Entry of technology vocabularies by industry
First fiscal year in which each of 20 technology vocabularies entered the 10-K disclosure of each of nine industries, fiscal 1996–2025, and the lag of construction behind the earliest entering industry. An industry enters a vocabulary when at least 5% of its operating filers and at least three firms mention it in two consecutive years.
First sustained entry by technology and industry
Each dot marks the first sustained entry of an industry; construction is shown with the larger dot. A lag marked “≥” is a lower bound, because the leading industry was already above the threshold in fiscal 1996, the first year of the panel. The slider limits the display to the entries up to the selected fiscal year, and the play button advances it year by year. Hovering over a dot shows the industry and the year.
Entry lag of construction by technology
Notes
For the vocabularies first dated after 2005, where the panel is not censored, construction entered 4 to 7 years after the leading industry: cybersecurity 4, IoT 5, cloud computing 6, big data 6 and blockchain 7 years.
For the internet vocabulary, construction was among the industries above the threshold in fiscal 1996.
Seven vocabularies (Six Sigma, RFID, wearables, autonomous equipment, VR/AR, 3D printing and quantum computing) entered at least one other industry but did not enter construction within the panel.
Of the 66 technology-industry entries dated 2006 or later, 11 are risk-first, that is, located mainly in the risk-factor section (Item 1A) in the first years after entry; seven of the 11 concern cybersecurity.
The artificial-intelligence vocabulary is not shown. Its matches before 2015 did not pass the precision audit (initials and outline labels such as “Part A.I”), and it is analyzed with passage-level methods in a companion study.
Filing inventory
One cell per company-year, fiscal 1996–2025, shaded by the number of technology vocabularies the filing mentions. Hovering over a cell lists the vocabularies; selecting it lists the first sentence of each vocabulary, with links to the filing on sec.gov (for most construction and construction-machinery filings, at the sentence).
Adoption by technology
Share of operating filers in each industry that mention the selected vocabulary, fiscal 1996–2025. Construction is shown with the heavier line.
Statistical tests
Tests reported in the paper. Under the headline entry rule (5% of filers), the mean entry rank of construction does not differ from chance; the industries ranked later than chance are pharmaceuticals, construction machinery, retail and utilities. At 10% of filers, construction ranks 8th of 9 under every entry rule tested. All estimates are associations in corporate disclosure.
Precision audit of the lexicon
Before a vocabulary was measured, sampled term contexts from the filings were coded by three open-weight language models (Qwen2.5-7B, Phi-4 and Ministral-8B) for whether the match denotes the technology. A vocabulary was retained if its precision was at least 80% overall and at least 70% in every era with 30 or more samples; the other candidate families are excluded from the adoption measures.
Method
Data
All original Form 10-K filings (including the 10-K405, 10-KSB and 10-KT variants) of US registrants in nine SIC-defined industry bands, fiscal 1996–2025: 48,180 filings by 5,238 registrants, frozen under a SHA-256 manifest. Fiscal 1996 is the first year because the EDGAR electronic-filing mandate was phased in between 1993 and 1996, and earlier electronic filings are a self-selected subset.
Technology lexicon
Twenty-four candidate families of case-aware, boundary-guarded term patterns. Each family was audited for precision before it was measured: 20 were retained (automation, TQM and Six Sigma, flagged as noisy in advance, passed the audit), and four are excluded from the adoption measures (lean and modular or prefabricated construction, confirmed as noisy; artificial intelligence and exoskeletons, which did not pass). Abbreviations (BIM, ERP, RFID, GPS, IoT, AI) are matched case-sensitively.
Entry rule
An industry enters a vocabulary in the first fiscal year in which at least 5% of its operating filers and at least three firms mention it, sustained for two consecutive years. The rule is one of a prespecified grid of 18 (2%, 5% or 10% of filers; one, two or three firms; one or two sustained years), all of which are reported in the paper. Entries at the 1996 floor are left-censored, and the corresponding lags are lower bounds.
Operating screen
Filings of fewer than 5,000 words are excluded. Because XBRL financial data do not exist for most of the years before 2009, the screen is textual.
Scope
Entry into disclosure dates the appearance of a vocabulary in an industry's annual reports, not the adoption of the technology in operations. All results describe corporate language, and firm-level estimates are associations.