Speaker
Description
Political commitment is easy to over-read from modal-word counts. The same form can realize prediction, obligation, discourse management, or a future undertaking, and a category can occupy a larger share of a functional inventory even while becoming less frequent in the discourse as a whole. This study examines those measurement problems in a longitudinal corpus of Donald Trump's monologic political speeches. The protocol-defined corpus contains 156 speeches from 2015 through July 2026 and 843,837 analysis-ready lexical tokens. A deterministic, precision-first annotation layer identified 12,325 commitment-related construction events: 5,530 epistemic, 1,859 deontic, and 4,936 commissive. A reserved stratified sample of 300 detected events was independently double-coded and adjudicated before final interpretation; the validation was used as a construct audit of domain assignment rather than as a claim of exhaustive recall. Composition was modeled with speech-clustered generalized estimating equations, while absolute event rates were modeled with negative-binomial regression using document token count as an offset. The most stable substantive pattern is a contraction of explicit first-person commissive constructions: density falls from 93.4 events per 10,000 tokens in the first-campaign period (P1) to 25.7 in the second-presidency period (P5), with P5 showing both lower genre-adjusted commissive membership (OR = 0.28, 95% CI [0.22, 0.36]) and a lower absolute rate (IRR = 0.23, 95% CI [0.19, 0.28]). The direction survives rally-only and formal-address-only comparisons, month-clustered covariance, exclusion of the longest speeches, and a conservative removal of formulaic first-person futures. By contrast, higher epistemic or deontic shares do not consistently imply higher absolute rates, and latent lexical-content adjustment materially changes some period coefficients. The paper therefore treats periods as historically situated contrasts, not causal effects of office or candidacy. The broader contribution is methodological: longitudinal political-language research benefits from separating form from function, share from rate, and event frequency from the dependence structure of speeches.
