Roution V3.1: Language-Specific Marker Grammar in Profile-Guided JP/EN Observation Planes
Abstract
This paper reports the V3.1 correction and extension of Roution, a framework for measuring how black-box language-model responses move across route-marker profiles under controlled wording and observation operators. The contribution is not merely that a Japanese profile was expanded or that four additional Japanese terms were tested. The contribution is the observation that the same marker names do not imply the same marker grammar. English and Japanese profiles must be read through different language-specific response grammars before they are mapped into the shared Roution profile space. The V3.1 data base combines a JP600 golden profile and an EN600 golden profile. The Japanese profile is a full-read, QA-crosschecked, exclusive marker profile: all 60 aggregate rows satisfy marker-sum = n. The English profile is a cleaner subject-attribution and axis-gate reference plane, mostly coded by mechanical or regex rules and intentionally not forced into the Japanese exclusive-count criterion. A superficial table reading would say that boundary denial (BD) is dominant in both languages. The corrected reading is deeper: boundary routing is conserved across languages, but its marker realization is language-specific. English tends to expose route shifts through compact subject-predicate and conclusion-first structure; Japanese distributes the same route through predicate-final syntax, indirectness, politeness, and pragmatic cushioning. V3.1 therefore treats JP and EN as separate observation planes: not because they are incomparable, but because pooling them would erase the observed marker-grammar difference. The central rule is: same Roution profile space, same coding anchors, different language-specific marker grammars, no pooled denominator. V3.1 is also a correction layer on a specific published V3 result. V3's candidate-coded Table 1 reported risk_pressure_uncertainty_shift (RS) at 255/960 (26.6%), of which 100 were Japanese-language responses. V3.1's full manual re-read of the JP480 corpus found RS = 0/480 (and RS = 0/600 in the combined JP600 profile). The V3 RS value is not carried forward as a verified JP marker result; see Section 9. Lineage: V2 (doi:10.5281/zenodo.20915489) defined Roution; V3 (doi:10.5281/zenodo.20994120) introduced the EXP-10 measurement protocol; V3.1 (this version) corrects and extends V3. Aggregate data: https://github.com/Tsune2034/roution/tree/public-data-release-20260709/data_releases
// Source
Authors: Daisuke Tsunemori