A semantic and context-dependent approach to the interpretation of ‘near’ in historical English Lake District narratives

A common and intuitive way of identifying the proximity relationship between two entities is to use the preposition ‘near’ (e.g. ‘near (inn, village)’). However, the ‘near’ relation is vague, often asymmetric and context-dependent, and hence incorporating factors such as the effect of type, size and scale of reference object and associated features is required for cognitive modeling. In this work, we interpreted spatial proximity as described in the historic Corpus of Lake District Writing comprising travel narratives as early as the 16th century. At a time when modern transportation modes were not available, it is interesting to explore how proximity has been perceived and recounted with various context factors explicit or implied. We utilized pre-trained BERT and its variants to first identify the broader semantics of ‘near’ and generate contextual embeddings. We further identified the contextual factors of spatial nearness. Finally, we used quantitative distances between entities to measure the departure of context-dependent proximities from objective notions of distance.