Nuhu Osman Attah

This dissertation is a collection of three independent but thematically connected papers on language models. The first assesses a currently popular account of the nature of the linguistic competence ostensibly manifested by LMs. According to this account LMs are “stochastic parrots”, whose linguistic outputs, though they look outwardly meaningful, are actually only cleverly stitched together bits and pieces of language which have no meaning at all except accidentally. This claim depends on what I call the Communicative Intention argument. In this chapter I demonstrate that the argument is largely driven by strong Gricean assumptions about the nature of linguistic competence that I argue are now known to be empirically untenable. Moreover, I show that once we give up these assumptions and fairly assess the architecture of both LLM and human cognitive architectures, we will find no good reasons to suppose large language models lack communicative intentions. The second paper addresses the larger debate in the context of which the Communicative Intention argument is made: the debate about meaning in LMs. I argue here that there are two senses of “meaning” in the debate, which I term semantic meaning and content meaning. A failure to distinguish between these two varieties of meaning has led to many of the characteristic confusions in the debate. The final paper deals with metalinguistic capacities in LMs. This paper introduces a novel way of accounting for any ostensible differences in the competences manifested by humans and LMs. The account has to do with what I call metalinguistic agency — roughly, the claim is that the agential structure of the linguistic competences in LMs and humans differ; any ostensible differences between them can be accounted for in terms of this difference in agential structure.

In some recent work, some psychologists, linguists, AI researchers, and philosophers (e.g., Shanahan, 2022; Bender & Koller, 2020; Montemayor, 2021; Bender et al., 2021) have argued that, despite producing convincing human-like linguistic output, large language models do not possess linguistic competence on the ground that they lack communicative intention. Among the proponents of this position, the notion of communicative intention is entertained as the liveliest candidate for a distinguishing characteristic of human cognition vis-a-vis LLMs and as such as at least one of the features which accounts for the characteristic mentality of humans. I demonstrate that such a position is largely driven by strong Gricean assumptions about the nature of linguistic competence that I argue is now known to be empirically untenable. Moreover, I show that once we give up these assumptions and fairly assess the architecture of both LLM and human cognitive architectures, we will find no good reasons to suppose large language models lack communicative intentions. I thus reject communicative intentions as a viable ground on which to understand the difference between LLMs and humans and thus as a characteristic mark of human mentality.