Digital humanities have promoted the digitization, sharing, and visualization of humanities resources. However, as Flanders, Piez and Terras stated in the first issue of Digital Humanities Quarterly, digital humanities “is by its nature a hybrid domain, crossing disciplinary boundaries and also traditional barriers between theory and practice, technological implementation and scholarly reflection” [Flanders, Piez and Terras 2007, par. 3].

Digital humanities is an ongoing effort to facilitate interaction between traditional humanities and digital/computational methods. However, continuing this effort is somewhat challenging. The main obstacles are difficulties in understanding each other's languages and methodologies. To promote interaction with humanities scholars, the digital humanities side should demonstrate how their methods can contribute to specific humanities research topics. Similarly, humanities scholars should expand their knowledge of digital/computational approaches. What is needed is not sophisticated tools or presentations but concrete results that address unsolved questions.

Flanders argues that humanities computing does not primarily concern itself with either the discovery of new knowledge or the validation of existing information. Instead, she suggests that its core function lies in the modeling of that knowledge, and in certain instances, the modeling of the modeling process itself. According to her, humanities computing is a method of exploring our modes of knowledge acquisition and the presentation of that knowledge for personal study [Flanders 2009, par. 10]. Going beyond Flanders's argument, this paper aims to model how we comprehend a type of artwork using a mixed quantitative and qualitative methodology, and furthermore discover new knowledge in the field of Japanese art history from a digital humanities perspective.

New discoveries through digital humanities approaches are still forthcoming. Regarding the knowledge discovery in art history by computer vision technology, Lang and Ommer discussed effective use cases of digital technology, which include a classification and visualization of age and social status of models in painted portraits. They conveyed the idea that it is not enough to just convert analog methods into digital ones, as this approach does not capture the unique attributes and potential of digital data. They emphasized that the generation of new knowledge only becomes possible when we devise new computational methods for handling digital data. This level of knowledge production, they argue, could not be achieved with analog data and its corresponding methods. The impact of this shift, they stressed, extends beyond art historians to also touch earlier generations of academics, particularly students [Lang and Ommer 2021, par. 26].

The use of digital methods, nonetheless, should be judiciously applied to facilitate seamless communication between scholars in humanities and digital humanities. An effective stragegy is to pursue a study on a topic that is not overly specialized, thereby eliciting broader interest and utilizing a language mutually comprehensible to each other. In order to meet these criteria, this paper undertakes an examination of well-known artworks in Japan, utilizing digital methods that are not excessively complex.

“Hyakki Yagyō (Yakō) Emaki”[1] (“Night Parade of One Hundred Demons” Picture Scroll; henceforth, Demons Scroll), which depicts yōkai (Japanese specters) marching at midnight, have been popular in Japan due to the enigmatic and humorous depictions of demons.[2] It is believed that numerous Japanese painters replicated this subject between the sixteenth and nineteenth centuries, resulting in the formation of several lineages of “Hyakki Yagyō” scrolls.[3] The creation of multiple copies of these scrolls is indicative of a significant “copy culture” in premodern Japanese art.[4]

Shinjuan, a sub-temple of Daitokuji temple in Kyoto, owns the most famous version of the Demons Scroll, attributed to Tosa Mitsunobu[5] (1434–1525) and designated as an important cultural property of Japan (Figure 1). The majority of research on Demons Scrolls has referenced this Shinjuan copy. Since the Shinjuan scroll is believed to be the oldest of the known Demons Scrolls, it is only natural that scholars have focused their attention on it. However, previous research has overlooked an important fact: although many copies depict the same demons as the Shinjuan painting, few scrolls have the same compositional arrangement.[6] If the Shinjuan scroll is the oldest Demons Scroll, then many more copies with the exact same arrangement should be preserved. The demon arrangement in the Shinjuan scroll is likely one of many variants of the Shinjuan lineage of Demons Scrolls. It is also possible that the configuration of demons in the Shinjuan scroll's prototype differed from the extant copy.

Researchers of the Demons Scrolls have noted the need to reevaluate the Shinjuan scroll's significance. Tanaka Takako, a scholar of Japanese medieval literature, stated, “The Shinjuan copy was likely not the prototype of the Demons Scroll, but rather a unique case. In other words, among the numerous Demons Scrolls that circulated between the sixteenth and nineteenth centuries, I believe the Shinjuan lineage was fortuitously propagated” [Tanaka 2002, 220]. Yumoto Kōichi, a collector and researcher of materials pertaining to Japanese specters, wrote, “If the Shinjuan lineage is the standard, other copies omitted certain demons and added new ones. To position the Shinjuan version, its relationship with other types of Demons Scrolls must be clarified” [Yumoto 2005, 53]. In response to these studies, the Japanese expert on specter culture Komatsu Kazuhiko stated, “When considering the lineage of Demons Scrolls, we must set the Shinjuan copy aside” [Komatsu 2008, 164]. Based on their perceptions, this paper will present a new theory regarding the Shinjuan scroll.

In addition to the Shinjuan lineage, there are several other variants with different demons and what appears to be a mixture of two lineages. Tanaka centered her attention on a copy known as the “Tōhaku mohon” (in this paper, the Tōhaku-A scroll). The Tōhaku-A scroll contains all of the demons depicted in the Shinjuan scroll, as well as a large number of additional demons. Tanaka presumed that the original Tōhaku-A lineage predates the Shinjuan lineage and that the latter is an idiosyncratic copy of the Tōhaku-A lineage created by extracting demons from it.

In 2007, the “Hyakki no zu” (“Picture of One Hundred Demons”) picture scroll acquired by the International Research Center for Japanese Studies (“Nichibunken”; in this paper, Nichibunken-A scroll) was discovered to be composed of demons from the Tōhaku-A lineage minus those from the Shinjuan scroll. Komatsu's investigation of the Nichibunken-A scroll led him to conclude that its prototype dates to the sixteenth century, the same era during which the Shinjuan version is believed to have been painted. Komatsu argued that, contrary to Tanaka's theory, the original Tōhaku-A was a combination of the Shinjuan and Nichibunken-A compositions. Moreover, Komatsu asserted that Demons Scrolls can be classified into four distinct lineages that do not overlap, and that these lineages were sometimes combined to produce a variety of copies. The four lineages are represented by the Shinjuan, Nichibunken-A, Kyoto City University of the Arts (KCUA), and Hyogo Prefecture Museum of History (Hyogo-A) scrolls [Komatsu 2008, 233].

The study by Komatsu was the first to investigate and classify sixty-four known Demons Scrolls, including those preserved in the United States and Europe. By examining a diversity of scrolls, Komatsu was able to clarify that there are combined copies of Shinjuan and Nichibunken-A; Shinjuan and KCUA; Shinjuan and Hyogo-A; and Nichibunken-A and KCUA. If we were to accept Tanaka's theory that the Shinjuan scroll is an idiosyncratic copy, we would have to recognize that demons in Tōhaku-A were neatly separated to form the Shinjuan and Nichibunken-A lineages, with no overlap. However, this is unlikely. In contrast to Tanaka's theory, Komatsu's “mixture theory” seems more plausible. The author agrees with the mixture theory positing that there are four distinct lineages of Demons Scrolls: Shinjuan, Nichibunken-A, KCUA, and Hyogo-A, and that these were intermixed to produce the Tōhaku-A scroll and other combined copies. However, previous studies, including those by Komatsu, have focused on the differences and interpretations of the images while ignoring the arrangement of the demons. Using the digitized photographs of scrolls that Komatsu collected and others that have recently been made available on the internet, it is possible to estimate the process by which Demons Scrolls were drawn and edited by successive generations of painters.

Utilizing a computational method, this paper will explore new approaches to this research topic. The author will not employ aesthetic criteria, such as line quality. Deliberately disregarding minor differences in depiction, the author will identify the demons based on the parade's sequential order. Similar to a DNA test, we will calculate the proximity between two scrolls and try to determine the genealogy of Demons Scrolls.

Several previous studies have examined the editorial process of Japanese cultural artifacts. Yano Tamaki has conducted research on the genealogy of esoteric texts on the tea ceremony (chadō) [Yano 1999] [Yano 2022], as well as the “Sanjūrokkasen Emaki” (“Picture Scroll of Thirty-Six Master Poets”) [Yano 2006, 36–42], using tools such as SpritsTree[7] and Hayashi's Quantification Theory Family III.[8] Nevertheless, these studies rely on textual arrangement, including those that focus on illustrated scrolls. There is a conspicuous lack of previous research that examines the genealogy of picture scrolls based on their visual representations.

To ascertain the picture scroll editing process, we propose the following estimation model: the fewer the differences between scrolls, the shorter the number of edits between them, and the closer they are to each other in the chain of copying. For instance, if one block differs between scrolls A and B, and between B and C, and two blocks differ between scrolls A and C, the most likely conclusion is that B was created by editing A, and C was created by editing B (or vice versa). Obviously, one cannot rule out the possibility that C was created by editing A without mediating B, but this is probably not the case.

This paper utilizes the concept of edit distance in order to quantify the differences between scrolls. Edit distance is a standard method for determining text similarity [Nerbonne 2005].[9] In recent years, the field of linguistic computing has developed a variety of applications for edit distance, such as historical name normalization [Hämäläinen, M. et al 2018] [Benito-Santos, Díaz and Sánchez 2019], text comparison [Bernholz and Zillig 2011] [Hyytiäinen 2022], and analysis of library catalog transmission [Baker, Salway and Roman 2022]. Its application to visual arts, however, is somewhat challenging because edit distance is rarely applied to non-textual data.

Typically, edit distance is calculated based on the number of edit operations per symbol, but to accommodate this application to Demons Scrolls, we will use symbol blocks (chunks of symbols) as the unit and count the numbers of swap, insertion, and deletion operations on the blocks. This estimation model differs from Levenshtein distance (including insertion, deletion, and substitution) and Damelau-Levenshtein distance (Levenshtein plus transposition of two adjacent characters). We will assume that the differences between the symbol arrays “123456789” (sequence-A), “145678239” (sequence-B), and “167845239” (sequence-C) are due to editing operations. The estimated editing procedure is described in Figure 2. The symbol block “23” in sequence-A first swaps with “45678”, then it turns into sequence-B. Next, the symbol block “45” in sequence-B swaps with “678”, then it becomes sequence-C. The edit distance between A and B (and also between B and C) is one, while the edit distance between A and C is two. In this example, it is plausible that the sequence-A is edited into sequence-B, which is then edited into sequence-C (or vice versa). This estimation model is incapable of determining a single conclusion; it can only indicate what are the plausible processes. Therefore, it is essential to confirm the estimation's accuracy by employing additional qualitative methods.

This estimation model posits that picture scrolls can be segmented and encoded. By identifying the individual demons depicted in a Demons Scroll, it is possible to create symbol blocks and segment the scroll. However, this estimation model is challenging to apply when parts of a scroll are missing or if the consecutively joined sheets of paper have become disordered.[10]

This paper will analyze seventy-one Demons Scrolls, which include the sixty-four works catalogued by Komatsu [Komatsu 2008, 249–254]. We excluded works that were evidently copied after the mid-nineteenth century, and limited our study to the previously mentioned four lineages, as well as works constituting mixed editions of them. Komatsu classified the lineages as follows: class-A (Shinjuan lineage), class-B (Nichibunken-A lineage), class-C (KCUA lineage), and class-D (Hyogo-A lineage).[11] In addition, he discovered four types of mixed copies, including class-AB (Shinjuan+Nichibunken-A), class-AC (Shinjuan+KCUA), class-AD (Shinjuan+Hyogo-A), and class-BC (Nichibunken-A+KCUA). However, a closer look at these scrolls reveals that Komatsu's class-AB can be divided further into two classes: class-AB (the Shinjuan scroll demons come first, followed by the Nichibunken-A parade of demons) and class-BA (the demons in the Nichibunken-A scroll come first, followed by those in the Shinjuan scroll). Class-AD can also be divided into AD and DA. Our revised list of works is presented in the Appendix in light of these findings.

As stated at the end of section 2, we identified the demons by focusing on their relative positions in the parade sequence as opposed to minute differences in their depictions. Similar demons appearing in a similar sequence in two scrolls are deemed identical, and the two arrangements are thought to be the same (Figure 3). This method is not applicable to picture scrolls with a strong suspicion of missing or disordered sheets of paper, or to folding screens and hanging scrolls. Therefore, such materials were excluded from the analysis. The target works are consequently limited, as shown in the Appendix. We should note that one of the target works (Kyōgaen in class-AC) is a woodblock-printed book and not a painted scroll. We decided to include it because the demons in the book are depicted in sequential order and it contains useful information such as the date of publication.

Based on the arrangement of demons, we will first attempt to estimate the editing process of the Shinjuan (class-A) lineage, the most popular lineage of Demons Scrolls. As depicted in Figure 4, we identified all the demons in the Shinjuan scroll and assigned numbers to them. Using these numbers, eight other class-A scrolls were converted into symbol sequences. The Shinjuan (No. 1 in Appendix), Itō (No. 7), Iwate (No. 9), NDL-A (No. 12), Nichibunken-B (No. 14), Nichibunken-C (No. 15), Rekihaku-A (No. 18), Rikkyo (No. 21), and Spencer-B (No. 23) are the nine scrolls examined here that show almost the same numbers of demons in comparison to the Shinjuan scroll (Figure 5). In the case of the Shinjuan and Rekihaku-A scrolls, for instance, if we swap two blocks of demons in Shinjuan, (No. 04−07 with 08−30, and No. 20−30 with 31−43 in Figure 4) the demons transform into the arrangement in Rekihaku-A. The edit distance between the two scrolls is therefore two (See Figure 6).

Of these nine scrolls, we omitted the Rikkyo scroll because of its evidently unique arrangement. It has an unusual layout compared to other scrolls and can be considered an atypical copy. The edit distances between eight of the scrolls are shown in Table 1. Among them, the arrangement of Nichibunken-B is particularly notable (Yumoto-A [No. 26 in the Appendix] has the same arrangement).[12] This copy is connected to five scrolls (Shinjuan, Itō, NDL-A, Nichibunken-C, and Rekihaku-A) via an edit distance of one.

Based on the matrix of edit distance, multiple possible lineage trees can be drawn. The simplest tree is a structure where the sum of the distances is minimized, or the “minimum spanning tree”, as depicted in Figure 7. Nichibunken-B (Yumoto-A), which is located at the “root” and has the shortest distance to other scrolls, is connected to the Itō, NDL-A, Nichibunken-C, Rekihaku-A, and Shinjuan scrolls at a distance of one, and to the Iwate scroll at a distance of two. Spencer-B is connected at two to the Shinjuan scroll, while Rikkyo is isolated. If the class-A lineage shares a common ancestor, this tree indicates that the Nichibunken-B scroll's demon arrangement is the closest to it. In other words, it is possible that the Shinjuan scroll is a rearranged version of Nichibunken-B's prototype.

This lineage tree is supported by a comparison of iconographic details. For example, let us investigate whether or not a handle is depicted on the pot lid of the “demon wearing a pot lid” that appears in the latter half of class-A lineage scrolls (Figure 8). Japanese art historian Komatsu Shigemi observed that the handle depicted in the Tōhaku-A (No. 49 in the Appendix) scroll was missing in the Shinjuan scroll. He stated that “such detailed copying strongly suggests that the Shinjuan scroll was painted based on a prototype” and that the Shinjuan scroll had another prototype [Komatsu 1979b, 132]. If we apply the presence or absence of the “pot lid handle” to the lineage tree depicted in Figure 7, we find that the “handle” is depicted in Nichibunken-B (Yumoto-A), which is at the “root”, and in the Iwate version, but not in other scrolls. This indicates that the presence of the “handle” was clearly distinguished on the lineage tree. While Nichibunken-B (Yumoto-A) retains the demon arrangement of the ancestorial copy relatively well, the Shinjuan and other six scrolls are copies of versions in which the “pot lid handle” was omitted.

In what ways do the demon configurations of the Nichibunken-B and Shinjuan scrolls differ? As illustrated in Figure 9, the two can be transformed by a single block swap. It is remarkable that some of the movable demon blocks in the Shinjuan scroll can be separated at the paper joins. Five joins in this scroll do not overlap the figures, and three of them are the boundaries where the demons are arranged differently than in Nichibunken-B. In contrast, only one join at the beginning of the Nichibunken-B scroll does not overlap the figure. In other words, the arrangement of the demons in the Shinjuan scroll can be transformed to that of Nichibunken-B by simply detaching the paper joins and rearranging the sheets. If the arrangement of Nichibunken-B is closer to the Shinjuan scroll's prototype, we can conclude that the order of paper sheets became disordered during a remounting and resulted in the current arrangement in the Shinjuan scroll. This new theory regarding Demons Scrolls emerged from the field of digital humanities. Close inspection reveals that the Shinjuan scroll's joined sheets of paper are stained along their edges. This indicates that the sheets of papers were separated and preserved for a certain period of time, and is evidence supporting our conclusion that the sheets were incorrectly arranged when the scroll was reassembled.

Because the majority of Demons Scrolls lack the names and descriptions of each demon, it is difficult to comprehend their iconographic meanings. Nonetheless, by numbering the demons, we are able to see how the individual demons collectively form meaningful units. If a block of demons appears in multiple copies, it indicates that it was recognized as a “unit” by multiple painters.

Researchers of Demons Scrolls have debated the iconography of No. 31 (running fox lady) in Figure 10: whether she is “running toward” the group of ugly women with black teeth[13] in front of her (No. 24−30) [Komatsu 1979a, 81] [Yumoto 2005, 40], or “escaping” from the cloud of demons that follows (No. 32−) [Yasumura 1987, 43] [Komatsu 2008, 39]. An examination of Figure 5 reveals that six scrolls (Itō, Iwate, NDL-A, Nichibunken-B, Nichibunken-C, and Rekihaku-A) contain the common separation of the block between No. 30 and 31. While it is impossible to determine the intentions of the original painter of the class-A lineage, this separation suggests that many painters who copied the scroll may have interpreted the “running fox lady” as “escaping” from the “cloud of demons”. This is another new theory developed through digital humanities. As previously discussed, misordering of the sheets of paper is suspected between section No. 30 and 31 in the Shinjuan scroll. In other words, the original compositional arrangement would also have intended the “fox lady” to be “escaping”.

Next, we will examine the ancestry of the class-AB/BA lineages. Class-AB comprises the combined scrolls that begin with Shinjuan demons followed by Nichibunken-A demons, and class-BA is the opposite. In terms of the arrangement of demons, class-AB/BA scrolls share a common characteristic. Near the end of the scrolls, there is a section where three demons from the class-B lineage are interwoven among class-A demons (Figure 11). Such shared characteristics would not exist if the prototypes for class-AB and class-BA scrolls were created independently by different painters at different times. Instead, it would be more natural to assume that class-AB/BA scrolls share a common ancestor and later copies followed this sequence as it appears in their ancestor scroll. The characteristics of class-AB/BA scrolls are summarized in Table 2.

At first glance, it is evident that the majority of demon configurations of the class-A part of the picture scroll are nearly identical to those of the Shinjuan scroll. In contrast, the demons in the class-B part contain more swaps, deformations, and omissions from the Nichibunken-A scroll. In the Kunaichō-A, OHRM, Rekihaku-B, and Tōhaku-A scrolls, a deformation refers to the change from “demon with spear and hat” to “octopus head demon” (Figure 12).

Among the class-AB scrolls, the only scroll discussed in this paper is the Beatty scroll, which retains the arrangement of demons relatively well. On another front, this method make it possible to trace the genealogy of class-BA. Based on the edit distance among the class-BA scrolls (Table 3), the minimum spanning tree is depicted as a linear connection: Geidai-A to OHRM to Tōhaku-A to Rekihaku-B (or vice versa) with distances of two, two, and one, respectively. As in the case of class-A scrolls, this lineage tree must be evaluated based on qualitative characteristics. The four class-BA scrolls contain four peculiar depictions: a demon with a “pot lid handle” attached or not, a fleeing monkey wearing a hat or not, a spear-carrying demon with an octopus head or not, and a demon holding two leaves depicted or not (Table 4). Using these binary categorical data, Figure 13 is generated as a dendrogram.

This dendrogram supports the premise that the root of the lineage tree is traceable to Geidai-A. As a result, the most straightforward iconographic transmission started from Geidai-A, which was inherited by OHRM, Tōhaku-A, and Rekihaku-B (Figure 14). First, the “demon with two leaves” motif was omitted from Geidai-A, creating a nested image for OHRM. The “demon with a spear and hat” appearing at the beginning of the scroll was transformed into a “demon with an octopus head” at this time.[14] When the ancestor of the OHRM scroll was swapped by two blocks, the arrangement of Tōhaku-A was produced. A further one-block swap resulted in the Rekihaku-B arrangement. In addition to OHRM, the Tōhaku-A and Rekihaku-B scrolls also exclude the “demon with two leaves” motif and transform the “demon with a spear and hat” into an “octopus head demon”.

Based on the preceding, Figure 15 depicts the results of estimating the editing process of the class-AB/BA lineages. Initially, these classes were created relying upon prototypes of the Shinjuan and Nichibunken-A scrolls, but at some point the “demon in a white robe” motif was omitted. The Geidai-A scroll is the closest to the class-BA prototype. The OHRM scroll was created using Geidai-A's ancestor as a template. In the process, the “demon with a spear and a hat” became an “octopus head demon”, and the “demon with two leaves” motif was omitted. The ancestors of Tōhaku-A and Rekihaku-B were born from the OHRM prototype.

Let us apply two qualitative characteristics of the iconographic specifics to the results of this estimate. The first characteristic is the presence or absence of a “pot lid handle” in the previously mentioned class-A lineage, and the second is whether or not the “fleeing monkey” in the latter half of the class-B lineage is wearing a hat. As seen in the Nichibunken-B scroll, it can be assumed that the “pot lid handle” was originally a part of the class-A lineage, but that it was eliminated in the Shinjuan scroll. As in the Nichibunken-A scroll, the “fleeing monkey” would have been depicted without a hat in the class-B prototype. These two characteristics were passed along to the Beatty and Geidai-A scrolls. In contrast, a “pot lid handle” was added after the Geidai-A scroll was created, and the “fleeing monkey” was covered with a hat after the OHRM. These characteristics are shared by the Tōhaku-A and Rekihaku-B scrolls.

The editing processes of the class-AC and AD/DA lineages remain to be estimated. As for class-AC, Kyōgaen (No. 55 in the Appendix), Sendai (No. 56), Tōdai (No. 58), and Yumoto-F (No. 59) will be considered. It is difficult to determine the edit distance between these works, however, due to the numerous additions, deletions, and swaps of demons from the prototype. The arrangement of demons in the class-A lineage appears to be based not on the Shinjuan scroll but on Nichibunken-B in all of these materials. Consequently, the “pot lid handle” is depicted in all the class-AC scrolls that have survived. It was combined with the ancestor of KCUA to form the class-AC prototype, from which the draft of Kyōgaen (dated 1775) was created.

As for class-AD/DA scrolls, the characteristics are indicated in Table 5. Similar to class-AC, AD shares many characteristics with Nichibunken-B instead of the Shinjuan scroll. Rekisaikan-B and Spencer-C, which retain the arrangement of Nichibunken-B, appear to be the closest to the class-AD prototype.

As an example, we will describe how the demons in Rekisaikan-B are arranged. Although the sequence is split and copied, the demons are identical to those in the Nichibunken-B scroll; however, the editing of the demons from Hyogo-A is more complicated. Moreover, the class-D part of the Rekisaikan-B scroll contains two demons that are absent from Hyogo-A. This suggests that the demons within the Hyogo-A scroll may have been repeatedly swapped and eliminated during the copying process from the class-D prototype.

On the basis of the preceding argument, the class-AD/DA editing process can be estimated as follows: First, the class-A component of AD/DA scrolls is based on the prototype of Nichibunken-B. The initial class-AD/DA scrolls were created from this and the ancestor of Hyogo-A. However, after the creation of the prototypes, it is difficult to ascertain the genealogies of this lineage. The “pot lid handle” is not depicted in existing class-AD/DA scrolls. Since the prototype of class-A appears to have possessed the “pot lid handle” motif as exemplified in Nichibunken-B, it was likely eliminated around the time the ancestors of these scrolls were painted. If this is the case, class-AD/DA scrolls which include the “pot lid handles” could be the missing link in this lineage.

Finally, let us summarize the estimation of the editing processes of Demons Scroll lineages (Figure 16). The most important point is that among the various scrolls in the class-A lineage, the compositional arrangement of demons in Nichibunken-B is older than that of the most famous Shinjuan scroll. In addition, the author argues that the arrangement of demons in the present-day Shinjuan scroll was created through the misordering of its paper sheets. In class-AB/BA scrolls, the arrangement seen in Nichibunken-A was combined with that of the current Shinjuan scroll, whereas class-AC and AD/DA scrolls are assumed to have combined motifs from the ancestor of Nichibunken-B, instead of the Shinjuan scroll. Since Kyōgaen, a class-AC printed book, has a colophon dated 1775, the original class-AC lineage must have been established prior to that year.

The other datable artifact is Tōhaku-B, a scroll of class-BC that was not examined in detail for this paper. Tōhaku-B contains demons from the KCUA scroll mixed with those from Nichibunken-A (the majority of which are inverted), as well as demons not found in either scroll. It is dated 1617, so it is likely that both the Nichibunken-A and KCUA scrolls were painted prior to that year. Based on the clothing and customs depicted in the scrolls, some scholars believe that their prototypes were painted in the sixteenth century. Since the Shinjuan scroll is also said to have been created in the sixteenth century, the prototype for Nichibunken-B must have been made around or prior to that time. On the other hand, since it is believed that the original Hyogo-A scroll was created during the Edo period (1603–1868), it can be assumed that the class-AD/DA lineages evolved during the same time frame or later.

The above-mentioned estimates are the result of using a digital humanities approach to study Demons Scrolls, focusing on the configuration of demons. The fact that we were able to deduce and reconstruct the rational editing process from the surviving scrolls provides additional support for the mixture theory of Demons Scrolls. However, there is one flaw in this paper's methodology. If the differences in arrangement are the result of changes in the order of the paper sheets occurring in later periods, the conclusion may be affected. Although scrolls with a high probability of misordered paper sheets were excluded from this study in advance, if a disorder is discovered in other scrolls, it will be necessary to reconsider our findings. However, if we consider this kind of disorder to be the result of editing by someone other than the painter, it may not be necessary to exclude all disordered scrolls.

To further strengthen the mixture theory of Demons Scrolls, two additional riddles must be solved. First, if there are four distinct lineages, why are the surviving non-class-A scrolls so few in number? There are at least twenty-seven copies in the class-A lineage, but only two in class-B and class-D, and three in the class-C lineages. Second, why are all the demons in the combined scrolls from the class-A lineage? If there were four distinct lineages, one would expect to find a greater number of combined scrolls from lineages other than class-A. The only exception to this rule is Tōhaku-B (class-BC, No. 71 in the Appendix); the absence of class-BD and CD scrolls is also a mystery. This may be due to the preferences of the group of painters who replicated the Demons Scroll and/or their patrons. We defer clarification of this point to future art historical research.

This digital humanities research methodology is applicable to other types of undated materials in which the separable unit in the content is ordered sequentially and there are multiple versions in which the unit is swapped, omitted, or a new unit is inserted. To name a few examples: handwritten catalogs of Japanese martial arts, other picture scrolls of which multiple copies were produced, and even drafts of novels. To broaden the potential application of this method, mutual understanding and increased collaborations between the digital humanities and other disciplines are essential.

This paper is based on an article originally published in Japanese [Yamada 2009]. The author added new data, corrected errors, updated the discussion, and edited the manuscript for an Anglophone audience. This research utilized digital images collected by the Database Project for Cultural Materials on Monsters and Specters directed by Komatsu Kazuhiko at the International Research Center for Japanese Studies (Nichibunken).