Sopa (the Spanish for "soup") is a game about a young boy who goes to fetch a potato for his grandma, then stumbles upon a magical world at the back of the food cupboard. "The pantry seems to get longer and longer," explains creative director Juan Castañeda. "And when you're about to grab the sack of potatoes, you get pulled into this other world of fantasy and magical realism. So you go on all these adventures, and meet all these different characters, but at the end of the day, you're really just trying to get that potato for your grandma's soup." As video game quests go, this is fabulously mundane and makes a refreshing change from rescuing princesses in castles and saving lands in peril.

Recurrent and convolutional neural networks comprise two distinct families of models that have proven to be useful for encoding natural language utterances. In this paper we present SoPa, a new model that aims to bridge these two approaches. SoPa combines neural representation learning with weighted finite-state automata (WFSAs) to learn a soft version of traditional surface patterns. We show that SoPa is an extension of a one-layer CNN, and that such CNNs are equivalent to a restricted version of SoPa, and accordingly, to a restricted form of WFSA. Empirically, on three text classification tasks, SoPa is comparable or better than both a BiLSTM (RNN) baseline and a CNN baseline, and is particularly useful in small data settings.