The neural network doesn’t understand ingredients

As the neural network learns to generate recipes, it’s very clear from the ingredients it produces that it has no clue what these words actually mean, or whether the substances involved are powders, liquids, solids, etc. It just knows that certain kinds of words tend to go after other certain kinds of words. Some of my favorites so far:

½ cup unbleached diced cold water
2 inch soup pot whipped to taste
2 eggs, separated – slightly
1 tablespoon grated fresh soy sauce
1 1 ½ cup crushed water
1 cup shelled whipping cream
½ cup sifted milk
1 tablespoon drained cloves
1 rosemary, peeled
2 cup cubed instant and round flour
½ cup packed sweet sherry
2 tablespoon sherry; julienned
2 squares raisins
½ cup flaked lemon juice
1 tablespoon whipping milk
1 tablespoon minced fresh milk
1 nuts, seeded and diced

My very favorite ingredient, though, found by the neural network that inspired mine, is the shredded bourbon.