Modern natural-language systems usually represent linguistic meaning through statisticalmodels trained on large text corpora with substantial graphics-processor (GPU) resources.The present work examines a different, instrument-based hypothesis: that semantic differences between written assertions may be registered directly as differential torsion-fieldresponses, without training a statistical model. The study is formulated within Shkatov’storsimetry framework, in which the torsion field (TF) of physical objects and processes isrecorded as the torsional contrast (TC) between the measured object and the instrumentby means of single- and multi-coordinate torsimeters. To reduce spurious informationalphantoms that can accumulate in object—instrument—operator systems and affect repeatedmeasurements, we use the Method of Differential Test Assertions (MDTA). In this method,the object under test is composed from paired textual assertions with opposite meanings,and the differential torsion response between the two assertions is analysed. In 19 independent measurement sessions, a high-threshold assertion ("the total score is greater than θ"),treated as false under the experimental conditions, produced a higher torsional contrast thanthe paired true assertion ("the total score is less than θ") in every session. The mean paireddifference was +112.1 arbitrary units, and the exact one-sided sign-test probability for 19positive differences out of 19 was p = 2−19 ≈ 1.9 × 10−6. These results show a reproducibledifferential response between the paired textual assertions in the present protocol. Withinthe torsimetry framework, they support the working hypothesis that semantic content cancontribute to the measured torsional contrast of a written text; independent replication andexpanded controls will be required to determine the generality and physical mechanism ofthe effect.