The statement view of testimony

The statement view, defended by Jennifer Lackey, holds that testimonial knowledge is acquired from what speakers state rather than from what they believe, so that a hearer can learn from a statement its speaker does not believe. It is Lackey’s alternative to the belief view, on which testimony transmits a speaker’s belief or knowledge, and it is the most permissive of the leading accounts on the speaker’s side.

Lackey’s central cases attack the idea that testimony is transmission. Stella, a devout creationist teaching high-school biology, has mastered the evidence for evolution, teaches it competently, and does not believe it; her students, Lackey argues, come to know on her say-so. If they do, a speaker need not know, or even believe, what she tells. In place of belief Lackey requires that the statement be a reliable indication of the truth and that the hearer have no undefeated defeaters; her fuller account adds positive reasons on the hearer’s side.

I grant Lackey her cases (see selfless assertion). But what the statement view puts in place of belief is itself a truth-connection. Reliable indication is a covariation between statement and truth, so even the most permissive account retains a substantive constraint between production and truth, and entails the Tracking Minimum.

If any contemporary account admitted language models as testifiers, it would be this one, since its project is to weaken the speaker side. I argue that its own materials draw the line that excludes them. Apply an intervention to Stella: plant in her textbook an error of a kind her competence can detect, and she does not assert it; vary the evidence within the reach of what she can assess, and her assertions follow. Her production tracks the evidence through her own competence, so the textbook does not screen it off. Nothing in the current evidence shows that a model’s open-domain production does the same. What Stella has is tracking reliability; what the models show is frequency reliability.