Feature ambiguity (source code)

= Feature ambiguity

A property of a discrimination in which an individual feature is compatible with more than one response, while combinations can disambiguate the stimulus. For example, <learning> that $AB$ and $CD$ are rewarded but $BC$ and $AD$ are not cannot be solved by assigning one fixed value to each feature alone. To prove the limitation for an additive threshold rule, suppose $w_A+w_B>c$ and $w_C+w_D>c$, while $w_B+w_C\leq c$ and $w_A+w_D\leq c$. Summing the first pair gives a total greater than $2c$, while summing the second gives the same total at most $2c$, a contradiction. Conjunctive representations can distinguish the pairs without this restriction.