The document presents a condition under which unbounded unions of languages can be learned from positive data using refinement operators. Specifically, it introduces two theorems:
1) Theorem 1 states that a concept class (C,R,L) is learnable if it admits a refinement operator satisfying properties [A-1] to [A-3].
2) Theorem 2 (the contribution of the paper) states that the union concept class (C*,R*,L) is learnable if (C,R,L) admits a refinement operator satisfying [A-1] to [A-3] and additional properties [C-1] and [C-2]. This allows learning of unbounded unions of languages.