Appendix 1 Experiment for Validating the Use of a Custom-Made Font

In: Material and Digital Reconstruction of Fragmentary Dead Sea Scrolls
Eshbal Ratzon
Search for other papers by Eshbal Ratzon in
Current site
Google Scholar
Einat Tamir
Search for other papers by Einat Tamir in
Current site
Google Scholar
, and
Rivkah Madmoni
Search for other papers by Rivkah Madmoni in
Current site
Google Scholar
Open Access

1 Introduction

Chapter 10 conveys a method for creating fonts by imitating the handwriting of a scribe in order to reconstruct lacunae in a scroll. We claimed that using a custom-made font, while potentially not as accurate as letter cloning, is accurate enough for reconstruction and is less time consuming than cloning. This appendix presents an empirical validation for this claim and provides the margin of error to be expected for it.

The font can be used to reconstruct a missing text in three different stages:

  1. the length of a line;

  2. the height of a column;

  3. long textual units spanning over several columns.

We examined the precision of the font in an experiment conducted on three of the comparatively well-preserved scrolls. The scrolls are 1QIsaa, 1QS, and 11QPsa. We prepared the font for these scrolls especially for this experiment. It is available for download at the SQE website.1 All three scrolls contain several consecutive columns with at least 17 complete lines each. Before we began the work, we examined scribal practices, such as the frequency of using vacats and the existence of dry rulings in each of the scrolls. While performing the experiment some additional nuanced practices were noticed. These will be summarized for each scroll at the end of the experiment report.

In order to achieve a margin of error for each of the experiment stages, we assumed that only part of the text was preserved, based on which we established the font size. We then typed the rest of the respective unit (lines or columns) using the font to simulate the original textual unit, and compared the size of the simulated text to the size of the original unit in the handwriting of the scribe. The results support the reliability of using a font as a method for reconstruction. All the raw materials for this experiment, including InDesign files on which the simulations were performed and excel files with measurements, are openly available for view and download.2

2 Materials: The Examined Scrolls

2.1 The Great Isaiah Scroll (1QIsaa)

The Great Isaiah Scroll is one of the first seven scrolls found in 1947 in Cave 1 and is the best-preserved scroll found in Qumran. It consists of 54 columns preserving the entire biblical book of Isaiah. Most of these columns are undamaged. It remains debated whether the entire scroll was written by the same scribe or whether columns XXVIIILIV were written by a second hand.3 Avoiding the need to decide on this matter, we chose the columns for this experiment from the second half of the scroll, which is better preserved than the first half and which contains fewer supralinear additions. While vertical ruling exists throughout the scroll,4 the scribe deviates from the borders of the column quite freely.

Choosing the right images for performing the experiment on this scroll proved to be problematic. The optimal images would indicate the accurate scale by means of a ruler next to the scroll. While 1QIsaa was photographed several times over the years, fully-scaled images are not easily available. The full set of images from the Shrine of the Book was not available to us, due also to the Covid-19 crisis (2020). Of the earlier sets of images, the set taken by John Trever in 1948 does include a scale but we encountered difficulties when digitally stitching the distinct images together.5 Therefore, we had to settle for an image from Wikimedia Commons.6 This stitched image of the entire length of the scroll has superb quality but it does not contain a scale, and provides no information about the way the stitching was made. Since the most important result for this experiment is the error in percentage, the effect of the scaling is less relevant.

The typed text of this scroll is taken from the Accordance program (Oaktree Software).

2.2 Community Rule (1QS)

This scroll is written over 11 nearly fully-preserved columns including top and bottom margins.7 The first and final two lines of each column are sometimes broken. The scribe was not very careful, and the scroll has many supralinear corrections and additions. In addition, the scribe’s tendency towards adding a comparatively large number of vacats may result in a significant error for the reconstruction. The scroll contains dry rulings; this is helpful for the reconstruction of the uneven column widths. The first column is significantly narrower than the rest of the columns.

Photographer Ardon Bar-Hama imaged some of the large scrolls for the Shrine of the Book at the Israel Museum in Jerusalem in 2010–2011, and the Shrine kindly supplied them to us.8 These high-resolution images include an industrial standard ruler. However, for some reason the files are not uniformly scaled. We therefore had to rescale each of them individually in order to stitch them together.

The text of the Community Rule used in this experiment is based on Abegg’s reading in the DSSEL.9

2.3 The Great Psalms Scroll (11QPsa)

The Great Psalms Scroll is designated 11Q5 or 11QPsa. The continuous part of the scroll comprises three whole sheets, plus an incomplete fourth at the end, which add up to nearly 24 columns. The handwriting is very neat and organized. Column and lines are marked by dry rulings. Between each psalm there is a vacat with a varied length between half a line and two lines. While most of the script is quite uniform, the scribe wrote the Tetragramaton with a Paleo-Hebrew script. We prepared a glyph in the font for the Tetragramaton, but it appears that its overall size varied throughout the scroll.

This scroll was imaged by the LLDSSDL in 2015 next to a standard ruler, with the camera set at a permanent position, planned in advance to create 1:1 scaling. However, apparently, over the years the scroll underwent shrinkage. Sanders measured the length of the sheets (with the last one first) to be 77, 72, 87, and 81 cm,10 while according to our measurements based on the 2015 images, they are now 75.5, 69.3, 83.0, and 72.4 cm, respectively. Since the shrinkage is not constant over the length of the scroll, it seems that reconstructing the length of the original scroll based on the images of its current state is problematic. Obviously, shrinkage and other damaging processes also occurred during the two millennia prior to the unrolling of the scroll, but these cannot be measured today. Therefore, it is important to keep in mind that the error measured in the following experiment reflects the error compared to the length of the scroll in its current state of preservation. An additional error exists in relation to the original length of the scroll because of shrinkage, but this error cannot be known.

We chose the final columns of 11Q5 for the experiment, as they were best preserved. The text of 11Q5 used in this experiment is based on Sanders’s readings taken from the DSSEL, with our own transcription of text when not represented in that electronic resource.11

3 Method

3.1 Stage 1: Length of a Line

This stage of the experiment simulates the quite common scenario by which a small fragment containing only a few characters is preserved, but the text of the rest of the line is known from elsewhere. We want to verify that the length of the reconstructed text using a custom font corresponds to the real length of the line.

For a given column, we simulated the situation in which a fragment of only eight characters (including spaces) from the beginnings of all lines of the column was preserved. The font size for the rest of the line was set according to these eight “extant” letters. If letter size was not uniform throughout the eight characters, we set it based on frequency, average, or the last letters in the “fragment,” depending on our evaluation of the most fitting size. The reason we do not always choose the average is because we wanted to ignore exceptional letters. This required some personal judgement.

After typing the text of the rest of the line, we measured and compared the length of the typed text to the length of the handwritten line of the original scribe. We wrote a continuous text within the line even in cases where the real line contained vacats or interlinear script, to simulate the situation of a real fragment, when one has no way of knowing in advance when to expect vacats or scribal mistakes. Having discovered that a comparatively large contribution to the error is our ignorance regarding the existence and size of vacats, we display the results for lines with no vacats separately.

3.2 Stage 2: Height of a Column

This stage of the experiment simulates the scenario by which only part of a column is preserved, but the text of the rest of the column is known from elsewhere.

We produced a virtual fragment with only the top three lines preserved and filled in the rest of the text in this column using the custom-made font. The three “extant” lines indicated the font size, the space between lines, and the width of the column.

Under these conditions, the text of the entire column is known but not the exact length and layout of individual lines. We typed the text according to the column dimensions known to us. As the space between lines may vary, we then compared both the number of lines and the height in centimeters of the simulated text to those of the original column.

3.3 Stage 3: Larger Textual Units

In some cases, a long continuous text is known from parallels. In some overlapping cases, a scholar will need to establish the amount of space this text would occupy in a given scroll where it is not extant. We would like to test the validity of a reconstruction of this long stretch of text using a custom font; the layout of the simulated columns is established based on the extant ones or on material reconstruction of the scroll.

In stage 3 of the experiment, we again created a virtual scroll, in which only the first three lines of the first column are extant but the rest of the text is known. We set the font size and interlinear space based on the first three lines of the first column of each scroll.

Stage 3 was performed in two versions. In version 3.A we assumed that none of the measurements of the following columns is known, and used the height and width of the first column for all of them. In this version the error is expected to be larger since the unknown measurements of the columns contribute to it. In reality, however, for a real scroll some information is available about the size of at least some of the columns. In order to isolate the contribution of the font to the error, we created another version (3.B). In version 3.B, the text of the next columns, the number of lines and column width for each column is known, but not the exact points in the text where each line and column end.

In both versions, we copied the entire text of the next columns continuously from the end of the “extant” lines to the end of the simulated columns. In version 3.A, we made a separate record for each simulated column, recording how many lines in it are in lack or excess vis-à-vis the last word of the original column. Lines that exceed the column are marked in positive numbers; the opposite case being marked by negative numbers. In version 3.B we were unable to use the lines as a measurement unit for the error, because the width of the lines was uneven (matching the actual width of the column, rather than the first column throughout). In this version, the unit for measuring the lack or excess of text was the number of characters.

This stage of the experiment was carried out only on 1QS and 1QIsaa. It could not be performed on 11Q5, as all of the columns of this scroll are missing a few lines at the bottom. The number of columns to be examined was dictated by the eleven extant columns of 1QS.

4 Results

4.1 1QIsaa

4.1.1 Stage 1: Length of a Line

Stage 1 was carried out on columns 41–44 of 1QIsaa. It is reported in Tables 2 and 3 below. As is evident from the table, the average error for each one of the columns ranges between 2.9–5.3%. The error is rarely more than five letters, and usually even less.

Figure 24
Figure 24
Stage 1 performed on 1QIsaa columns 41 and 42. The first eight characters of every lines (yellow) are treated as extant, whereas the rest of the line is simulated.
© Shrine of the Book, Israel Museum. Ardon Bar Hama
Table 2
Table 2
Margin of error for the reconstruction of all lines in 1QIsaa
Table 3
Table 3
Margin of error for the reconstruction of lines with no vacats in 1QIsaa

Exceptions, however, do exist. The two largest errors in col. 41 are caused by vacats (lines 1 and 12 with 14.6% and 13.2% respectively). While the first vacat occurs also in MT (49:4), the second one does not (49:12).12 Vacats cannot always be anticipated, and the error caused by them cannot be reduced by different means of reconstruction such as letter cloning. Even without considering vacats, the handwriting in col. 41 is less regular than the handwriting in other examined columns. Other large errors in this column are 10.4%, 10.1%, 9.9%, 9.4% (lines 15, 9, 5, 17 respectively). All of these errors are caused by the inconsistent handwriting of the scribe.

In column 41, in approximately 20% of the lines the error was below 3%, while in col. 42 it was true for nearly 50% of the lines, over 65% of the lines of col. 43, and approximately 60% of the lines in col. 44. Column 41 thus produced a large number of lines in which our reconstruction did not match the true length of the lines. We may speculate that this incongruity is due to the scribe’s fatigue in this particular column. The scribe would have restored his usual habits in the next column after taking some rest.

Exceptionally long or short lines may affect the error. The largest error is in Col. 42, line 22 (12.4%). This is an exceptionally long line, where the scribe made an effort to fit a large amount of text. In column 41, the largest error occurs in line 17, which is a comparatively short line, and the scribe was generous with spaces between words. The same is true for cols. 43 and 44. In col. 43, only one comparatively short line (1) has an error of approximately 10%. All the rest of the errors are significantly lower. In col. 44 four lines show an error of 8–9% (6, 15, 19, 22). Three of these lines are comparatively short, while all the rest show a lower rate of error. Apparently, when scribes leave a long vacat at the end of a line, they tend to write the text more spaciously than usual.13

The columns examined in this scroll, and the scroll in general, does not contain many vacats, and therefore they do not contribute much to the average error.

4.1.2 Stage 2: Height of a Column

This stage was examined on cols. 41–51. We chose exactly eleven columns in order to match the number of columns in 1QS. These particular columns were chosen because they are not damaged, contain a minimum of second-hand additions, and were written by the same scribe.

We compared each simulated column to the original one with regard both to their height in centimeters and the number of lines contained in them. Differences in both cases are reported in percentages, in Table 4 below. The average error in both factors is approximately 6–7%, but the standard deviation of the difference in the number of lines is higher.

Table 4
Table 4
Margin of error for the reconstruction of column height in 1QIsaa

The largest error occurs in column 51. It seems that the first three lines of this column, according to which the width of the column was set, were narrower than the average line in the column. A similar situation occurs in column 46, the final column of its sheet. The lines grow wider from the upper part of the column to its bottom together with the physical shape of the sheet, with the seam going down in a diagonal rather than straight line. In this particular column, the space between the first three lines was larger than average, which caused the error in centimeters to be lower than the error in the number of lines. The large error in column 41 was caused by significant vacats. In column 44 the number of simulated lines was very close to the number of original lines, but the upper lines were denser than the following lines. This anomaly created a large error with regard to the column’s height. In terms of the absolute number of lines, columns 46 and 51 were exceptional with an additional 4.3 and 6.5 lines respectively. Columns 41 and 43 were missing 3 lines. Column 45 had an additional 2 lines. All the rest of the simulated columns show a difference of one line or less.

While the average error in stage 2 of the experiment is higher than that of stage 1, it remains within a reasonable level of error, error which may be unavoidable.

4.1.3 Stage 3: Larger Textual Units

In version 3.A, all columns were assumed to show the same measurements (height and width) of the first column. We then examined how much the final words of the simulated columns deviate from their original position. The error is expressed in the number of lines (Table 5).

Table 5
Table 5
Margin of error for the reconstruction of several consecutive columns in 1QIsaa – version 3.A

In column 41, the largest contribution to the error is caused by large vacats separating the paragraphs. The second column, 42, is wider than the first one. The large difference between the two created an error in the opposite direction, thus mostly annulling the error from column 41. Columns 43–47 and 50 are much narrower than column 41, a fact that drastically enlarged the error. While the error in the absolute number of lines grows larger with the progression of the reconstruction, the error in percentage stabilizes after a few columns around 10% with a standard deviation of approximately 3%. The specific error for each column mostly depends on the width of that column compared to the width of the first one.

Table 6
Table 6
Margin of error for reconstruction of several consecutive columns in 1QIsaa – version 3.B

In version 3.B we assumed that the size of each column is known. In this version of stage 3, the average error is significantly reduced to a little more than 4%. The first columns show a negative error resulting from vacats. However, the size of the script in the original diminishes as the columns progress, which creates the opposite kind of error. These two phenomena nullify each other, creating a very good reconstruction around column 45. The very low error of this column does not attest to outstanding consistency but is rather the outcome of the nullification of the errors by the contradicting causes, as can be seen in stage 2 above.

As expected from the results of stage 2, a comparatively large error occurs in column 41. While a significant error existed for column 46 when examined by itself in stage 2, in this stage it is lower thanks to the lower error of the previous columns. In stage 2, column 51 produced the largest error, but it was then dependent on the exceptionally narrow first lines of this column. These did not affect the simulation in stage 3, thus keeping the error for this column in stage 3 close to the average.

4.1.4 Unique Scribal Practices

While working on the simulation of 1QIsaa, a few scribal practices of this scroll became evident. We summarize them here. These are small-scale practices not evident to the naked eye, which have been revealed by means of the minute numerical attention in this experiment. They are thus additional to the practices enumerated by the editors of DJD XXXII.14

Generally speaking, in columns 41–51 the size of the script is reduced as the columns progress. The handwriting of some columns is less consistent than others. This problem was especially evident in columns 41, 46, and 51. The consistent gap of five columns may indicate that this phenomenon is caused by the scribe’s fatigue. Perhaps after paying attention to the reduced quality of the work, the scribe took a break, and was able to retain the earlier standard for the next few columns.

Some of the inconsistency of the handwriting is related to exceptionally long or short lines. In long lines, the scribe made an effort to fit a large amount of text, thus making the script denser. On the other hand, when the scribe left a long vacat at the end of a line, the text was more liberally spaced than usual.

4.2 The Community Rule (1QS)

4.2.1 Stage 1: Length of a Line

This stage of the experiment was carried out on columns 8–11 of 1QS.

The average value of the errors in the reconstruction of 1QS is similar to that of 1QIsaa. Here too some of the largest errors are caused by unpredictable vacats (11.4% in col. 8 line 5; 13% in col. 8 line 7; 8.8% in col. 11 line 2). The phenomenon of spacing the letters or crowding them in shorter and longer lines respectively, encountered in 1QIsaa, was not noticed in 1QS. Some further errors are caused by defects in the skin that forced the scribe to deviate from the usual handwriting (e.g. col. 8 line 19), or from omissions that are completed above the line (e.g. col. 8 line 8; col. 10 line 19).

Table 7
Table 7
Margin of error for the reconstruction of all lines in 1QS columns 8–11
Table 8
Table 8
Margin of error for the reconstruction of lines with no vacats in 1QS columns 8–11

As in 1QIsaa, here too an exceptionally irregular column was noted. In col. 10, two of the errors are above 10% (lines 13 and 23) even without vacats, and only 25% of the errors are less than 3%. But in col. 11 the scribe returns to the consistent writing mode, with approximately 68% of the lines showing less than 3% errors. Only two of col. 8’s lines that do not contain vacats (19 and 22) show errors larger than 3%, and one of them (line 19) includes a supralinear correction. Col. 9’s maximal error is only 6.3%, and nearly 60% of the error are less than 3%.

4.2.3 Stage 2: Height of a Column

This stage was examined on all eleven columns of 1QS. Since the first three lines of column 1 are damaged, we established the font’s parameters based on lines 4–6 of this column. In addition, we ignored the line 7:27, which contains only one word. This scroll contains vertical dry rulings for every column, thus does not raise the problem of the unknown measurements of column width.

For the most part, the use of a font gives very good results for the reconstructed number of lines in this scroll. The error amounts to only 1–2 lines (and an average for all lines of 1.81), which is a common variation even between columns of the same scroll. This is the highest precision a scholar can expect from a reconstruction of a column.

Column 7 is exceptional with an error of 4.5 lines. This column includes many scribal mistakes, erasures, and corrections in addition to large vacats, the largest of which is approximately 2–3 lines long. The error incurred in this line is thus not caused by the use of a font but rather by unpredictable variables. Similar error would have occurred had we used other means too.15 Other smaller vacats also have a significant effect on the error. All the simulated columns are shorter than the original ones. This should be explained by the existence of frequent vacats in this scroll.

In this scroll the vacats of every column amount to an average of one blank line per column. But the average result at this stage of the experiment is larger (1.81 lines), and thus cannot be explained by the presence of vacats only. The rest of the error should therefore be explained by the fact that the scribe of 1QS had a tendency to use smaller script in the first few lines of the column, enlarging the letters down the column.16 The source for this error is thus definitely related to the use of font. It is difficult to avoid the problem, since in fragmentary scrolls scholars depend on the material that survived, having no way to determine the size of the letters that was used in the non-extant parts. Thus, even if the problem caused by vacats and blank lines is accounted for, the reconstruction must acknowledge an extra error of ca. 5% for each column (in this case only one line per column).

Table 9
Table 9
Margin of error for the reconstruction of column height in 1QS

Unlike the reconstructed number of lines, the error of the reconstructed height in cm is lower. This is due to the above-mentioned phenomenon: the scribe’s tendency to enlarge the script towards the end of the column and the many vacats. In addition, the space between lines in 1QS usually increases towards the end of the column. Since we established the simulated space based on the upper lines, the simulated lines turned out to be denser than the original ones.

The average error for both the height and number of lines is approximately 7–8%, but the standard deviation in this scroll is larger for the height error than for the number of lines. The error in column 7 is exceptional not only with regard to the number of simulated lines, but also with regard to the simulated height, with the error reaching over 30%. Large errors occur also in columns 3 and 11.

Figure 25
Figure 25
1QS Column 3 with the top three lines extant and the rest of the column simulated. In this visual representation, the “red” column ends ca. 4.5 lines before the real one, but the line count shows only 2.5 lines difference.
© Shrine of the Book, Israel Museum. Ardon Bar Hama

4.2.4 Stage 3: Larger Textual Units

In 1QS the first column is significantly narrower than the rest of the columns. This oddity created huge mistakes in stage 3.A of the experiment, in which the size of the columns is based solely on the measurements of the first column. Had we encountered a scroll in which only this exceptional column is preserved, the results would indeed be substantially wrong. Not only the absolute number of lines is in error, but also the relative error (expressed in percentage of the overall text) increases as the simulated text is longer.

Table 10
Table 10
Margin of error for the reconstruction of several consecutive columns in 1QS – version 3.A
Table 11
Table 11
Margin of error for the reconstruction of several consecutive columns in 1QS – version 3.B

The situation is significantly improved in version B, under the assumption that the height and width of each column is known. The average error is around 6%, slightly higher than for 1QIsaa.

Since in Stage 2 of the experiment all simulated columns of 1QS turned out to be shorter than the original ones, it is expected that over a longer text comprised of several columns the error will sum up and increase as the amount of simulated text grows. However, since the script was smaller in the first columns and grew larger as the scroll progressed, the simulated text held more characters than the original one in the first three columns, but that changed in the following columns. After the accumulation of enough columns the simulated text became shorter than the original one. The simulated text in fact ended after ten columns, leaving no column corresponding to the last, 11th column of 1QS. Note that this column is shorter than the rest of the columns. While 1QS is anomalous in the amount of uninscribed space, this is still something to consider when reconstructing long sections. Interestingly, the relative error decreases in the first few columns, and then gradually increases. with the numbers behave similarly to 1QIsaa, but the positive and negative numbers are opposite. In both scrolls, the larger the simulated text is, the more prone it is to errors in absolute numbers; however, while in 1QIsaa the relative error becomes stable after a few columns, in 1QS it keeps growing at a small rate.

4.2.5 Unique Scribal Practices

Some unique scribal practices were noticed during the simulation. As in 1QIsaa, here too some columns were less regular than others. Thus, columns 3, 7, and 10 were inconsistent. This supports the suggestion that the inconsistency is affected by the scribe’s increased fatigue every three or four columns.

In addition, as Herbert has already noted, the scribe of 1QS has a tendency to use smaller script in the first few lines of the column, enlarging the letters down the column.17 Together with the script, the space between lines in 1QS usually increases towards the end of the column.

The phenomenon of spacing the letters or crowding them in shorter and longer lines respectively, encountered in 1QIsaa, was not noticed in 1QS.

4.3 11Q5

4.3.1 Stage 1: Length of a Line

This stage of the experiment was carried out on columns 20–24 of 11QPsa. Since fewer lines of this scroll’s columns were preserved, we decided to include five columns rather than four in this stage.

Since the columns in this scroll are comparatively narrow, the contribution of the vacats to the relative error is larger than usual: 20.7% in col. 20 line 8; 25.2% in col. 21 line 1; 28.2% in col. 22 line 1; more than 8% in col. 22 lines 4, 7, and 16. Removing these lines from the statistics gives similar results to those of the first two scrolls. The phenomenon of higher errors in shorter or longer lines (observed in 1QIsaa) was not noticed. In addition, there were no conspicuous second-hand corrections.

The least regular column of this scroll is col. 22, with less than a third of its lines containing errors lower than 3%. After removing the lines with the vacats, the more accurate lines increase to 50%, similar to the rest of the columns. The irregularity is thus explained by the large number of vacats rather than by the scribe’s fatigue.

Table 12
Table 12
Margin of error for the reconstruction of all lines in 11Q5 columns 20–24
Table 13
Table 13
Margin of error for the reconstruction of lines with no vacats in 11Q5 columns 20–24

4.3.2 Stage 2: Height of a Column

Since all of the columns of 11Q5 are damaged at their end, we worked with only 11–17 lines in a column, much less than the previously examined scrolls. Vertical dry rulings are preserved, facilitating the reconstruction of the column’s width. The script is highly uniform. Vacats are regularly added between psalms. In most columns, despite the vacats, no error was found in the simulated number of lines. Only in two cases is the error larger than a few words. The error of the simulated height in cm is larger, due to irregularity in the spaces between lines. In cases of an error of a few words, the effect on the height of the column is larger, because the addition of the few words adds another line to the column. A space of two lines was left by the scribe between psalms in column 18, which created a larger error in the reconstruction of this column. It is difficult to explain the error in column 23. A vacat of half a line is not enough as an explanation, because similar vacats appear in other columns as well. An additional explanation may result from the many occurrences of the Tetragrammaton in this column, which in this scroll is written in Paleo-Hebrew script, whose size is difficult to predict.

Table 14
Table 14
Margin of error for the reconstruction of column height – 11Q5

4.3.3 Unique Scribal Practices

This scroll is highly formal and regular, and no previously unknown scribal practices were noticed during the experiment.

5 Discussion

Stage 1 of the experiment was tested on all three scrolls (1QIsaa, 1QS, and 11Q5). The results exhibit average errors of 3.7–4.2% for the reconstructions of a length of a line. Errors are sometimes caused by vacats, which are not always predictable, but the effect of this factor is not high: computing the average error without those lines that contain vacats only reduced the error to 3.1–3.45%. Other causes for the error include the scribes’ inconsistency likely due to fatigue, some tendency to crowd or space the writing at the end of lines, damages to the skin, and scribal corrections. Whatever the reasons, usually the errors are not large, and rarely reach or exceed 10%. In order to improve the results, a scholar must pay close attention to the scribal practices in the extant fragments before the beginning of the reconstruction. We anticipate that additional practices may surface during the reconstruction work, and these should be taken into account as well. When reconstructing a column width using a font, we recommend to give a range of ±4% for each column, and to draw further conclusions accordingly.

Stage 2 examined the precision of the reconstruction of column height. While in 1QIsaa and 1QS full columns were preserved, the bottom lines of 11Q5 were destroyed, and we had to reconstruct only the preserved lines. In this stage we measured the error for the column height both in centimeters and in the number of lines. The average error in cm was 6–8% for all three scrolls, while the average error for the number of lines was approximately 7% for the first two scrolls, and less than 3% for 11Q5. In all three scrolls the error for most columns was less than 2 lines. Larger mistakes were caused by large or numerous vacats, inconsistency of column width and inconsistency of handwriting. As for the height of the column in centimeters, inconsistency in the spacing of lines has a significant contribution to the error. Large errors for the number of lines of over 10% were rare, occurring in only 1–3 columns in each scroll. When reconstructing the height of a column whose text is known with a font, a scholar should be aware of an error of approximately two lines in most scrolls,18 but perhaps only one line if the scroll is exceptionally formal and consistent.

Due to the damages to the bottom part of 11Q5, we were unable to include it in stage 3 of the experiment, which involves the reconstruction of text along a large number of columns. We performed this stage in two versions. In version 3.A, we reconstructed the space required for copying the text of the scroll based on the size of the first column without assuming the height and width of the rest of the columns. In this version, the results were not good. The average error in 1QIsaa was approximately 10%, and in 1QS it was over 30%, and it seemed to increase the longer the reconstruction goes. However, if the measurements of the following columns are indeed known (version 3.B), the average error is approximately 4% and 6% in in 1QIsaa and 1QS respectively. Reality is somewhere in the middle between these two versions, as scholars usually have clues for the dimensions of at least some of the columns.

In both scrolls, establishing the space required for a small number of columns is different than for a larger number of columns. While in the first columns, vacats and inconsistencies of the handwriting are more significant, sometimes annulling each other, later on the relative error stabilizes around a certain number.

To summarize this stage, the reconstruction of a large number of columns with unknown dimensions may not be considered reliable, unless some information may be culled for the dimensions of the missing columns. At the same time, it is important to set this caveat in perspective. Even without any prior knowledge of the dimensions of the columns, the error incurred by the procedure described here is smaller than the error obtained by other methods commonly accepted in our field. For example, the error incurred when using the Stegemann Method for the reconstruction of a length of a scroll.19

Overall, the outcomes for the reconstruction of all three sizes: width of a line, height of a column, and space required for longer textual units, yield comparatively small errors. When considering the precision of the “font method” compared to other reconstruction methods, it has been demonstrated here that most of the errors are not caused by the use of font, and are in fact unavoidable even with other means of reconstruction, such as letter cloning. Such errors are caused by unpredictable scribal practices such as vacats, defects in the skin, inconsistencies in column width, etc.

One cannot avoid errors, but these errors are usually small enough to keep the reconstruction reliable. Nevertheless, errors should be kept in mind, especially when using the reconstruction for other purposes. Scholars should remember that a reconstruction does not give an absolute result, but a range of possibilities, and they should therefore conduct their reconstructions accordingly. The reconstructions provided in this book (chapter 15) are aware of this factor and will demonstrate its correct usage.




For a summary of both opinions see Tov, Scribal Practices, 27 and Eugene Ulrich and Peter Flint, Qumran Cave 1.II: The Isaiah Scrolls, DJD XXXII (Oxford: Clarendon Press, 2010), 61–64. Ulrich and Flint claim that a single scribe wrote the entire scroll. See also recently Mladen Popović, Maruf A. Dhali MA, Lambert Schomaker, “Artificial Intelligence Based Writer Identification Generates New Evidence for the Unknown Scribes of the Dead Sea Scrolls Exemplified by the Great Isaiah Scroll (1QIsaa),” PLoS One 16(4) (2021): e0249769.


Ulrich and Flint, DJD XXXII, 2.59.


When matching overlapping portions of the scroll from various images one against the other, they were never quite the same. We can only assume that something was not uniform in the photographing process. For a survey of early images see Ulrich and Flint, DJD XXXII, 2.15–21 and 59–61.



The experiment did not include 1QSa and 1QSb, regardless of the question whether they constitute part of the same scroll as 1QS. For this question see recently Michael B. Johnson, “One Work or Three? A Proposal for Reading 1QS-1QSa-1QSb as a Composite Work,” DSD 25.2 (2018): 141–77.


We would like to thank Hasia Rimon for her help in facilitating the images.


Martin G. Abegg, “1QS,” in Brigham Young University: The Dead Sea Scrolls Electronic Library, ed. Emanuel Tov, rev. ed. (Leiden: Brill, 2006).


James A. Sanders, The Psalms Scrolls of Qumran Cave 11 (11QPsa), DJD IV (Oxford: Clarendon Press, 1965), 3–4.


James A. Sanders, “‎11Q‎5 (11QPsa),” in Tov, Brigham Young University: The Dead Sea Scrolls Electronic Library.


For the correlation of vacats in this scroll vs. the medieval Masoretic codices, see Yeshayahu Maori, “The Tradition of Pisqā’ôt in Ancient Hebrew MSS. The Isaiah Texts and Commentaries from Qumran,” [Hebrew] Textus 10 (1982): אנ; but see the opposite conclusion in Ulrich and Flint, DJD XXXII, 2.82–86.


For other scribal practices in this scroll, see Herbert, Reconstructing Biblical Dead Sea Scrolls, 64–68.


Ulrich and Flint, DJD XXXII, 63–64.


See Herbert, Reconstructing Biblical Dead Sea Scrolls, 19–20.


Herbert, Reconstructing Biblical Dead Sea Scrolls, 72–74, already noticed this tendency.


Herbert, Reconstructing Biblical Dead Sea Scrolls, 72–74.


In exceptionally high or short columns the number may change accordingly.


For criticism on this aspect of the Stegemann method see Eshbal Ratzon and Nachum Dershowitz, “The Length of a Scroll: Quantitative Evaluation of Material Reconstructions,” PloS One, 2020. doi: 10.1371/journal.pone.0239831. The criticism pertains only to the assessment of the length of scrolls, while other parts of the Stegemann method are used in this book with a smaller level of potential error.

  • Collapse
  • Expand