Semantic Space of Eurysemic Verbs: Vector Representations

The paper proposes a quantitative approach to semantic eurysemy based on Old English data, applying methods of computational linguistics and diachronic distributional semantics. The object of the study is the semantic space of the Old English eurysemic verb dn “to do” in comparison with the semantically concrete verb etan “to eat.” The aim of the research is to identify and describe the structural and semantic characteristics of the eurysemic verb dn in Old English using methods of distributional vector semantics, specifically through the construction of context vectors based on Positive Pointwise Mutual Information (PPMI). The methodological framework includes three complementary techniques: Shannon entropy for measuring the diversity of context distributions, DBSCAN clustering for detecting the presence of dense semantic groups (clusters) versus non-clustered noise, and Principal Component Analysis (PCA) for evaluating the geometric shape of the distribution (whether it is spherical or elongated). The analysis reveals that the qualitative features of eurysemic words—high semantic generalization, a syncretic (diffuse) structure, and wide lexical combinability—are systematically reflected in the parameters of the vector space. In particular, for the verb dn, these features correlate with a high entropy value (7.47 bits), indicating diverse and evenly distributed contexts; an exceptionally high proportion of non-clustered noise usages (93.86%), suggesting the absence of stable semantic clusters; and a spherical geometry (PC1 ratio = 0.057), showing that the contextual variation is evenly distributed across multiple directions. In contrast, the verb etan exhibits lower entropy (4.98 bits), a low proportion of noise (27.89%), and an elongated geometry (PC1 ratio = 0.189), all of which point to a more compact, structured semantic organization. These findings contribute to a more precise understanding of the status of eurysemy in diachronic linguistics and allow for the identification of its system-defining properties. The novelty of the approach lies in applying methods of distributional semantics to the study of broad meaning in historical languages, thereby expanding the methodological toolkit of historical semantics and providing quantitative criteria for diagnosing degrees of semantic generalization in ancient texts.