Linguistics Department Colloquium (please note special time and place: Tue. March 6, 5pm, Merrill 315):

Andreas Haida, Hebrew University of Jerusalem

“Logical Reasoning and Scalar Inference Computation” (joint work with Luka Crnič and Yosef Grodzinsky)

Abstract:

It seems that logical reasoning is a necessary component of human activities such as science, engineering, and legislature. However, psychological studies found failure rates of up to 80% and more when subjects perform logical reasoning tasks such as forming (in)validity judgments about the arguments in (1) and (2), where only (1) is valid in Aristotelian logic.

(1) All desires are sins (p1) and some beliefs are desires (p2)
Therefore: some beliefs are sins (c) (valid)
(2) All desires are sins (p1) and some beliefs are sins (p3)
Therefore: some beliefs are not desires (d) (invalid)

These findings have often been taken to show that human reasoning is illogical. Other researchers concluded that reasoning is logic-based but also involves non-logical inferential methods such as scalar-inference (SI) computation, which yields the logically invalid inferences in (3).

(3) a. some beliefs are sins (p3) ↝ not all beliefs are sins (s1)
b. some beliefs are not desires (d) ↝ not all beliefs are not desires (s2)

Recent studies suggest that there is individual variation in the use of SIs. We identify five ‘SI profiles’ of logical arguments such as the profile of (2): the premises p1 and p3 validate the conclusion d if and only if the SI of p3 is taken into account and the SI of d is supressed (that is, p1 ∧ p3 ∧ s1 ⇒ d but p1 ∧ p3 ∧ s1 ⇏ d ∧ s2). In a validity-rating experiment, we tested for individual variation in the response to the five SI profiles. Our data show, for instance, that there are subjects that systematically accept arguments with the profile of (2) but reject logically invalid arguments that are inert to SI computation. Thus, they compute SIs for premises but not for conclusions. Overall, the data show that a sizable proportion of ‘logical failures’ are in fact successfully performed logical inferences on the basis of sentence meanings that are enriched by SIs. I will discuss how we may explain the existence of different groups of reasoners and the preference to perform SI computations in premises over conclusions.