תִשְׁמְרוּ לַעֲשׂוֹת לֹא תֹסֵף עָלָיו וְלֹא תִגְרַע מִמֶּנּוּ.

^{Deut 13:1} Be careful to observe only that which I enjoin upon you: do not add to it nor take away from it.

In contrast, in LBH, the prohibition uses אֵין (ʾēn) plus an infinitive construct, i.e., “one may not” instead of “you may not.” This is seen, e.g., in the similarly phrased statement in Ecclesiastes:

^{קהלת ג:יד} יָדַעְתִּי כִּי כָּל אֲשֶׁר יַעֲשֶׂה הָאֱלֹהִים הוּא יִהְיֶה לְעוֹלָם עָלָיו אֵין לְהוֹסִיף וּמִמֶּנּוּ אֵין לִגְרֹעַ...

^{Eccl 3:14} I realized, too, that whatever God has brought to pass will recur evermore: Nothing can be added to it and nothing taken from it…

A sharper example of a contrasting pair is in two texts that prohibit entrance to a place, each using the common Hebrew root ב.ו.א, “to come or enter.” Deuteronomy forbids a creditor from entering someone’s house and taking their only clothes as collateral:

^{דברים כד:י} כִּי תַשֶּׁה בְרֵעֲךָ מַשַּׁאת מְאוּמָה לֹא תָבֹא אֶל בֵּיתוֹ לַעֲבֹט עֲבֹטוֹ.

^{Deut 24:10} When you make a loan of any sort to your countryman, you may not enter his house to seize his pledge.

In Esther, however, after Mordechai learns of Haman’s decree, he dresses in sack cloth but doesn’t enter the palace because of the prohibition:

^{אסתר ד:ב} וַיָּבוֹא עַד לִפְנֵי שַׁעַר הַמֶּלֶךְ כִּי אֵין לָבוֹא אֶל שַׁעַר הַמֶּלֶךְ בִּלְבוּשׁ שָׂק.

^{Esth 4:2} And he came up to in front of the palace gate; for one may not enter the palace gate wearing sackcloth.

The phrasing ʾēn plus infinitive appears a total of eight times in LBH texts.^[12] In contrast, it appears only once in a CBH text:

^{שמואל א ט:ז} וַיֹּאמֶר שָׁאוּל לְנַעֲרוֹ וְהִנֵּה נֵלֵךְ וּמַה נָּבִיא לָאִישׁ כִּי הַלֶּחֶם אָזַל מִכֵּלֵינוּ וּתְשׁוּרָה אֵין לְהָבִיא לְאִישׁ הָאֱלֹהִים מָה אִתָּנוּ.

^{1 Sam 9:7} “But if we go,” Saul said to his servant, “what can we bring the man? For the food in our bags is all gone, and there is nothing we can bring to the man of God as a present. What have we got?”

On account of this one CBH example, Daniel Fredericks argued that this grammatical structure can’t be LBH.^[13] This is an overstatement for two reasons: First, as noted by Hurwitz, in Samuel the phrase is descriptive: it isn’t that they aren’t allowed to give the prophet a gift, or that it isn’t proper behavior, but that they simply don’t have anything to give him. Thus, usage here cannot be considered parallel to that of LBH.

Even putting this aside, the question is really a statistical one, as Hurvitz notes:

Statistically, the LBH corpus comprises but a small fraction of the entire Hebrew Bible. Consequently, if eight out of nine occurrences of אֵין [ʾēn] + inf. are documented in this limited corpus, one is more than justified in regarding the phenomenon under examination here as a feature typically characteristic of LBH.^[14]

Indeed, Hurvitz has statistical support, as the chance of ʾēn + infinitive appearing 8 times out of 9 in the LBH corpus if the grammatical form was randomly distributed across the LBH + CBH corpora is approximately 0.000045 (or 4.5 in 100,000).^[15] Thus, we must conclude that the form was not randomly distributed, a point that demands explanation. Without any cogent explanation, the statistical analysis strongly supports the conclusion that the form is an LBH development.

A Useful Model

Statistical analysis itself cannot provide an explanation for why the statistically significant differences in distributions may have occurred. Nonetheless, long historical development has shown that modelling numerical outcomes as random occurrences can offer powerful tools to shed light on scholarly questions.

Specifically, statistical analysis shows that observed differences between CBH and LBH in numeric measures of several linguistic features can not be explained on the basis of random variation. Thus, to claim that the bodies of writings are not distinguishable requires cogent reasons, independent of occurrence in CBH or LBH, that would explain why the linguistic features differ. Without such convincing reasons, the statistical evidence strongly supports that Hebrew developed over time from Classical Biblical Hebrew to Late Biblical Hebrew, and that different biblical authors wrote in each.

An insightful quip attributed to the famous statistician, George E.P. Box (1919–2013) is that “all models are wrong, but some are useful.” In the two cases analyzed above, statistical modelling is quite useful in highlighting linguistic differences between texts and supporting the regnant theory that biblical Hebrew developed over time.