Canada’s Translation Bureau reported 253 million professionally translated words in 2025-2026, down from approximately 325 million a year earlier. Its internal machine-translation service, GCtranslate, processed about 380 million words by June 30, 2026. The figures are not directly additive or perfectly comparable, but together they document a rapid change in how federal language demand is being handled.

Two volumes, two different measures

The Translation Bureau says its professionals translated approximately 253 million words during fiscal 2025-2026. Of that total, 43 million words served Parliament and 210 million served federal departments and agencies. Separately, GCtranslate processed approximately 380 million words from its launch on June 10, 2025 through June 30, 2026 across 69 federal organizations.

The periods, users and production methods differ. The 380 million words should not be added to the 253 million to create a single measure of finished translation. GCtranslate volume represents text processed through an internal tool, while the professional total represents work translated by Bureau professionals. Some machine output may have been used only for comprehension, revised elsewhere or abandoned.

Even with that caveat, the scale is striking. A protected internal machine-translation service handled more words in a little over a year than the Bureau reported as professionally translated during its latest fiscal year.

Human production and staffing moved lower

The previous year’s official data reported approximately 325 million professionally translated words. The latest total of 253 million is about 22 percent lower. Translation for Parliament fell from 59.7 million to 43 million words, while work for departments and agencies declined from 262.7 million to 210 million.

The workforce also contracted. Total employment declined from 1,269 to 1,172, a decrease of 7.6 percent. The number of translators moved from 730 to 683, down 6.4 percent. Interpreters increased slightly from 73 to 75, and terminologists remained at 31.

Financial activity followed the lower-volume pattern. Revolving-fund revenue declined from $166.7 million to $141.6 million, or about 15 percent, while expenditures fell from $163.1 million to $146.3 million. These calculations use the Bureau’s published rounded totals, so percentage changes are approximate.

Interpretation and accessibility also changed

The Bureau delivered more than 37,800 hours of interpretation in 2025-2026, compared with 42,700 hours a year earlier, a decline of about 11.5 percent. Closed-captioning hours fell from 7,216 to 5,926. Sign-language interpretation, tactile interpretation and oral transliteration fell from 11,673 to 8,325 hours.

One area moved in the opposite direction. Indigenous-language interpretation increased from 320 to 668 hours, more than doubling, even though the number of reported assignments fell from 20 to five. This suggests that a small number of engagements may have been substantially longer, although the published data do not provide enough detail to confirm the reason.

The Bureau also processed more than 1,150 Indigenous-language translation requests totaling 1.2 million words. That work should not be lost in a debate focused only on English-French automation.

What the figures do and do not show

The data do not establish that 380 million words of professional translation were replaced. They do show that federal users have adopted machine translation at considerable scale and that professional volume fell during the same broad period. Election-related parliamentary activity, departmental demand, procurement choices, staffing and productivity may all affect the annual comparison.

For Canadian LSPs, the commercial signal is clear. High-volume, low-risk comprehension work is becoming easier for clients to handle internally. External providers will need to demonstrate value where accountability, domain expertise, accessibility, Indigenous languages, publication quality and defensible review matter. The most useful future metric would be the share of GCtranslate output that receives human revision or enters public-facing material. The Bureau does not currently publish that breakdown.