Did a Government Loanword Paraphrase List Change Parliamentary Speech? A Staggered-Adoption Study of 176 Words in the Japanese Diet Record, 1998-2012

Between April 2003 and March 2006, the National Institute for Japanese Language (NINJAL) published four rounds of ``loanword paraphrase proposals'' (「外来語」言い換え提案), covering 176 hard-to-understand English-derived loanwords in public documents, each with a recommended native paraphrase. This study asks whether Diet speech, a complete record of elite language use, moved away from them. The data are 4,733 hit-count queries to the National Diet Library's Diet record API: speeches per year containing each proposed word and 60 never-proposed control loanwords, 1998-2012. Each proposal round is compared with never-proposed policy and business loanwords, using cohort-specific differences against a five-year base. The author's self-recorded time stamps, not a publicly verifiable pre-registration, place the design 35 seconds before the first analysis query. The registered primary estimate is -0.47 log points in log(1 + speeches) over six post-round years, relative to control B: a 38% decline in the geometric mean of 1 + speech count (95% bootstrap interval -54% to -18%), not a 38% fall in the number of speeches. Registered falsification tests reject this comparison: the treated-minus-control gap was already falling before the proposals, by 0.10 log points per year (interval -0.20 to -0.01), and everyday loanwords given placebo dates show a spurious ``effect'' of the same size (-0.39 log points). Against everyday loanwords the registered robustness estimate is -8% (-19% to +4%), an interval that holds only under parallel trends with that group; the not-yet-treated comparison gives -10% (-29% to +16%), but an exploratory check shows a falling pre-period gap there too. Paraphrases did not gain ground: the loanword share of loanword-plus-paraphrase speeches was 65% before and 66% after. The study finds no credible evidence that the proposals reduced loanword use in the Diet.