Digital Methods on Ancient Texts - using Fear as an Example

Team1 in ANEE has developed digital tools and workflows to study the semantic domains of different lexemes in the Akkadian language. This methodological development process is now more or less ready, and the tools and workflow can be applied to study real-life research questions.

The history of emotions in ancient Mesopotamia is a fascinating but relatively little studied topic that can benefit from computational analysis of large textual datasets. In a recent article, a group of the Team 1 researchers in ANEE applied their set of digital tools to study the semantics of five fear verbs in Akkadian.

The English language has a number of verbs that belong to the semantic domain of fear, such as to be afraid, to be frightened, and to be scared. The same applies to Akkadian, and there are five common verbs that can be translated roughly as “to fear” in English: adāru, galātu, palāhu, parādu, and šahātu. It is evident, however, that these verbs are not synonymous, being used in different contexts and having a wider range of meanings than “to fear.”

These five verbs were analyzed in a snapshot of the Open Richly annotated cuneiform corpus (Oracc) from February 2019. The snapshot consisted of ca. 7300 lemmatized texts mostly from the Neo-Assyrian period of various genres excluding lexical lists. We produced two variants of the data set: one with standard lemmatization as in Oracc with only minor modifications such as replacing various names with placeholders, and one where the roots of the words were combined and homonyms were separated. This prevented the methods from confusing similar looking words with completely different meanings and uses (e.g. rabû “big” and rabû “set”, now rabû_1 and rabû_2).

We used two language technological methods, Pointwise Mutual Information (PMI) and fastText to analyze the semantic contexts in which these fear verbs and their derivatives occur. PMI enables us to detect the neighbouring words that typically occur in the same context with our word of interest. Accordingly, the PMI results for the word to fear might include words like spider and dark, as they might occur in sentences (from modern era) like; “He feared the spiders he knew were lurking in the dark room.”  FastText can be used to find words that appear in similar contexts as our word of interest, i.e. the detected words have neighbouring words in common. These words can be synonyms (to be afraid), antonyms (to dare), or other lexemes that are used in a similar way. To continue the previous example, we could have sentences like “He was afraid of the spiders in the dark room” or “Do you dare to go into the dark room filled with spiders?” where “afraid of“ and “dare to” both have “the dark room” and “spiders” in their context.

The results of the PMI and the fastText were visualized as networks. We used network analysis to compare and analyze our results, and the search interface Korp in the Language Bank of Finland to study the words of interest in their full context.

In addition to developing the methodological approach and making that publicly available, the article has implications for Akkadian lexical semantics. For example, the close connection between certain anger and fear verbs (especially in martial contexts) was apparent in our results. The interesting relationship of šaḫātu “to fear” with “love” is worth mentioning here as well. Based on our analysis, this connection is related to piousness toward gods. In this instance, the semantic connotations reach the modern world as well. The idea of “fear of god” is well established in the Bible, e.g, a well-known proverb in the Hebrew Bible (Proverbs 9:10) says that fear of God is the beginning of wisdom. Connections between ancient Near Eastern literature and the Biblical Proverbs (as well as other parts of the Hebrew Bible) are well documented. The influential idea of fear of god as a positive emotion therefore has a strong background in Akkadian language and Mesopotamia three thousand years ago.

The digital methods, such as the ones we used in this article make it possible for us to analyze large amounts of text. Although our statistical methods cannot significantly improve the existing lexical semantics (esp. CAD, but also other philological work on Akkadian lexical semantics) in terms of adding new nuances and new “meanings” to already known ones, what this approach can do is analyze the quantitative “meaning” of a word of interest. Thus, the digital methods make it possible to execute a comparative analysis between words that is based on an openly available textual dataset. The method makes it possible to trace the common and uncommon patterns in the way words are used in the Akkadian language. This suggests interesting possibilities for examining the sociolinguistic aspects of language, for example, where and when a certain vocabulary is used.  At last, the methods employed in this paper provide an efficient way to highlight various uses of words without the need of close-reading all their occurrences in the texts individually. 

In terms of the study of emotions, there has been very little previous research which systematically reviews a group of frequently occurring words, in our case semantic fields of five verbs of fear. The article is therefore contributing both a new methodology for studying lexical semantics of Akkadian, but also on the emerging field of study of emotions in Mesopotamia. We hope that our results will raise new questions and provoke readers to study some of these words from a new perspective with the help of the dataset we have published on Zenodo.

Bibliographic details:
Svärd, Saana; Alstola, Tero; Jauhiainen, Heidi; Sahala, Aleksi & Lindén, Krister (2021). Fear in Akkadian Texts: New Digital Perspectives on Lexical Semantics. In The Expression of Emotions in Ancient Egypt and Mesopotamia, Shih-Wei Shu, Jaume Llop-Raduà (eds.). Brill, Leiden. pp. 470-502. (Open access)

Here is link to the article, and here is link to the data.