Experimental paradigms
Separating world knowledge from factivity
A key challenge in studying factivity is that our judgments about whether someone is certain of a proposition depend on two factors:
- The properties of the predicate (whether speakers interpret it as being associated with a presupposition)
- Our world knowledge about how likely the proposition is
Degen and Tonhauser (2021) developed a two-stage experimental design to tease these apart:
Stage 1: Norming study (Experiment 2a)
First, they measured how world knowledge affects belief in various propositions. Participants saw items like:
Fact: Sophia is a hipster.
Question: How likely is it that Sophia got a tattoo?
Response: [slider from 0 to 1]
For each proposition (e.g., Sophia got a tattoo), they created two contexts–one making it more likely (Sophia is a hipster) and one making it less likely (Sophia is a Mormon). This gives us a measure of world knowledge independent of any embedding predicate.
Stage 2: Projection study (Experiment 2b)
Next, they embedded these same propositions under various predicates and asked about the speaker’s certainty:
Context: Isabella said that Sophia is a hipster.
Utterance: Noah knows that Sophia got a tattoo.
Question: Is Noah certain that Sophia got a tattoo?
Response: [slider from 0 to 1]
They reasoned that, if speakers categorically interpret know as triggering a factive presupposition, Noah should be certain regardless of the prior probability; but, they reasoned, if projection is gradient, his certainty should vary with world knowledge.
Key experimental details
The experiments tested 20 predicates that theory suggests might pattern differently:
- Canonical factives: be annoyed, discover, know, reveal, see
- Non-factives:
- Non-veridical non-factive: pretend, say, suggest, think
- Veridical non-factive: be right, demonstrate
- Optionally factive: acknowledge, admit, announce, confess, confirm, establish, hear, inform, prove
Each participant saw 20 items (one per predicate) plus controls, with predicates paired with different complement clauses and prior contexts across participants.