Being able to parse code-switched (CS) utterances, such as Spanish+English or\nHindi+English, is essential to democratize task-oriented semantic parsing\nsystems for certain locales. In this work, we focus on Spanglish\n(Spanish+English) and release a dataset, CSTOP, containing 5800 CS utterances\nalongside their semantic parses. We examine the CS generalizability of various\nCross-lingual (XL) models and exhibit the advantage of pre-trained XL language\nmodels when data for only one language is present. As such, we focus on\nimproving the pre-trained models for the case when only English corpus\nalongside either zero or a few CS training instances are available. We propose\ntwo data augmentation methods for the zero-shot and the few-shot settings:\nfine-tune using translate-and-align and augment using a generation model\nfollowed by match-and-filter. Combining the few-shot setting with the above\nimprovements decreases the initial 30-point accuracy gap between the zero-shot\nand the full-data settings by two thirds.\n