Our team approached the secret of love from a language-technological perspective. Instead of reading a corpus of texts and looking for relevant passages, we used statistical methods to find the most typical contexts in which the word râmu (“love; to love” in Akkadian) appears. We downloaded a dataset of 7,346 Akkadian texts from the Open Richly Annotated Cuneiform Corpus (Oracc). Most texts are Neo-Assyrian, but other historical periods are attested as well. As genre has a huge impact on the themes and vocabulary used in a text, we studied the genre-specific semantic contexts of the word râmu and its derivatives.
We applied two language technological methods, pointwise mutual information (PMI) and fastText, to study the semantic similarity of different words in our corpus. PMI excels in finding words that typically appear in the same context and fastText in finding words that appear in similar contexts, like synonyms and antonyms. For example, “chocolate” and “cats” are words that often co-occur with the verb “to love” in English and could be found using PMI. At the same time, words that are used similarly to “to love” and could be detected using fastText are “to like” and “to adore.” From PMI and fastText results we created semantic networks which show the words that appear in the same contexts (PMI) and in similar contexts (fastText) as love words. You can read more about our methods and their application to the study of emotions in this blog post.
As expected, love words appear in different semantic contexts in different genres. In royal inscriptions, râmu and its derivatives primarily characterized the relationship between the king and the gods. The gods love the king who obeys the divine will and maintains the correct order in his realm. In love literature, our results highlighted words related to erotic and sexual love between earthly or divine couples. Typical PMI collocates of râmu in love literature include ṣīhtu (“laughter”) and dādu (“darling”), both being associated with lovemaking in Akkadian. The results in the genre of letters were again different, and the characteristics of Neo-Assyrian official correspondence were obvious. Here râmu characterizes instrumental love between a superior and the subordinate, something that the servant both desires and wants to show to his master.
One of the most intriguing results of our research does not relate to love as such but to the way how emotion words are generally used in Akkadian. In a fastText network, it turned out that emotion words do not cluster together according to the emotion they represent but according to the genre in which they are attested. For example, different love words do not constitute a cluster, but the different emotion words (love, anger, and fear) attested in royal inscriptions cluster together. This suggests that the same word is used in very different contexts in different genres, but the vocabularies used in a given genre are rather consistent. Furthermore, this result strengthens the validity of the genre labels given by modern scholars to ancient Mesopotamian texts, as the genres appear to constitute cohesive groups from a statistical point of view.
Above: Semantic network of emotion words in Akkadian, created using word vectors from fastText. The graph shows how different emotion words cluster together according to the genre rather than to the emotion they represent. Each node in the network is an emotion word, and love words are underlined. Numbers equal to different genres (e.g., 11 is royal inscriptions). The figure was first published in Alstola et al. 2022.
As “love” has various nuances and contexts of use in English, so has râmu in Akkadian. Different aspects of râmu become visible in different textual genres ranging from the servant’s desire of his master’s love to sexual joy of two lovers. This emphasizes the need to pay attention to the composition of the textual dataset one is studying: only a diverse corpus can adequately highlight the most important semantic contexts of a word.
The article can be freely downloaded on the publisher's website.
Reference:
Alstola, Tero, Heidi Jauhiainen, Saana Svärd, Aleksi Sahala, and Krister Lindén. 2022. “Digital Approaches to Analyzing and Translating Emotion: What Is Love?” Pages 88–116 in The Routledge Handbook of Emotions in the Ancient Near East. Edited by Karen Sonik and Ulrike Steinert. London: Routledge.