How Moncton is using AI to deliver authentic bilingual communications at scale

Moncton, a city in southeastern New Brunswick, Canada, used its participation in the City Data Alliance (CDA) to strengthen its capacity to deliver clear and culturally accurate communications in both English and French. The city developed and tested an AI‑assisted translation and response prototype tuned to Moncton’s French dialect, with humans reviewing all outputs. Early testing showed measurable improvement over a baseline model, validating the potential of fine‑tuning with local data. Alongside the prototype, Moncton advanced workforce training, clarified AI principles, and mapped risks and mitigations, laying the foundations for responsible scale‑up.

Providing timely, high‑quality bilingual service is central to equity and trust in Moncton. Residents depend on the city for clear information in both English and French. At the same time, bilingual communication relies on a small group of specialized staff and manual translation workflows, creating bottlenecks as demand grows.

Mayor Dawn Arnold set a clear objective: give residents confidence that they will receive accurate information in their language, even as demand grows. This meant designing AI tools that assist staff rather than replace them, clearly disclosing AI use to residents, and ensuring outputs reflect Moncton’s linguistic and cultural context.

Moncton’s work built on capabilities strengthened through the CDA and the city’s broader data strategy, including emerging governance structures, data warehouse development, and growing cross‑departmental collaboration. With these foundations in place, the city focused on a resident‑facing service area with clear equity implications. The approach centered on three priorities:

1. Human-in-the-loop bilingual drafting: The city developed a prototype that supports resident inquiries in English and French, using AI to draft responses while staff retain authority to review, edit, or regenerate content before sending. The system incorporates Moncton-specific knowledge, tone and style matching, explicit AI labeling, source citations, and feedback mechanisms. Fine-tuning with local French data improved alignment with Moncton’s dialect, reinforcing the importance of local language data and human oversight.
2. Governance and risk mitigation by design: Alongside the prototype, Moncton articulated a clear AI vision and values centered on transparency, equity, privacy, and accountability. The city conducted risk mapping focused on mistranslation, offensive language, and misinformation, and embedded safeguards such as AI disclosures, standard response patterns, and escalation paths. Rather than creating parallel structures, Moncton plans to integrate AI directives into existing privacy, security, and data governance frameworks.
3. Workforce and data readiness for scale: Moncton secured GovAI licensing for staff, conducted training in 2025, and began assessing internal capacity needs. The team identified challenges in accessing high‑quality Moncton‑French documents and started systematizing document flows into the data warehouse to support ongoing model improvement and reproducible evaluation.

Moncton’s early work shows how bilingual communication can be strengthened at scale without sacrificing human oversight:

• Improved translation quality: Fine‑tuned models produced measurable gains, with quality scores rising from 88.7% to 92.1% and broader tests stabilizing near 92.7%.
• Human-centered safeguards: Staff remain in control of final outputs, and resident‑facing responses include AI disclosure and source citations.
• Stronger readiness for responsible adoption: Citywide licensing, training plans, and cross-departmental engagement shifted AI from isolated experimentation to an organizational capability.

Moncton now has a tested prototype, a clearer understanding of data requirements, and a roadmap to embed AI within its broader digital and data strategy.

Moncton’s experience shows that AI can strengthen bilingual service delivery when it supports human expertise rather than replacing it. Data quality and local relevance determine whether AI adds value or risk, especially in a multilingual environment where tone and nuance matter. Governance and transparency must be established before scaling tools, not after, ensuring that adoption is safe, consistent, and trusted by staff and residents. Finally, building workforce capacity and shared understanding turns pilots into repeatable practice, helping Moncton use AI to reinforce clear communication and resident confidence.