2021/06/11 by Saujas Vaduguru, Aalok Sathe, Vaduguru, Saujas +5 · 1 citation
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Software Engineering Research #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2106.06566
openalex publication_date 2021/06/11 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Neural models excel at extracting statistical patterns from large amounts of\ndata, but struggle to learn patterns or reason about language from only a few\nexamples. In this paper, we ask: Can we learn explicit rules that generalize\nwell from only a few examples? We explore this question using program\nsynthesis. We develop a synthesis model to learn phonology rules as programs in\na domain-specific language. We test the ability of our models to generalize\nfrom few training examples using our new dataset of problems from the\nLinguistics Olympiad, a challenging set of tasks that require strong linguistic\nreasoning ability. In addition to being highly sample-efficient, our approach\ngenerates human-readable programs, and allows control over the generalizability\nof the learnt programs.\n