Feminist Marxist thought has long been looking into motherhood as labour, defending its role in perpetrating patriarchal capitalism. That role has been examined, consciously or not, by many artists producing socially engaged art in the 20th century, as it has happened in Social Realist enthusiasts in Brazil. To analyze images produced by those artists through a data driven analysis that accounts for the depiction of motherhood, it is critical to understand how to best identify it in images. From previous research conducted by us, we have obtained a manually annotated and structured dataset consisting of 372 engravings of 57 Brazilian socially engaged engravers. Drawing upon this dataset, we aim to gain insights into how multimodal networks understand the concept of motherhood. Does the output of such networks, which can automatically label images, agree with human annotators for motherhood? Are there shared characteristics within engravings automatically tagged with motherhood? We used a pre-trained multimodal network for two main reasons: (1) multimodal networks combine multiple types of data, for example, text and images, and (2) the interest in working with a pre-trained model, as opposed to finetuning a preexistent model, as doing so grants insights into how these models have been trained and how they would understand motherhood. These models generate a probability score indicating the likelihood of a word's association with an image and vice versa. We supplied the pre-trained model with our image corpus with the task of determining which engravings were most likely to represent motherhood. The preliminary results demonstrate how the model assigns high scores of similarity to engravings that portray children held by women. However, not all the engravings that were manually tagged as motherhood are given a high score by the model, for example, the engravings that depict women working while children are around.