Jonathan Ben-Dov
Search for other papers by Jonathan Ben-Dov in
Current site
Google Scholar
Asaf Gayer
Search for other papers by Asaf Gayer in
Current site
Google Scholar
, and
Eshbal Ratzon
Search for other papers by Eshbal Ratzon in
Current site
Google Scholar
Open Access

In chapter 8 we discussed the reconstruction of small lacunae using letter cloning.1 That method is rather accurate because it accounts for variations in the letter-shapes when they constitute part of different combinations. On the other hand, it is also time consuming, and thus not practical for the reconstruction of large sections of text.2 A custom computer font designed for each scroll may prove more useful for that purpose. In addition, as discussed in chapter 9, several ways have been proposed in the past for estimating the column width of fragmentary scrolls using parallel text from other copies. We find the use of a tailored font fitting for this latter task, as well as for reconstructing the height of a column and estimating the amount of space occupied in several consecutive columns. This is, to our mind, a cost-effective compromise between the need for utmost precision and limitations in the amount and cost of work that could be invested. To be sure, the production of the font is time consuming, but the investment is paid off when using the font in the reconstruction.

Already in 2005 as part of the “Electronic Boethius” Project, Kevin Kiernan presented the transcription of an Old English manuscript that is now lost by using the letters from other manuscripts with similar paleographical features. That program allows replacing transcribed letters for images of letters digitally cut out of manuscripts, choosing between a variety of letters and using ligatures. This project did not, however, perform kerning for the letters.3 More recently, several scholars used a custom-made font, imitating the handwriting of the scribe, a method we will also use. In 2016 Bronson Brown-deVost created a font for the reconstruction of 4Q204 1 col. i. He measured all the letters in this column and chose the ones closer to the average for his font. In cases where a letter was not represented in this scroll, he completed it from the inventory of letters in 4Q203, which was probably written by the same scribe. He then adjusted the kerning of the required pairs of letters to what is seen on the fragments. Since handwriting is never as consistent as a computer font, Brown-deVost still had to adjust individual pairs using image manipulation software.4 In their 2017 publication, Ben-Dov, Stökl Ben Ezra, and Gayer used a font imitating the Cryptic A script to reconstruct a cave 4 copy of Serekh haEdah. This font was designed by a graphic designer, rather by replicating letters from the images of the scroll’s fragments.5 In 2018, Sacha Stern and Jay Birbeck reconstructed a Genizah fragment in a similar way. They chose the most representative letters from the images (without measuring) and then fixed the kerning to match that of the manuscript. According to their description, they too had to adapt certain letters to the image even after the kerning.6 Similarly in 2019 Ratzon reconstructed 4Q208 with a custom-made font. She had to complete the missing letters from the similar script of 4Q28 (4QDeuta), although in this case it was not written by the same scribe.7 For the creation of the font used in this book we followed Brown-deVost’s guidelines, although like Stern and Birbeck we did not find it necessary to measure every letter but rather chose what seemed to our trained eyes as the average size. While the use of custom fonts has started making inroads in scholarly practice, this chapter offers a rigorous theoretical discussion for the validation of the method, the preferred ways to apply it, and its margins of error.

Admittedly, long reconstructed sections are subject to some margin of error, due to two kinds of factors. The first arises from our lack of knowledge of the properties of the scroll, such as the width of the columns, their height, etc. The second is related to the fact that handwriting is not as uniform as a computer font: the shape of every letter and its relation to the surrounding letters vary; spacing between lines is not always constant throughout the height of a sheet;8 and the size and style of handwriting changes through various parts of the scroll. While the former is unavoidable, and will remain unknown for any method of reconstruction – whether letter space counting or letter cloning or fonts – the effect of the latter factor can be, to some extent, controlled.

Control is achieved both by means of the method of font creation and by limiting the kind of scrolls that can be securely reconstructed using this method. Unlike with letter cloning, when using a font all occurrences of the same letter are uniform. Therefore, this method can only be accurately used for the reconstruction of scrolls with comparatively stable features, such as non-cursive script, consistent letter-shapes, scrolls written by an expert scribe, using straight lines and columns, etc. In other cases, a font can be used for visualization only, but cannot be used to accurately reconstruct the measurements of the scroll.9

In the following chapter we describe a method for controlled use of a custom computer font for the reconstruction of specific scrolls. This methodology is intended for reducing the error to a minimum, by choosing the best letters, adapting the relations between chosen pairs of letters to the scribe’s practice, and properly using the font on a digital canvas. The suggested methodology may reduce the margin of error but it cannot eliminate it. We empirically checked the precision of such reconstructions on comparatively intact scrolls (1QIsaa, 1QS, and 11QPsa [= 11Q5]). The experiment involves, first, using custom-made computer fonts for typing the text of scrolls, and then comparing the results with the actual length of the lines, columns, and a sequence of columns of these scrolls. The present chapter recounts the mode of operation in the experiment and its results, as well as a discussion of these results. A detailed description of the experiment appears in Appendix 1.

1 Designing the Font

As in other chapters of this book, with the rapid change in technology, and the wide variety of font design software, we do not provide a detailed manual how to run the procedure with a specific program. We do provide general instructions for the process, emphasizing its methodological issues.

1.1 Choosing Letters

The first step in font creation is choosing the letters from the scroll which will serve as the basis for the font. While the most accurate way for choosing the letters is measuring a substantial number of examples of any letter and then choosing one that is the closest to the average,10 the main point of creating a font is to avoid this kind of effort. Such an effort is more easily done when the font is created semi-automatically, as in Appendix 2. When creating the font manually, there are several features to consider while choosing the letters:

  1. A practical condition for the success of creating the glyphs is that the chosen letters are complete.

  2. After gaining close acquaintance with the manuscript the scholar should choose letters that seem typical.

  3. Dark letters with a distinct black color are preferable as their outline and various strokes are more easily processed. Some programs overcome this obstacle better than others.11

  4. The handwriting of each scribe is not consistent throughout the manuscript, and in order to avoid choosing non-representative letters, it is best to examine as many fragments as possible. If the scaling of the images of a specific fragment or an entire scroll is problematic, the scholar must choose all the letters from the same image. For the same reason, fragments that went through shrinkage should not be included in the reservoir of letters.

  5. In cases of rare letters such as ט or ס, and in cases of highly fragmentary scrolls, there may not be many complete examples of every letter. One then has to settle for the best-preserved letter rather than the most typical one. For some scrolls, not every letter is documented. In these cases, the missing letters will be taken from another scroll with the closest script possible. For example, in the case of the very fragmentary 4Q418a, we had to use the close handwriting of 4Q418.12

  6. Letters should preferably be isolated enough in order not to include traces of the surrounding letters in the glyph. It is easier to find spacious writing in fragments coming from the middle of the sheet, whereas in the first and last columns the script tends to be denser.

  7. Designing a font involves not only designing the letters but – just as important – also the size of the space glyph. While font-design software will assign a standard width for the space between words, adapting its width to the particular scroll significantly increases the font’s precision. Unlike the choice of the letters, it is difficult to estimate the width of the space since it greatly varies throughout the scroll.13 In this case, we recommend measuring a substantial number of examples and computing the average size, which will be fixed as the size of the space glyph in the custom font. While the average does not correspond to any one specific space in the scroll, it is functional when reconstructing large stretches of text, as the average evens out. This hypothesis was tested in the experiment described below.

1.2 Creating the Glyphs

For the creation of the glyphs, it is extremely important to copy all letters from evenly scaled images (see chapter 4). After that, no zooming in or out is allowed, in order not to tamper with the scaling. All copied letters should be pasted on the same file (see figure 23) with their background removed and all additional ink marks, whether from neighboring letters or otherwise, removed.

Figure 23
Figure 23

An image file of the letters of 4Q417.

Graphics: Einat Tamir

The image file with the letters is uploaded to the font program as a background file. The next step is to connect each glyph on this file to the letter it represents, identifying each image of a letter as a character in the Hebrew alphabet. When connecting a glyph to its letter, it is placed inside a grid, setting the exact position of the letter within the rectangle dedicated to it. It is important that the left and right boundaries of the grid be attached to the glyphs, so that their relations to other letters will not be disturbed later. Letters should be suspended from the upper line, rather than standing on a straight baseline.

1.3 Kerning

After all letters and the space glyph have been accurately placed in their grids, the next stage is adapting the kerning, i.e. the distance between specific pairs of characters. These small spacing differences add up to a significant error when reconstructing a whole line, not to mention a whole column. They should thus be designed with utmost care. As mentioned above, each glyph is designated with its own rectangle, which will define the space it occupies on the canvas. Without any adaptation, the right-most edge of the next typed letter will begin after the previous rectangle ends. In handwriting, however, the glyphs do not occupy the entire rectangle, allowing other letters to penetrate to their designated space. Normally, scribes of the DSS tend to align their letters according to certain rules, depending on the characteristics of types of letters. For example, letters with diagonal strokes overlap other letters to some extent (e.g. א, ‪צ‬); pairs of letters, comprising a letter with a base (like ב, ‪כ‬, נ, ‪פ‬), followed by a letter with a right leg (such as ד, ‪ה‬, ו, ‪ז‬, ח, ‪ר‬, ת), touch or even overlap in a way that the leg “stands” on the base of the previous letter (for example: בו).

While general rules may be formulated for the kerning of many DSS scripts, one should also fine-tune the kerning to conform with the handwriting of the specific scribe. The initial kerning should be tested on a digital canvas, by typing letters in the custom font in a separate layer above the layer of the (scaled!) image. If, after adjusting the size of the font, each typed letter does not fully cover its corresponding original letter, the scholar should suspect that additional kerning adjustment is required. The suspicion will be confirmed only if the relationship between the two letters is consistent throughout the scroll. Otherwise, the discrepancy indicates mere variations in the handwriting of the ancient scribe.

When using the font on a digital canvas, some programs carry out their own automatic kerning that may interfere with the adaptation to the scribe’s handwriting. This feature should be turned off.14 After the kerning is ready, when the font is used in the reconstruction, it is important not to adjust every letter or pair of letters individually, but to set the parameters based on the preserved fragments, and to maintain them in the reconstructed text. This is the only way to preserve its precision.

The kerning is the final stage of font creation. After it is accomplished, the font can be exported from the font-creator program. Designers should indicate their copyright in the font file to secure credit and proper licensing. The font is now ready to be exported and then installed on any computer for the work to continue.15

Work on the kerning is time consuming, not only due to the need to examine and re-examine the relations between letters in the manuscript, but also due to a variety of practical problems in the capacity of the programs to adjust right to left (RTL) orientation. The community of Hebrew font designers is not large, and finding help online is not easy. When a good program is found, later versions do not always consider the small market of RTL font designers, and a new learning curve is required from the designer. The semi-automatic production of fonts described in Appendix 2 may solve some of these problems.16

2 Margin of Error

As in every reconstruction, a margin of error exists, which means that the suggested result does not produce one absolute reconstruction, but rather a range of possible solutions. In Appendix 1 we describe an experiment conducted to check the validity of the custom font as a working method. The experiment also provides the measure of error that can be expected as a result when applying this method. The experiment was carried out in three stages, checking the reconstruction of three sizes pertinent for the verification: the width of a line, the height of a column, and the space required for long textual units.

We tested the method on three comparatively intact scrolls, type-setting their known text in the newly-produced custom font and comparing the required space with the length of the same text in the original scrolls. The difference between the measure of the reconstructed and the original text equals the margin of error. The results vary slightly between the three scrolls, but remain in the same approximate range. While detailed results are provided in Appendix 1, here we provide general impressions.

In the first stage of the experiment, we reconstructed the length of a line based on a few preserved letters at its beginning. The error for all three scrolls was around 4% with a standard deviation of approximately 3–4%. As these scrolls differ in many features, this persistent result may be used as the margin of error for the reconstruction of other scrolls as well.

In the second stage, we reconstructed an entire column based on three preserved lines at its top. We then measured the height difference between the reconstructed column and the original column by means of two units: height in centimeters and number of lines. The error for the reconstruction of the height of a column in centimeters was between 6–8% in all three scrolls, but with a standard deviation of approximately 5% for 1QIsaa and 11Q5 and of nearly 10% for 1QS. When the height of the column is measured in number of lines the error for 1QIsaa and 1QS was around 7% with a standard deviation of approximately 7%, but the error for 11Q5 was less than 3% with a standard deviation of a bit more than 4%. The higher margin of error in the former two scrolls is due to many intervening factors. In 1QS one finds many vacats, significant variations in font size, and many interlinear corrections. The scribe of 1QIsaa was more careful, but the length of the lines in the respective columns was less consistent. In contrast, the scribe of 11Q5 was very careful, keeping highly organized script and columns. When reconstructing the column’s height in other scrolls, the scholar must first examine the above-mentioned intervening factors in order to assess the margin of error.

The third stage tested the reconstruction of longer textual sections spanning several columns. It had two versions. The first version assumed that the text of the scroll is known as well as the measurements of the first column, but the size of the rest of the columns is unknown; it had to be copied from the first one. The second version assumed that the height and width of all columns were known. For both versions we typed the text of up to eleven consecutive columns, and then measured the space that the text occupied from the beginning of the first column to the final word of each one of the eleven columns. We then compared the difference between the reconstruction and the original scroll. In the first version the average error was quite high: approximately 10% for 1QIsaa and over 30% for 1QS with a standard deviation of 3% and 17% respectively.17 In contrast, when the width of the columns is known from elsewhere, the error is only 4–5% with a standard deviation of 2–3%. Obviously, the more information a scholar has about the reconstructed columns, the more sound the reconstruction is. The variance from the average did not spread equally between columns. The relative error for the first few columns may vary, but the error stabilizes after a few columns. Therefore, while it is difficult to predict the error for a reconstruction of a few columns, a reconstruction of over five or six columns will most likely be as accurate as the reconstruction of the number of lines in a column, that is around 5–8%.

The meaning of these results varies for each purpose. If the purpose of a reconstruction of an entire line based on a few characters is to determine if there was enough space for one additional word, the margin of error will probably not allow such resolution. Alternatively, in material reconstruction, knowing the width of a column with a precision of up to one word is quite good. In contrast, in key columns, on which much of the reconstruction is dependent, this error can be more significant, as explained in chapter 2 and Appendix 3. Eventually, using a font for the reconstructions of long sections of relatively consistent scrolls is the best option, but the results of any reconstruction should be taken as a range of options, and conclusions should be drawn accordingly.


We are grateful to Dr. Bronson Brown-deVost (Göttingen) for kindly sharing with us the initial knowledge how to design a font using the Bird Font and Microsoft Volt programs, and for his patient instructions during later stages.


For the difference between the reconstruction of small lacunae and of longer sections see Zuckerman, “Dynamics of Change,” 13–19, and Herbert, Reconstructing Biblical Dead Sea Scrolls, 7–11.


Kevin Kiernan, “The Source of the Napier Fragment of Alfred’s Boethius,” Digital Medievalist 1 (2005). doi:


Brown-deVost, “4QEnc,” 60–84. We received this point from him through private correspondence.


Ben-Dov, Stökl Ben Ezra, and Gayer, “Reconstruction of a Single Copy.”


Sacha Stern and Jay Birbeck, “Reconstructing folios from text editions: Lévi (1900) + T-S NS 98.18 and Bodl. MS Heb d.74.27,” in Fragment of the Month, November 2018:


Eshbal Ratzon, “4Q208: A New Reconstruction and Its Implications on the Evolution of the Astronomical Book,” RevQ 31.1 (2019): 51–110, especially 110.


To be clear, even if the scroll contains dry rulings, the space between lines within one column is not fixed. However, the spaces between corresponding lines in adjacent columns contained in one sheet will remain constant. In other words, the distances between lines 1–2, 2–3, 3–4, etc. will be fixed in all columns of the same sheet.


For example, 4Q208 is written in an archaic non-formal script, which is more prone to error than a formal script. In that case, the use of the font was mainly for the purpose of visualization, and the reconstruction itself was based mainly on textual considerations. See Ratzon, “4Q208,” 51–110.


Herbert, Reconstructing Biblical Dead Sea Scrolls, 7–11, 62–63.


The font program allows for binarizing the color of the glyphs and filling their surface in case there are gaps or holes in it. It is important, however, to verify the integrity of the letter’s surface after the automatic processing by the program, and manually enhance it where necessary.


See chapter 16. For a similar case see Ratzon, “4Q208,” 51–110. Ratzon used some letters from 4Q28 for the reconstruction of 4Q208.


For inconsistency of the spaces between words, see Tov, Scribal Practices, 106–7, 133–35.


For example, in the software Adobe InDesign it is important to choose V/A “metrics” instead of “optical.”


We would like to stress here the importance of making the fonts available to the wide scholarly public. Open access is the spirit of our age, which has significantly enhanced the advance of many disciplines. The fonts prepared in the SQE project are available for download in


There are several font-design programs on the market. It is important to make sure that the chosen program supports Right-To-Left languages. At an early stage we worked with the free software Bird Font, with the kerning done with Microsoft Volt. However, when the workflow did not go smoothly, we switched to High Logic Font Creator (, a one-stop program with improved user experience. This software works on PC, while the standard font software for Mac users is Glyphs.


We were not able to perform this stage on 11Q5, because its bottom part was not preserved.

  • Collapse
  • Expand