The Language of the Inexplicable: Computational Analysis of Project Blue Book (USAF, 1947–1969)

Project Blue Book (USAF, 1947–1969) constitutes the only systematic governmental

programfortheinvestigationofunidentifiedaerialphenomenawithaccessibleprimary

data. This work applies a natural language processing (NLP) pipeline of 13 analyses

to a raw corpus of 1,267 processed reports and a labeled analytical subset of 1,247

declassified reports, addressing three research questions: (1) are there quantifiable

linguistic differences between cases classified as Unknown and explained cases?; (2)

what thematic and rhetorical structurecharacterizes the institutional USAF discourse

in the face of the inexplicable?; and (3) can the five anomalous physical performance

observables described by the Pentagon’s AATIP program be detected through lexical

analysis? The results show that Unknown reports are significantly longer (x = 2,538

words vs. x = 1,737 in explained cases) and exhibit lower lexical diversity as measured

by length-independent MATTR (MATTR=0.496 vs. x = 0.521 in explained cases;

H = 36.2, p < 0.001). The hedging analysis reveals a counterintuitive pattern: it is

the categories with conventional identification (Aircraft: 25.7 hedges/1kw; Balloon:

25.1) that show the highest density of epistemically hedged terms, while Unknown

(17.7) and Insufficient Data (17.5) fall below (H = 15.3, p = 0.018), suggesting that 

epistemic hedging is employed rhetorically to certify conventional identifications,

not to describe the inexplicable. Unknown reports also exhibit a higher presence of

instantaneous acceleration and propulsion-free flight descriptors (O1: 47%, O2: 27%,

compared to 7–11% in explained cases). A supervised classification model achieves

AUC=0.61, confirming that the differences are real but not determinative. The

inter-corpus analysis with ASRS (civil aviation, n = 47,000) and NTSB (accidents,

n = 24,124) reveals that the Blue Book corpus is 5.6 times longer and exhibits hedge

rates 5–8 times higher than civil aeronautical reports. The findings suggest that

the corpus reflects a militarized institutional uncertainty discourse quantitatively

distinct from the civil technical register, and that Unknown cases leave measurable

though non-classificatory linguistic markers.