Framing AI in Canadian political documents, 2017–2026.
“How is AI framed in Canadian political discourse, by which actors, and how do these framings transform between 2017 and 2026?” — a supervised machine-learning approach, transposing the CCF methodology to the federal parliamentary and strategic corpus.
2017: a turning point.
The Pan-Canadian AI Strategy marks the first time the federal government treats AI as an object of public policy. It is the anchor from which the corpus is built — the equivalent of the CCF project’s “1978” — and the reference point against which the evolution of framings is measured.
-
2017
Pan-Canadian Artificial Intelligence Strategy
CIFAR / ISED. The founding document: AI becomes a federal policy object.
-
2019
Directive on Automated Decision-Making
Treasury Board rules for algorithmic decisions within the federal administration.
-
2022
AI Strategic Plan · AIDA (Bill C-27)
First attempt at a Canadian legislative framework for artificial intelligence.
-
2024
National AI Strategy consultation
Public consultation and briefs feeding into the future national strategy.
-
2026
National strategy “AI for All”
The closing milestone of the period under study.
A four-phase computational pipeline.
The pipeline is taken directly from the CCF methodology and transposed to the scale of the Canadian corpus. Validation relies on intercoder agreement — Cohen’s κ and Gwet’s AC1.
Corpus construction
Collection of the sources and sentence-level segmentation with spaCy; structured storage in PostgreSQL.
Semi-manual annotation
A gold standard of 500–1,000 passages, annotated with LLM-assisted tooling and Doccano.
Supervised classification
BERT (EN) and CamemBERT (FR) trained on the gold-standard sample.
Diachronic analysis
Quantified measurement of the shift in dominant framings across 2017–2026.
A corpus in five strata.
The corpus is hypothetical at this stage — volumes remain to be confirmed, and passages genuinely related to AI will be filtered beforehand.
Parliamentary debates
Main corpus. Hansard — House, committees (INDU, ETHI, SECU, RNNR), Senate · 50,000–200,000 passages.
ourcommons.ca · parl.ca · XMLStrategic documents
2017 Strategy, 2022 Plan, Directive on Automated Decision-Making, AIDA (Bill C-27), “AI for All” 2026.
canada.ca · ised-isde.canada.caConsultations & briefs
AIDA, 2024 National Strategy consultation, briefs submitted to committees.
consultations · canada.caSpeeches & addresses
Speeches from the Throne, federal budgets, ministerial addresses (ISED, TBS, PS).
canada.ca · House archivesAdvisory reports
AI Advisory Council, Office of the Privacy Commissioner, Senate committee reports.
lop.parl.caThe annotation scheme.
The first AI × politics annotation scheme adapted to the bilingual Canadian context. Each sentence can carry multiple labels: thematic frames, messengers and evaluative stances.
Artificial intelligence represents a major economic opportunityEconomic · Opportunity for Canada, but we must protect privacyEthics and prevent algorithmic discriminationRisk as well as disinformationSecurity.
The analyses this corpus makes possible.
Beyond the when: the corpus captures who frames AI, how, why and with what effects.
Actors & framings
Who speaks and how: Liberals vs Conservatives, ISED vs Treasury Board, Mila vs Microsoft.
Authority networks
NER and co-citations: Bengio framed as “ethics”, OpenAI as “threat” or “opportunity”.
Federal geography
Framings by province and institution; jurisdictional tensions over AI.
Agenda-setting
Trigger events and attention peaks: the ChatGPT effect in the House.
Priority at the top
Which frames reach the Speech from the Throne and the federal budget.
Inter-party comparison
NDP / labour, Bloc / sovereignty, Conservatives / competitiveness.
Discursive polarisation
Is AI becoming divisive like climate, or does it remain consensual?
Discursive cascades
Tracing a framing from brief to committee, to minister, then to bill.
Contributions and next steps.
What the thesis brings
The first AI × politics annotation scheme adapted to the bilingual Canadian context · a quantified measurement of the shift in framings, 2017 → 2026 · an open, reusable computational corpus.
Roadmap
01 Build the Hansard corpus (XML) · 02 Annotate the gold standard (Doccano) · 03 Train and validate the classifiers · 04 Analyse the diachronic trajectory.