Published subject indices are typically constructed of alphabetized subject headings and sub-headings, with the latter indented, as demonstrated in Figure 2. Therefore, in order to include all the information in the index, it was necessary to correctly identify subject headings and sub-headings. This was done algorithmically in Python 3.10, with human input and supervision. The stages were as follows:

At the end of the process, the subject headings and sub-headings were joined by a comma, and page ranges were expanded.

Table 1 provides an example of data entry for the output appearing in Figure 2.

Every reference in the source index was parsed and divided into the following columns: 1. author and title identifying number, 2. reference, and 3. Page number, as seen in Figure 3.

Table 2 provides an example of data entry for the output appearing in Figure 3.

Clearly identifying the various ancient authors and works was challenging as publications use highly varied conventions to refer to ancient authors, works, and references [Romanello 2018]. The source indices contained more than 12 thousand unique ancient works from different traditions and languages. For example, The New Testament book, First Corinthians, can be written in various abbreviated forms such as “1 Cor.”, “First Corinthians”, “I Cor.”, etc. Many ancient works have several naming conventions for the title and for the author. Furthermore, the chapter and verse number can also be written according to different conventions, using Roman or Arabic numerals, and different types of dividers. To meet this challenge, we constructed an author-title database of modern naming conventions and abbreviations of ancient authors and works, which the program consults to identify the references included in the indices. This database utilized and collated existing free databases such as the TLG canon of Greek Authors and Works, abbreviation tables from various Greek and Latin dictionaries available online, the Classical Works Knowledge Base, SBL handbook, and other lists. Many additional authors and titles for Greek, Latin, Coptic, Hebrew and Syriac works were added incrementally to the database as more indices were processed and additional authors and titles were identified manually. The script checked each line in the source index against this database to identify authors and titles. When a match was made of both author and title, a work number was assigned. Fuzzy matches were not used because different works can have very similar titles (e.g., 1 Corinthians and 2 Corinthians). Lines which were not identified as authors or titles were assumed to be of references and page numbers. These were distinguished by manually inputting the differentiating character (usually a whitespace) between them. This script generally produces good results, but with some errors as a result of unrecognized authors or titles, or incorrect differentiation of reference and page numbers, usually in cases of unusual references. Therefore, an expert examined the outputs (in csv format) to delete or correct wrong rows.

To perform network visualization, we first manipulated the files to create a network graph based on shared subjects and references. The validated table was grouped by references and the number of shared references between each subject was found. For example, the rows in Table 7 would produce the network graph in Table 8:

Using Gephi [Bastian et al. 2009] we visualized the network graph of subjects, as seen in Figure 7. In this visualization of the graph, words from subject tags which share the most references are adjacent on the network, while subjects not sharing references are far apart. The size of each node is determined according to the number of incoming and outgoing links (edges), so that subjects (nodes) with the most source references appear largest. Furthermore, communities of closely connected subjects are determined using the Louvain method for community detection and marked by color. This creates a mapping of the whole field based directly on the scholarly research, intuitively displaying the centrality or marginality of specific subjects, the relationship between subjects, and the different sub-disciplines of the study of ancient religion. The visualization is also instructive concerning subjects which are on the borders between sub-disciplines, that is, which are shared by one more of these sub-disciplines, as opposed to subjects which are in the core area of each. Figure 7 illustrates the part of the subject network dealing mainly with core subjects of two ancient religions - Jewish (in purple) and Christian (green), and some subjects connecting both areas.

Though the general subject network is instructive, it is very large and therefore not so informative concerning specific subjects or subject areas. We therefore constructed an interface for online users to create smaller networks ad-hoc, based on the same principles, using the Python module Networkx. In both cases, to reduce congestion, no more than 70 of the largest nodes (i.e., subjects with the most references) are shown. There are two options for network construction:

A heatmap is a visual representation of data based on color allowing quick identification of high, medium and low values of a variable. Using the Seaborn [Waskom 2021] visualization library for Python, we constructed a tool through by which users can produce heatmaps of the subjects sharing many references with a specific subject, per century of the reference. This is based on the same network graph file from the network visualization, but divided according to the century in which the source reference was written. The color shade corresponds to the number of references tagged by unique subjects, divided by all references from this century, serving as a normalized measure of the salience of the connection between the two subjects in a certain period. Figure 9 displays the heatmap of the word “oath” with other words through the centuries. Since some centuries are overrepresented in the database, the number of references tagged by each subject in each century was divided (normalized) by the total number of references of that century in the database to provide an index for the color shade, shown on the right column.

With the expansion of the database, heatmaps can serve to understand how the connection between subjects changed historically over time, providing insight on historical processes or on the scholarship on which the database is based. For example, Figure 6 indicates that most of the studies of oaths focused on Classical Greece rather than on later periods. It may also indicate that the sources speak of women swearing oaths only in sources from the 5th century BCE and not earlier, or that oaths by goddesses Persephone and Kore were less popular before this period. The heatmap cannot, of course, demonstrate actual historical trends, but it can provide pointers, questions and intriguing phenomena which can then be followed up by research.

A senior scholar on ancient religion is writing a book on oaths and swearing in antiquity. Searching for “oath” or “swear” and their derivatives in full text databases returns thousands of results. Therefore, to start their research, together with reading scholarship on the subject, they turn to Tiresias. Here they use several dimensions of the database to gain a general view of the subject in contemporary scholarship.

First, they search for “oath” among the subjects indexed by Tiresias. “Oath” appears in 621 different subjects. This list provides the scholar with a detailed overview of the different sub-headings in which oaths in antiquity have been discussed in the scholarship: for example, “language of oaths, and gender”, “homicide trials, oaths in” or “blood rituals, surrounding oaths”.

Second, they create a linked subjects heatmap for “oath” by entering the keyword in in the relevant search form and receives a heatmap graph showing the subjects sharing most references with “oath” in the database, sorted by century of the reference (see Figure 9). Here, they see that “sacrifice”, “women” and “witnesses” and “law court” are subjects strongly linked with oaths. Moreover, they see that almost all the references linked to oaths are from before the 4th century BCE. This can lead to the research question: are oaths in decline after this period? Or is there not enough scholarship on oaths after the 4th century BCE?

Pursuing these questions and a select set of sub-headings, they decide to limit their research to oaths by women. They search Tiresias for references tagged with both “women” and “oaths”. Currently there are 13 validated results for this search, and 220 unvalidated results (see above for an explanation of the validation process).

They start from investigating the validated results using the texts provided and can also go directly from the results to the modern publications in which they were discussed. After studying these, they turn to the unvalidated results, which were mentioned only in one book as connected to the subject. Some of these results do not appear relevant to the subject and can be discarded, but some of them are relevant, and considering their large number this facilitates location of relatively less known texts on the subject.

A BA student in religion, writing a term paper on conceptions of resurrection from the dead in early Christianity. The student reads scholarly literature on the subject, but as this is a large subject is finding it difficult to cope with the large and varied literature. Furthermore, the term paper instructions called for engagement with primary sources on the subject, and (beyond the New Testament) they are finding such sources difficult to locate. They turn to sites for searching the full text of Greek and Latin literature, and search for “resurrection”. This supplies them with some texts for analysis, but these are limited to cases in which the keyword explicitly appears in the text, and only in translations of Greek and Latin texts; furthermore, some of the cases turn out to be not actual discussions of resurrection but random mentions.

Turning to Tiresias, they search the same keyword among the validated results. They receive 76 results, with texts from the Hebrew Bible, second temple literature, the New Testament, Church fathers and rabbinic literature. They can read these texts in the original language (Greek, Latin, Hebrew and Aramaic) and in facing English translations. These results are not limited to cases in which the keyword appears in the text, but include texts discussing the subject without the keyword. Furthermore, all the results are relevant to the subject. Of these, they choose two ancient sources for further study, assisted by the detailed sub-heading tags. After choosing these, they click on to the modern publication in which these ancient sources were discussed, to find scholarship on the subject.

An instructor is building a new module on the history of emotions in antiquity. The study of the ancient sources is central in the module, as they are also interested in a broad, comparative view of several ancient cultures. When thinking of the general subjects and structure of the module, they turn to the heatmap and network visualizations to gain a general feeling of the subject. Here they see connections with Stoic and Platonic thinkers, but also the connections between emotion, the body and nature (Figure 10). Searching for “anger”, “love” and other emotions provides more information.

More concretely, searching the database for “emotion” leads to 54 validated results, from a wide range of periods and genres, as well as links to the modern publications. Some of these they find useful for background discussion, others for study together with their students, with texts and translations already available. Here too, searching for several other terms linked to emotions (some of them obvious such as “love” or “fear”, others found from the network, the heatmap or the subject subheadings of the validated results) leads to further texts on more specific subjects.

In class, the instructor asks their students to search for the term “anger” on Tiresias and to explore the results (for digital humanities usage in the classroom, see [Locke 2017], [Fyfe 2018]). This leads to a discussion of the validity of using english terms of emotion to describe ancient phenomena and to how using different types of indices can direct our focus and structure our worldview, as well as more specific questions concerning the validity of the the method used by Tiresias to locate subjects and texts.

In summary, the usage scenarios were developed based on our experience in academic research and teaching. Scholars are likely to become highly skilled in using the continually developing functionality of Tiresias and adopt it as one of their standard tools for aiding research based on primary sources. Users with simpler needs can also benefit substantially from access. For example, the BA student will realize that studying in the humanities can be supported uniquely by innovative digital technologies that were developed in academia, thus forming a mental link between humanities and innovation. Further, usage of Tiresias in addition to library databases will sharpen the understanding regarding the difference between primary and secondary sources. The teacher would be able to show younger learners that novel technology can help uncover historical knowledge and that the study of history may involve interactive assignments to promote curiosity and literacy, while problematizing the methods used to create such technologies and their impact on our thought.