1.0. Introduction

1The scope and methods of scribal education in ancient Israel, and even the very existence of scribal education in ancient Israel, have long been discussed and debated. Various scholars have taken different tacks in approaching this question, invoking, among other things, comparative evidence, biblical texts, archaeological data (and its absence), and palaeographical evidence.1 In this paper I would like to suggest that the spelling practices evident throughout the Hebrew Bible can themselves suggest something of the extent of such scribal education—both its extent and its limitations. To get there, however, we need to begin with a discussion of early writing systems, and then turn to the alphabetic revolution. As we will see, considering these developments from the perspective of writing systems can open entire vistas in considering the question of scribal education.

2.0. The Invention of Writing: Historical and Conceptual Framework

2We start with the invention of writing, not so much for historical context as much as for conceptual context: for orientation as to how writing systems can work, and how they did work in the ancient Near East. While the history of writing is a well-trod field, the question of how writing systems represent language (s) is far less studied (Daniels 2018, 3). We now know that writing was invented at least a few times in world history, certainly by the Maya and likely by the Chinese, in addition to its invention in the Near East, apparently in southern Mesopotamia in the late fourth millennium BCE, among the Sumerians. This Mesopotamian invention, like cylinder seals and other contemporaneous ideas and artifacts, thence spread rapidly to Egypt.

3Denise Schmandt-Besserat (1992) argued that the first writing emerged from tokens in envelopes, a system developed entirely for accounting purposes. Some of the specifics of her theory rest on very little data (Zimansky 1993), but it does seem clear from the texts themselves that the invention of writing was primarily for bookkeeping and accounting (Robinson 1995, 11–12; Woods 2020). It cannot be a coincidence, however, that writing developed in the same time and place (late fourth-millennium Sumer) as the first states, and so the association between writing and statecraft should not be ignored either (Scott 2017, 139–42).

4How did the early writing system work? The first writing consisted essentially of numbers, metrics, and common nouns (for the system, see Nissen, Damerow, and Englund 1993). There is no grammar; there were no verbs, adjectives, or prepositions, and even the plural on nouns, for example, is not marked. Only around 2800 BCE, nearly half a millennium after writing was invented, was the meš sign developed to indicate plurality of nouns. Until then, a text would simply say the equivalent of ‘donkey 1’ and ‘donkey 8’. The spoken language did, of course, distinguish between singular and plural nouns, but the earliest writing does not reflect this.

5The writing in the earliest texts is so distant from speech, that Englund (1998, 73–81) denies that it is provable that the language being written is Sumerian. Others (Steinkeller 1995; Cooper 2004) argue that there is just a little bit of phonetic writing, enough to prove that the script reflects Sumerian (probably enough to prove that the script was invented in order to write Sumerian, rather than borrowed for this purpose from some other linguistic group; Daniels 2018, 93–94). But this discussion establishes the key point. It is conventional, and useful, to think of writing as potentially reflecting two different levels of the underlying language: the sounds of the language, which are meaning-independent, and the meaning conveyed by those sounds. That is, the writing can reflect the phonemes (the sound-units) or the morphemes (meaning-units). Early cuneiform says nothing about the phonemic level altogether, and instead maps directly onto the morphemic level. In fact, Sumerian, Egyptian, and Mayan writing all use semantic classifiers, which have no phonetic value whatsoever, and whose meaning is entirely on the levels of semantics and grammar.

6Writing begins as primarily conveying meaning, and only marginally conveying sound. This was possible because the earliest cuneiform consists almost entirely of ‘morphograms’, signs representing complete, individual morphemes (up to the size of a word).2 Early writing was not meant to reflect spoken language at all, but to do jobs for which spoken language is actually quite poor, primarily lists and bureaucracy. And these it did very well: Cooper (2004, 77–78) observes that as soon as the idea struck, there was a full-blown system in place:

The idea that commodities, titles, names, and transaction types could be represented graphically led almost immediately to the elaboration of an entire system of signs, and, in contrast to the very simple enumeration of the earlier numerical tablets, we are confronted with an irrationally exuberant metrological system with over a dozen different sets of numerals for recording amounts of various kinds of discrete objects, weights, area, liquid and dry measures and time.3

7Morphograms comprise only a “partial writing system” (DeFrancis 1989). Theoreticians of writing argue that a script that is entirely morphographic is a dead end (DeFrancis 1989; Liberman 1992, 120; Rogers 2005; Joyce 2013, 70), so fortunately, cuneiform later develops into a morphophonemic system, utilising morphograms—signs for individual, complete morphemes or words—alongside phonograms—signs for sounds.

8For this move, so crucial in the history of writing, the needed step was the development of signs that reflect the sounds of the language, rather than the meanings. Actually, what was needed was a whole system of such signs. The motivation for this, however, was not to begin to write sentences. Still very much in the world of accounting, the motivation was to keep track of people’s names (Schmandt-Besserat 2015). Thus, with these developments, which took place between 2900 and 2700 in Mesopotamia, it was possible to write not only ‘8 donkey’, but ‘Kushim 8 donkey’, bringing bookkeeping practices fully into the third millennium BCE.

9Both cuneiform and early Egyptian accomplish this transformation through the rebus principle. Once complete, this move—which created the Egyptian writings of ‘son’ as a duck, both śɜ, ‘mother’ as a vulture, both mwt, and ‘to enter’ as a cormorant, both ʿq—allows the writing of anything in the language.

10We now have a “full writing system” (DeFrancis 1989). At this point, anything can be written, not only phrases such as ‘8 donkeys’, but sentences like ‘the sight of the royal boat on the great green sea yesterday filled me with feelings otherwise provoked only by the acquisition of 8 donkeys’.

3.0. The Alphabet as Orthographic Revolution

11Let us turn now to the invention of the alphabet. The steps involved must have been something like the following:

1. Introspection regarding the phonemes in the inventor’s language. This is not easy. Studies have shown that illiterate people (children who have not yet learned to read, and adults who never did) generally lack phonemic awareness, and cannot break down the word /bag/ into [b], [a], and [g]. This is one of the stronger reasons to doubt that the inventors of the alphabet were illiterate, although this alone cannot resolve that question.

2. Concomitantly, the decision was made to include only consonants—a decision which takes some linguistic sophistication, since the categories of consonants and vowels are not all that stable. This, too, suggests, prior literacy, as the exclusion of vowels was likely inspired by the practices of Egyptian writing (Daniels 2006, 375–76). In any event, this decision yielded approximately 28 or 29 consonantal phonemes to account for.

3. Somehow keeping the resulting list of consonantal phonemes in mind, the inventor then utilised the acrophonic principle to assign a graphic sign to each phoneme. The rules of this game were: for each, find a common noun that was easily drawable and easily recognisable that started with a given sound; from now on, if you see that picture, it means only the first sound. Arm is a good candidate; freedom is not. House works; to fly does not. Here again, the Egyptian writing system was probably helpful.

4. Finally, this had to be taught to others.

12Steiner (2015) observed that the earliest corpus of alphabetic texts, the inscriptions from Serabit el-Khadim, are radical in their phonemic transparency. They map one-to-one, sign to sound, with no attention paid to the semantic level of the language at all.

13Even the concept ‘word’ is foreign to the corpus of the Sinai inscriptions, as there is no indication in these texts where one word may begin or end. Early alphabetic inscriptions are much less clear than Egyptian at indicating where words end. This is linguistically interesting (see further below), but for now I want to draw attention to the implications for reading. Readers of English quickly identify familiar words. This works not just in the case of short words. Eye-tracking studies show that readers’ eyes settle on each word, typically once, at a spot relatively close to the middle of the word, before moving on to the next one.

14This is empirical evidence of what we all know intuitively: we read word by word, not syllable by syllable or letter by letter. In fact, studies have shown that good second- and fourth-grade readers read familiar object words such as <MAN>, <CAR>, <DOG>, <BALL>, <HAT> as quickly as they named digits such as 2, 6, 3, 5, 4, 9: in other words, they are reading the words as single units rather than by sounding out the separate letters (Ehri and Wilce 1983; Ehri 2005). In theorising reading today, a regnant theory is the ‘dual-route’ theory, which posits that the brain can pronounce words correctly either by mapping graphemes to phonemes, or by recognising words; the key distinction between these routes is familiarity (Share 2008). Already in kindergarten, children are taught to identify ‘sight words’. For adults, the vast majority of words are sight words.

15Reading hieroglyphs certainly relies on quick recognition of the boundaries of each word, along with rapid word recognition. This is enabled by the presence of classifiers, which besides classifying, also mark the end of the word, and regularised spelling. One may not spell nfrw ‘beautiful young people’ () the same way as one spells nfrw ‘foundations of a house’ (). See also these pairs, taken from Faulkner’s dictionary:

16In these cases and many others, orthographic practices convey visual clues regarding semantics that are not found in the spoken language: these pairs are homophonic, but not homographic.4 This type of orthographic distinction was not possible in the Sinai texts, however, where the words are not separated from each other in any graphic way, and the written signs convey nothing other than the sounds. Just how aberrant this is within writing systems can be seen from the fact that some scholars of writing systems (e.g., Liberman 1992), not cognizant of the Sinaitic texts, have argued that to be usable, a script cannot be pitched at the narrow phonemic level, but must be oriented at the more abstract phonological or morphemic levels.

17There are thus serious disadvantages to this type of system, foremost among them the difficulty of fluid reading. I assume this means that either these texts had to be read aloud or they were not meant to be read at all. Paul Saenger (1982; 1997) showed that the norm in the classical world in antiquity was for oral reading, and he connected that method of reading to the practice of writing with no spaces. The key insight is that when one reads aloud, the aural experience allows the reader to make sense of the text even through the graphic representation of the text is difficult to decode. In the Middle Ages—beginning in the eighth century on the Irish margins of Europe, and reaching France and Spain in the twelfth and thirteenth centuries—reading became silent, and writing concomitantly gained spacing. This type of approach to the question of word divisions relates the practice of writing, especially the peritextual elements in the written text, to the practices of reading. This framework needs to be incorporated into the story of Northwest Semitic writing (Boyarin 1993; Dobbs-Allsopp 2012). For the early alphabetic texts, with no word dividers and a purely phonemically-oriented writing system, reading would have been a slow, laborious task, likely only possible when read out loud.

18The other possibility is that these texts were not meant to be read at all. Recall that at Serabit el-Khadim the texts were primarily votive inscriptions, on objects dedicated to the gods, and especially to Ḥatḥor, lady of the turquoise. Other early alphabetic texts are graffiti, or labels—all plausibly never meant to be read again, except perhaps by gods, who are presumably untroubled by the lack of word dividers.

19Whether the alphabetic texts were read aloud or not read at all, this highlights a major drawback of this new script: its unreadability by a scribe sitting by himself. It is common to laud the invention of the alphabet not only as an act of genius, but as a transformative development with the ability to change the world of texts (Logan 1986; see Koller 2019 for discussion). While this may be true of the idea of the alphabet, the radical alphabetic script of the Sinai inscriptions actually goes ‘too far’, so to speak: while the total phonological transparency of these texts is remarkable and seems admirable, the script has thereby forfeited much of what makes reading efficient. It may not be surprising, then, that the alphabet did not quickly spread as a technology, and instead seems to have remained extremely marginal for more than half a millennium after its invention (contra Sass 2004–2005).

4.0. Spelling in the Hebrew Bible and Its Implications

20With these two polar models—the fully sign-to-meaning model of earliest cuneiform, and the fully sign-to-sound model of the earliest alphabet—in mind, we can approach the orthographic practices in evidence in the Hebrew Bible to see what we can learn from them. Other scholars have noted that spelling practices can tell us a lot about scribal training. For example, Rollston (2010; 2015) has argued that the hallmarks of Hebrew orthography evident in the epigraphic record are “synchronic consistency and diachronic development” (2010, 103) This is entirely accurate for the topics he studies, primarily תומא האירקה , matres lectionis. Here I want to emphasise some elements that are consistent both synchronically and diachronically: historical spellings and other morphophonemic spellings.

21To take a simple example discussed by Steiner (2015): the ʾalef in the word ןאצ ‘sheep, goats’ is etymological, but of merely historical relevance after the quiescence of the glottal stop and the resulting lengthening of the /a/ vowel, which was then subjected to the Canaanite shift. This chain has to have been completed prior to the end of the Late Bronze Age, when the last-mentioned sound change stopped being productive—and in fact the word ṣunu, with an <u> for /o/, is attested already in Amarna Canaanite. The same is true for שאר ‘head’ (but see Qimron 2003 for a suggested modification).

22So how did Israelite scribes know to spell the word ןאצ with an ʾalef, rather than spelling it like רוב ‘cistern’, סוכ ‘cup’, רוד ‘generation’, and the like? This requires education, in the form of sheer memorisation. This is actually the way we all learned to spell correctly: out of fear of our teachers, or a desire to do well in school, we studied for spelling quizzes and memorised the difference between to, too, and two, and so on.

23Many other standard spellings in Biblical Hebrew are not historical, but reflect meanings rather than sounds. In other words, these spellings are morphemic rather than phonemic. Some common examples are the yod in many plural forms with suffixes ( ךירבד ‘your words’, וניכלמ ‘our kings’, וישנא ‘his men’) and the yod in construct or suffixed forms of -ay- segolates ( י ֵ ךני ‘your wine’, ע ֵ ןי ‘eye, spring [construct]’), and the usual distinctions made between הרק ‘happen, meet’ and ארק ‘call, read’, or ארב ‘create’ and הרב ‘eat; choose’.5

24Closely related are the ʾalef in a form such as ל ָ ש ֵ תא ‘to bear’ and a form such as באת ְַ נד ָ ה ‘will perish (fpl)’—where the ʾalef has fully quiesced, but is useful for visually conveying the verbal root. These spellings are morphemic in the sense that the writing of the ʾalef preserves visually the existence of the three-letter root which no longer exists phonemically. It is far easier to realise that באת ְַ נד ָ ה ‘will perish (fpl)’ is related to א ָ ב ְ ד ּ ו ‘perished (3pl)’ than it would be if it were spelled בת ְַ נד ָ ה . Similarly, there is good reason for ָ שׁאר ִ םי ‘heads’, ֵ שׁאר ִ תי ‘beginning’, and ִ ןוֹשׁאר ‘first’, to be spelled as they are, despite the non-phonemic ʾalef, which synchronically serves as a mater lectionis for four different vowels in these words: the spellings transparently link the words to שׁאר ‘head’, and therefore give the reader a graphic clue to the semantics of the word that would not otherwise be available from the phonological shape of the word alone.

25Other examples are the spelling of the plural noun ח ַ תואט / ח ַ תאט (which appears dozens of times), and the singular noun ח ַ ט ַ תא / ח ַ ט ָ תא (which appears 55 times).6 All of these have a non-phonemic ʾalef, which is nevertheless useful, as it makes the respective connections to מ ָ ל ֵ א ‘full’ and ח ָ ט ָ א ‘to sin’ transparent.

26Such spellings are clearly useful, and therefore repay the extra time and effort needed to learn them in school. The same principle is in evidence in English spelling. For example, the past tense of read is spelled <read>, homophonous with the colour <red>. Why not merge the spellings? There are two benefits to these non-phonemic spellings: (a) the avoidance of ambiguity and (b) the conveyance of semantic information by graphic association with other words. For the first, it is convenient that read and red are spelled differently, just for the sake of spelling them differently, and avoiding confusion. For the second, the spelling <read> immediately makes transparent the relationship to the verb read.

27Similarly, consider English electric, electricity, and electrician, wherein the <c> is pronounced three different ways: /k/, /s/, /ʃ/) Although we tend to complain about such inconsistencies, there is great benefit to seeing the word ‘electric’ within the two other words. Another ubiquitous example (discussed by Steiner 2015) is the plural morpheme /s/, which is pronounced differently in <dogs>, <cats>, and <horses>. It would be only slightly exaggerated to say that the grapheme <s> represents the meaning [+plural], not the sound, and is in that sense equivalent to the Sumerogram meš in Akkadian texts or the three lines marking the plural in Egyptian writing.

28The other major reason for preserving non-phonemic spellings is to visually convey semantic distinctions, as already mentioned in the example of read (≠ red) (see also Daniels 2018, 17– 18). For example, if we spelled the opposite of day <nait>, the pronunciation would be more transparent from the writing. We could do the same with the /nait/ in shining armor. But <night> contrasts with <knight>, neither of which is spelled in a way that reflects pronunciation. What is gained in the lack of phonological transparency is semantic transparency: now the sentence ‘the boy was scared of the knight’ is unambiguous, as ‘the boy was scared of the nait’ is not.7

29Comparable examples in Biblical Hebrew are the consistent distinction between אל ‘no, not’ and ול ‘to him, for him’ and between הרק ‘happen, meet’ and ארק ‘call, read’—homophonic, but semantically distinct.8 The benefit is instant recognition of the meaning; the price paid is extra quizzes in scribal school. These examples show how spelling can convey semantic distinctions.

30The spelling of Biblical Hebrew takes a step away from the radicalism of the Sinaitic inscriptions; it is not strictly oriented towards the phonology of words, but contains morphemic information, as well. Of course, Biblical Hebrew spelling is still primarily phonemic, but the morphographic elements should be noted. Phonemic spellings are the skeleton and the foundation for learning to read, but competent readers of ancient Hebrew, like those of English, would ideally arrive at the end result of word recognition (Ehri 2005).

31This necessitates a return to one question alluded to above: did Iron Age Hebrew writing represent words, unlike in the early alphabetic texts? We have, of course, the Masoretic division of the text into words by use of spaces.9 For the Iron Age itself, word dividers are well attested, but to judge from the epigraphic record, there may have been some inconsistency. Within Biblical Hebrew, single-letter prepositions are obligatorily cliticised to the following word, but two-letter prepositions can stand alone. In inscriptions, monosyllabic nouns of two or even three letters are often cliticised to the following word, especially if the word is in the construct state, such as bytdwd ‘house of David’ and brbʿr ‘son of Beʿor’.10

32Interestingly, there are exceptional cases in the other direction, as well, where ‘word dividers’ divide graphemic strings that are not normally understood to be words. In a number of cases, the third-person masculine plural object suffix is divided by a word divider from the verb or noun to which they are grammatically suffixed: wʾsḥb.hm ‘and I dragged them’ (Mesha 18), wmḥnt.hm ‘and their encampments’ (Zakkur A9), ʾrq.hm ‘their land’ (Tel Dan 10) (Millard 1970, 15; also Lehmann 2016, 44* fn. 2).

33Thus, although the idea of words was certainly established in the writing system of the Iron Age Levant, it may well have been a fuzzy category. This is intuitively true, since it is not at all obvious conceptually if the definite article is a separate word (as in English), a prefix (as in Hebrew), or somewhere in between (as in French), and the same is true for suffixed pronouns and other features of language. These are good examples of things that had to be learned in scribal school.

5.0. Extra-Biblical Evidence for the Value of Scribal Training

34As is clear, then, much of the foregoing argues for the widespread, but not fully consistent, training of the scribes who produced the biblical text. The lack of full consistency argues for the genuineness of the training, since if order had been imposed where chaos had once reigned, we would not expect to find the many exceptional spellings that we find.

35How would scribes of ancient Hebrew write if they were not forced to memorise spellings? This is fortunately an answerable question, since we have epigraphic texts written by people who were self-evidently literate, but not trained scribes. This is an instance of the rule that the less-trained the writer is in the language he is writing, the more revealing the writing will be (see Steiner 1995, 202–3). The soldier-writer of Lachish 3, for example, shows that he pronounced the 2ms ending of the perfect verbs with a final /ā/, and so writes התחלש ‘you sent’ (for BH תחלש ) and התעדי ‘you know’ (for BH .) תעדי11

36This writer is also not entirely sure about the division of words, so he writes רמאיכו ‘and because he said’ and הוהיח ‘by Yhwh’s life’ as single words. He writes דגהל ‘to say’, because he never learned the rules behind plene and defective spellings of long i. And he never learned that masculine plural nouns with the 3ms possessive suffix are written with a silent yod, a morphographic spelling of the plural, so he writes ושנא rather than וישנא for ‘his men’.12

37It will be noted that the spellings found in Lachish 3 are not foreign to the Bible, and are actually found more often the later we go within the biblical tradition. Forms like התלעפ for תלעפ occur well over a hundred times (Tur-Sinai 1940/1987, 37–42; Barr 1986, 114–31).13 Barr rightly points out that the occurrence of this spelling does not date the text in which it is found, but Tur-Sinai is surely correct in arguing that as a spelling practice, it is later than the defective תלעפ , and never replaced it in the scribal tradition of the Bible.14

38The spelling of hifʿil verbs is not consistent in the Bible, either, although here a diachronic solution does not seem to fit the data (contra Barr 1986, 81–84). The specific word used in Lachish 3, דגהל , is spelled plene, i.e., דיגהל , more than two dozen times, without a single defective example like that in Lachish 3. But there are more than seven hundred examples of the defective spelling of hifʿil verbs overall, and I can discern no pattern in their distribution.

39The spelling of the 3ms possessive suffix on masculine plural nouns as ו - rather than וי - is also encountered in the Bible (see Barr 1986, 131–37, although he combines many disparate issues into one discussion). These are scattered throughout the corpus, generally limited to the scribal tradition (the ketiv); for example, ketiv ושנא , qere א ֲ נ ָ שׁ ָָׁ֨ וי ‘his men’ (1 Sam. 23.5).15 This likely shows that the plural was sometimes written with only a final waw, not the yod-waw sequence. These waw-only spellings increase in Transitional Biblical Hebrew texts. In Ezekiel 40 alone there are some 34 examples (Ariel 2013). The first three words of Ezek. 40.22 are ketiv ונולחו ומלאו ורמתו , qere ו ְ ח ַ נוֹל ָָ֤ וי ו ְ א ֵֵֽ ל ַ מ ָ וי ְ ו ִ ת ִ֣ מ ָ ר ָ֔ וי ‘and its windows and its porches and its palm trees’. (For reasons that are not clear to me, only the first and third words are marked as qere in the Aleppo Codex, while the second is allowed to simply be revocalised.) However, ‘proper’ spelling in the Bible insists on the morphemic yod to indicate the plural, and to distinguish between the singular and the plural possessive endings (note that a spelling such as ונולח is entirely ambiguous as to number, ‘his window’ or ‘his windows’).

40It is not surprising that some of the spelling practices in evidence in Lachish 3 become more common in later biblical texts, as well. The explanation is straightforward: these are all developments away from morphemic spelling and towards phonemic spelling. In general, this is the way spelling reforms tend to move, as writers and readers more easily see the lack of phonemic transparency than they grasp the benefits of morphemic spelling.16

6.0. Conclusions

41The conclusion that emerges from this all is that the training that went into being an ancient Hebrew scribe was real, but not all that extensive (contrast Schniedewind 2019). On the one hand, the very presence of morphemic or morphophonemic spellings is indicative of scribal training, as only someone who had gone to school (whatever that school looked like) could know to write התשע with a heh, but תבתכ without one, to write ןושאר and תאטח with ʾalef, and to write וישנא with a yod. Only a scribe could know to write ןאצ with an ʾalef, or to distinguish אל from ול , and so on. On the other hand, the lack of consistency on some of these points shows that even this low bar was not always met. Certainly, there is no reason to think that scribal training in ancient Israel (or any of its neighbouring alphabet cultures) was nearly as intensive or extensive as that needed for cuneiform or hieratic scribes.