Macho: Writing Programs with Natural Language and Examples

Current natural language programming systems avoid the difficulties of dealing with abstract and ambiguous concepts by restricting the input words to those comparable to a normal high-level programming language. Our system, Macho, can write programs from significantly more abstract language by asking the programmer to provide a unit test: one or more examples of correct input and output. This may seem unnecessarily complicated, but we show that natural language and examples have a surprising synergy both in constraining the ambiguity of the specification and in generating correct solutions.published or submitted for publicationnot peer reviewe