Do Language Models Communicate? Communicative Intent and Reference from a Derridean Perspective

This paper assesses the arguments of Emily M. Bender, Timnit Gebru, Angelina McMillan-Major and Margaret Mitchell in the influential article “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” These arguments disputed that Language Models (LM) can communicate and understand. In particular, I discuss the argument that LMs cannot communicate because their linguistic productions lack communicative intent and are not based on the real world or a model of the real world, which the authors regard as conditions for the possibility of communication and understanding. I argue that the authors’ view of communication and understanding is too restrictive and cannot account for vast instances of communication, not only human-to-human communication but also communications between humans and other entities. More concretely, I maintain that communicative intent is a possible but not necessary condition for communication and understanding, as it is oftentimes absent or unreliable. Communication need not be grounded in the real world in the sense of needing to refer to objects or state of affairs in the real world, because communication can very well be about hypothetical or unreal worlds and object. Drawing on Derrida’s philosophy, I elaborate alternative concepts of communication as the transmission of an operation of demotivation and overwhelming of interpretations with differential forces, and of understanding as the best guess or best interpretation. Based on these concepts, the paper argues that LMs could be said to communicate and understand.