I was reading Graham's On Lisp which works a nice example of Augmented Transition Networks which were popular in the 1970s for writing parsers that could parse controlled vocabularies. For that matter I remember similar kinds of grammar to control voice response applications for platforms like TellMe circa 2001. I bet it would do fine for food ordering.

(This is why a non-native speaker can do this job well, you don't need to know a lot of the language to parse orders like 芥兰牛肉)