We study two multiple-access scenarios with encoders that are informed only intermittently. The first is the Gaussian multiple-access channel with an intermittent feedback link. Here we assume that, depending on the current binary state which evolves in a memoryless fashion, the previous channel output is either revealed to the two encoders or not. For this scenario we obtain an outer bound on the capacity region that approaches the capacity region without feedback when the probability that the channel output will be fed back approaches zero. We also propose an inner bound that converges to the capacity region with ideal feedback and the capacity region with no feedback in the associated extreme cases. In the second scenario the encoders always observe ideal feedback, and in addition they can crib intermittently. For this scenario we establish the capacity region for the special class of semi-deterministic multiple-access channels. The capacity is achieved using the Superposition Block Markov Coding technique of Cover and Leung. For both scenarios the outer bounds are tighter than those obtained by revealing the underlying state sequence non-causally to the encoders.