Journal of Systems Thinking · Volume 6, Issue 1 · 2026-03-13
Measuring Context as Structural Organization in Natural Language
A DSRP-Based Empirical Analysis Using WinoGrande and SQuAD 2.0
Derek Cabrera, Laura Cabrera
DOI 10.54120/jost.v25i1.dc6mcnl6 (registration pending)
Abstract
ABSTRACT: Context is central to language interpretation, yet it is rarely defined in a way that permits direct measurement. In linguistics, cognitive science, and artificial intelligence, context is often treated informally as surrounding text, background knowledge, or situational framing. This study proposes a structural operationalization of context grounded in DSRP theory, in which context is defined as the organization of information relative to a target element, C = O(I | t₀). Using 600 items from WinoGrande and SQuAD 2.0, we coded structural context with DSRP-derived features and computed context scores mechanically from those features. Inter-rater reliability ranged from κ = 0.42 to κ = 0.83. Structural context scores were significantly associated with interpretation accuracy ( r = 0.347, p < 0.001). Accuracy increased from 4.7% in the low-context bin to 64.0% in the mid-high-context bin, a 13.7-fold increase. Information quantity alone did not significantly predict interpretation accuracy. In addition, low-context items frequently exhibited underconstrained interpretation spaces, suggesting that structural context may also function as a measure of interpretive determinacy and benchmark ambiguity. These findings support the view that interpretation depends not simply on how much information is present, but on how that information is organized, and that weak contextual organization leaves interpretation less determinate.
© 2026 Journal of Systems Thinking (JoST) · CC BY-NC-ND 4.0
