Systematic Analysis of Image Schemas in Natural Language through Explainable Multilingual Neural Language Processing

Proceedings of the 29th International Conference on Computational Linguistics, pages 5571–5581 October 12–17, 2022. 5571 Systematic Analysis of Image Schemas in Natural Language through Explainable Multilingual Neural Language Processing Lennart Wachowiak King’s College London lennart.wachowiak@kcl.ac.uk Dagmar Gromann University of Vienna dagmar.gromann@gmail.com Abstract In embodied cognition, physical experiences are believed to shape abstract cognition, such as natural language and reasoning. Image schemas were introduced as spatio-temporal cognitive building blocks that capture these re- curring sensorimotor experiences. The few ex- isting approaches for automatic detection of image schemas in natural language rely on spe- cific assumptions about word classes as indi- cators of spatio-temporal events. Furthermore, the lack of sufficiently large, annotated datasets makes evaluation and supervised learning diffi- cult. We propose to build on the recent success of large multilingual pretrained language mod- els and a small dataset of examples from image schema literature to train a supervised classi- fier that classifies natural language expressions of varying lengths into image schemas. De- spite most of the training data being in English with few examples for German, the model per- forms best in German. Additionally, we anal- yse the model’s zero-shot performance in Rus- sian, French, and Mandarin. To further inves- tigate the model’s behaviour, we utilize local linear approximations for prediction probabili- ties that indicate which words in a sentence the model relies on for its final classification deci- sion. Code and dataset are publicly available1. 1 Introduction In the tradition of embodied cognition, image schemas have been proposed by Lakoff (1987) and Johnson (1987) as spatio-temporal cognitive build- ing blocks that capture recurring sensorimotor ex- periences. For instance, in early infancy we expe- rience many objects with the properties of a CON- TAINER, i.e., having an inside and an outside sepa- rated by a boundary. The image schema CONTAIN- MENT captures this experience and is subsequently used to make sense of new experiences while at the same time also influencing how we think and talk 1https://tinyurl.com/24haedv5 Figure 1: Example of the image schema CONTAIN- MENT: From experiencing different types of a CON- TAINER in early infancy (left); to the development of the schema (middle); and the usage in language on ab- stract topics (right) about abstract concepts, such as thinking, emotions, or life (see Figure 1). In order to systematically analyse the occurrence of image schemas in natural language, we pro- pose to build on the recent success of multilin- gual pretrained language models and a small set of examples from image schema literature (Hurti- enne, 2017) to train a supervised classifier based on XLM RoBERTa (XLM-R) (Conneau et al., 2020) to classify natural language expressions into image schemas. An image schema detection model as ours could help linguists to explore the use of im- age schemas efficiently and effectively in large text corpora. It can guide researchers who, for instance, investigate how the use of image schemas differs across languages and cultures (e.g., Choi and Bow- erman, 1991; Papafragou et al., 2006), how the lan- guage of children with …

Content truncated to 500 words