Description
Identifies what users are likely to infer beyond the literal content of interface messages, checking for unintended implications that mislead decisions.
Implicature is meaning communicated beyond what is literally said, inferred from the assumption that the speaker is cooperative. 'Some of your files were backed up' implies not all; 'Your password was changed recently' may imply a security problem. Levinson identified generalised implicatures that arise by default, such as scalar inferences. Interfaces generate implicatures constantly, often unintended: an error message that blames a component implies the user did nothing wrong; a confirmation without a warning implies there are no consequences. An implicature audit lists each message's literal content and probable inferences, checks them against truth and intent, and tests with users. Rigour comes from eliciting inferences from users, not only expert judgement, and focusing on decision-relevant implications.
Attributes
- Discipline
- Linguistics, Semiotics & Discourse
- Loss type
- Unmigrated
- Stage
- Design, Evaluation
- Data
- Mixed
- Effort
- Days
Steps
- Select messages that inform decisions: errors, warnings, status, confirmations, recommendations and AI answers.
- For each, write the literal content and expert-predicted implicatures (scalar, relevance, manner-based).
- Elicit actual inferences from 10-20 users with questions such as 'What does this tell you?' and 'What would you do next?'
- Code inferences as intended, unintended but harmless, or unintended and harmful; flag any harmful inference held by more than 20% of users.
- Rewrite to cancel harmful implicatures explicitly or change what is said.
- Re-test rewritten messages with new participants.
Sources
- Gleason, J. B., The Development of Language
- Hall, E., Conversational Design (2018)
- Gopinath, V., Methods Atlas (Gricean audit)
- Development of Language (Gleason)
- Conversational Design (Hall)
- Methods Atlas — Gricean audit (Gopinath)
- The Inquiry Series, Ch 18 — Pragmatics — What Language Actually Does
Also known as
- Implicature analysis
- Unintended inference audit
Categories
- Content
- Linguistics, Semiotics & Discourse