To what extent are the preferences and biases exhibited by the language model language-dependent, based on its training data in each specific language, versus being language-independent underlying tendencies upon which various languages are superimposed?