The Singleton Fallacy: Why Current Critiques of Language Models Miss the Point

This paper discusses the current critique against neural network-based\nNatural Language Understanding (NLU) solutions known as language models. We\nargue that much of the current debate rests on an argumentation error that we\nwill refer to as the singleton fallacy: the assumption that language, meaning,\nand understanding are single and uniform phenomena that are unobtainable by\n(current) language models. By contrast, we will argue that there are many\ndifferent types of language use, meaning and understanding, and that (current)\nlanguage models are build with the explicit purpose of acquiring and\nrepresenting one type of structural understanding of language. We will argue\nthat such structural understanding may cover several different modalities, and\nas such can handle several different types of meaning. Our position is that we\ncurrently see no theoretical reason why such structural knowledge would be\ninsufficient to count as "real" understanding.\n