Does linguistic alignment work the same way across cultures?

The 2020–2025 SLR's most useful contribution may be its gap inventory. The reviewers flag four limitations of the literature they synthesize: cultural background, language proficiency, personality, and prior AI experience are known moderators of alignment effects, yet barely studied; outcome measures are unstandardized across studies, blocking meta-analysis; the psychological mechanisms underlying alignment effects are typically asserted rather than measured (eye-tracking, physiological measurement, think-aloud protocols are rare); and domain breadth is uneven, with healthcare, education, and mental-health support called out as particularly under-investigated.

The methodology consequence is non-trivial. Existing claims about "users prefer aligned AI" or "alignment increases trust" are best read as Western-WEIRD-sample local truths until cross-cultural replication arrives. Communication norms vary substantially across cultures — what counts as appropriate accommodation, the politeness-formality calibration, the threshold at which mirroring reads as warmth versus mockery — and an SLR drawn primarily from English-speaking samples cannot adjudicate which findings travel.

This connects to Can AI systems learn social norms without embodied experience?. Models can predict aggregate norm distributions superhumanly, but the alignment literature is asking a different question: whether adapting to a particular interlocutor's idiolect produces consistent effects across cultures. Predictive competence at the population level does not entail that a single alignment policy will read the same way in Seoul, Lagos, and Helsinki.

For writing about conversational AI design, the operational hedge is small but worth deploying: claims of the form "users prefer X" should be qualified by who the users were and where they were sampled. The SLR makes this hedge defensible — it is the literature's own self-assessment, not an external critic's complaint.

The deeper methodology point: an SLR that names its own limitations honestly is more useful than one that papers over them, because the named limitations become the agenda for the next round of work.

Source: Conversation Topics Dialog Paper: Linguistic Alignment in Conversational AI: A Systematic Review of Cognitive-Linguistic Dimensions, Measurements, and User Outcomes (2020–2025)