Canada's commitment to bilingualism in government services assumes semantic equivalence across English and French. This study tests that assumption by analyzing 47 Algorithmic Impact Assessments (AIAs) using computational text analysis and LLM-powered semantic matching.

We discovered three forms of translation failure. First, discoverability gaps: while 38% of AI systems have bilingual documentation, inconsistent title translation makes French versions effectively invisible (initial analysis detected only 4%). Second, availability gaps: 23% of systems remain genuinely monolingual, violating Official Languages Act requirements. Third, semantic drift: when bilingual versions exist, governance concepts transform across languages.

Analyzing 18 bilingual pairs, we found systematic divergences in key governance terms. English versions emphasize "oversight" and "human judgment," while French versions use "surveillance" and emphasize "client explanation." The term "accountability" appears 8 times in English documents but 0 times in French, suggesting fundamental differences in how responsibility is framed. Data input descriptions in English provide legal justification ("lawfully collected"), while French versions offer only technical descriptions ("audio recordings").

These divergences reveal that translation is not neutral transmission but governance transformation. Following Venuti's translation politics and Baker's semantic non-equivalence, we argue that bilingualism in algorithmic governance creates differential citizenship—where language determines not just access to information, but the very meaning of accountability, transparency, and oversight. We conclude that Canada's AI governance suffers from an "untranslatability" problem requiring not just better translation, but conceptual alignment across languages.

The Canadian aporetic condition names a patterned field of public contradiction produced when shared civic terms carry incompatible obligations across institutions, publics, and historical inheritances. This paper advances that framework by placing it in direct dialogue with Barbara Cassin’s account of the “untranslatable,” defined as terms and syntactical turns that remain translatable in practice while repeatedly generating a problem because conceptual networks do not superimpose across languages, histories, and idioms (Cassin, 2014, p. xvii). Cassin’s definition matters for the aporetic condition because it identifies a recurring mechanism: translation persists, and the trouble persists with it. The paper’s central claim is procedural: within Digital Humanities, translation includes digitization, modelling, classification, and editorialization. When civic terms enter databases, metadata fields, controlled vocabularies, interface labels, search facets, and language technologies, institutional choices stabilize some senses, compress others, and set an interpretive tempo for users. These translation layers operate as public infrastructure. They shape what becomes searchable, comparable, and teachable, and they distribute public standing by determining which senses appear authoritative within public-facing systems. Methodologically, the paper proposes an “aporetic lexicon” workflow for Digital Humanities research on public institutions. The workflow assembles a curated corpus of Canadian policy documents, parliamentary debates, court decisions, and institutional public records that already function as translation interfaces, including classification schemes, metadata templates, access notices, bilingual terminology guidance, and interface copy. Computational assistance supports concordance mapping, collocation analysis, and diachronic drift tracking for key terms. Close reading then treats each candidate term as a case, in Cassin’s sense, and reconstructs the institutional settings in which translation choices acquire governance force. The analysis highlights three clusters of untranslatability that concentrate Canadian aporias. First, sovereignty vocabularies linked to Crown and nation generate recurring disagreements over authorization, jurisdiction, and representation. Second, rights vocabularies linked to treaty, recognition, and reconciliation generate recurring disagreements over the status of obligation and the meaning of consent, particularly as terms move across legal, administrative, and public communication contexts. Third, pluralist vocabularies linked to multiculturalism, public, and security generate recurring disagreements over belonging, risk, and the boundaries of legitimate public protection. The conclusion develops counter-policy as a response to untranslatability in DH: design moves that keep contested terms legible as contested through documentation, multilingual annotation, plural metadata strategies, and transparent revision histories that preserve ambiguity as public evidence. The paper advances the Canadian aporetic condition by specifying a mechanism of persistence—governance contradictions travelling through translation layers embedded in digital infrastructure—and offers DH practitioners practical criteria for building public-facing systems that remain accountable to linguistic and civic plurality.

This paper argues that meme humour represents a form of cultural expression that resists seamless translation into universal digital, linguistic or computational frameworks. While memes circulate on global platforms and appear visually accessible, their humour depends on culturally embedded knowledge, shared social memory and affective familiarity that cannot be fully understood outside specific communities. Drawing from examples of linguistically hybrid memes (eg. code-switching, slang, Hinglish), culturally situated jokes, family structures, education systems, and gender norms. Moreover, this paper shows that meme humour operates through recognition rather than explanation. One must already inhabit the cultural world, the meme vernaculars, in other words, “you had to be there”.

This paper positions humour as a site of cultural untranslatability, where meaning exceeds what can be rendered into metadata, keywords or algorithmic classification. While digital platforms process memes as images, text strings or engagement metrics, they cannot access social memory, lived experiences and cultural codes that generate humour. Building on Limor Shifman’s understanding of memes as participatory cultural units and Ryan Milner’s concept of memetic resonance, this paper treats memes not as isolated jokes but as socially situated communicative practices that depend on shared frameworks of interpretation. Roland Barthes’ semiotic theory helps explain how everyday signs become naturalised cultural meanings, while Stuart Hall’s encoding/decoding model illuminates how meme humour invites negotiated and oppositional readings across audiences.

Methodologically, the paper combines semiotic analysis with digital ethnography. Rather than analysing memes as detached texts, it examines them within their circulation contexts: comments, shares, remix practices and audience interaction patterns on social media platforms. This approach allows the study to observe how humour is collectively recognised, contested or misunderstood, revealing the boundaries of cultural belonging. Moments where humour fails, that is where users ask for explanations or misread jokes become significant, as they mark points where translation breaks down.

By framing meme humour as culturally untranslatable, this paper contributes to debates in digital humanities and platform studies about the limits of universalising models of communication. Memes demonstrate that meaning online is not only encoded in text or image but in shared cultures that cannot be fully rendered into data.