This paper proposes a conceptual framework exploring how text embeddings provide an alternative approach to addressing Patrick Wilson’s “indeterminacy of subject” issue. The purpose of this paper is to demonstrate that embeddings represent a significant shift from rule-based subject indexing to a geometric model of semantic representation. The analysis engages with recent interpretability work suggesting that large language models may exhibit cross-lingual regularities in their internal representations; however, the scope and stability of such effects remain debated, and we do not assume language-independent conceptual universals.
The paper critically reviews Patrick Wilson’s notion of “sense of position,” outlining traditional indexing and classification limitations. It introduces embeddings as mathematical translations of textual content into multidimensional semantic vectors, emphasizing their ability to represent documents through “semantic centroids.” The discussion integrates empirical evidence from recent literature, highlighting real-world applications and evaluations of embedding effectiveness.
While embeddings do not resolve the philosophical indeterminacy of “subject,” they provide an operational reframing: a document can be represented by a dense vector that supports retrieval without committing to a single, context-free subject designation. Read in Wilson’s terms, this amounts to adding a fifth way to obtain a possible subject representation – a dense vector or “semantic centroid” – alongside his four (author’s purpose, dominant figure and/or figure-ground, counts of key terms and an inferred unifying principle).
This paper is original in explicitly connecting Wilson’s philosophical critique to contemporary embedding technologies and in articulating embeddings as a fifth way to determine a possible subject representation (a dense vector or “semantic centroid”) alongside Wilson’s four. By reconceptualizing the “sense of position” from a hierarchical logic to geometric proximity, the work offers a novel theoretical framework and practical insights for information science.
