Fintan Mallory

This paper applies a consumer-based teleosemantic framework to give a detailed analysis of a particular algorithm for generating word embeddings. In the process, it addresses several of the challenges facing teleosemantic approaches to artificial neural networks. It distinguishes the content of embeddings from the contents of their features and argues that word embeddings should be seen as pushmi-pullyu representations whose descriptive contents are facts about the contexts in which words appear. Furthermore, it argues that embeddings meet Millikan’s conditions for iconic representation as transformations between embeddings correspond to transformations of the contents of their representations.

The ability to control the direction of a conversation, which topics are raised, which questions are asked, and which lines of inquiry are followed, is a basic and powerful form of social control that has been under studied within the philosophy of language. This paper draws on work from formal pragmatics, social epistemology, and critical discourse analysis to identify the mechanisms by which social power influences who gets to set the question under discussion. In the process, it introduces the category of inquisitive injustice and identifies several underlying structural causes behind it. By making these mechanisms explicit, the paper aims to empower speakers to challenge existing but subtle forms of manipulation.

This paper argues for a pluralist approach to representation in large language models. There are two parts to this pluralism, the first is that we should recognise more than one vehicle of representation in transformer models. Call this vehicle pluralism. Rather than identifying the vehicles of representation with a single component of a system, e.g. individual neurons, patterns of activation, regions in the activation space, we should acknowledge multiple systems of representation within a network operating with different vehicles. The second claim is that we should recognise that there are different formats of representation in transformer models. Transformer models do not operate with a purely analogue, structural, or symbolic architecture but are a hybrid system of representation. Finally, I will discuss how this relates to several working hypotheses about representation that have become adopted in the field of mechanistic interpretability.

The ability to control the direction of a conversation, which topics are raised, which questions are asked, and which lines of inquiry are followed, is a basic and powerful form of social control. This paper analyses this phenomenon in relation to a category of discursive injustice called ‘Inquisitive Injustice’. Inquisitive injustice concerns a speaker’s ability to make their questions the question under discussion in a conversation. The paper presents examples of this injustice, identifies the mechanisms by which it occurs, and several causal factors giving rise to its occurrence. In the process, it draws connections between the fields of social philosophy, formal pragmatics and critical discourse analysis.

Generative linguistics is widely claimed to produce theories at the level of computation in the sense outlined by David Marr. Marr even used generative grammar as an example of a computational level theory. At this level, a theory specifi es a function for mapping one kind of information into another. How this function is computed is then specifi ed at the algorithmic level before an account of how this is algorithm is realised by some physical system is presented at the implementation level. This paper will argue that generative linguistics does not fit anywhere within this framework. We will then look at several ways researchers have attempted to modify either the framework of generative theory to reconcile the two approaches. Finally, it presents and discusses an alternative position, anti-realism about generative grammar. While this position has attracted some recent support, it also runs into some of the problems that earlier modifications faced.

For billions of people, the internet has become a second home. It is where we meet friends and strangers, where we organise and learn, debate, deceive, and do business. In some respects, it is like the town square it was once claimed to be, while in others, it provides a strange new mode of interaction whose influence on us we are yet to understand. This collection of papers aims to give a short indication of some of the exciting philosophical work being carried out at the moment that addresses the novel aspects of online communication. The topics range from the expressive functions of emoji to the oppressive powers of search engines.

Generative linguistics is widely claimed to produce theories at the level of computation in the sense outlined by David Marr. Marr even used generative grammar as an example of a computational level theory. At this level, a theory specifi es a function for mapping one kind of information into another. How this function is computed is then specified at the algorithmic level before an account of how this is algorithm is realised by some physical system is presented at the implementation level. This paper will argue that generative linguistics does not fit anywhere within this framework. We will then look at several ways researchers have attempted to modify either the framework of generative theory to reconcile the two approaches. Finally, it presents and discusses an alternative position, anti-realism about generative grammar. While this position has attracted some recent support, it also runs into some of the problems that earlier modifications faced.

According to widely accepted views in metasemantics, the outputs of chatbots and other artificial text generators should be meaningless. They aren’t produced with communicative intentions and the systems producing them are not following linguistic conventions. Nevertheless, chatbots have assumed roles in customer service and healthcare, they are spreading information and disinformation and, in some cases, it may be more rational to trust the outputs of bots than those of our fellow human beings. To account for the epistemic role of chatbots in our society, we need to reconcile these observations. This paper argues that our engagement with chatbots should be understood as a form of prop-oriented make-believe; the outputs of chatbots are literally meaningless but fictionally meaningful. With the make-believe approach, we can understand how chatbots can provide us with knowledge of the world through quasi-testimony while preserving our metasemantic theories. This account also helps to connect the study of chatbots with the epistemology of scientific instruments.

A familiar argument goes as follows: natural languages have infinitely many sentences, finite representation of infinite sets requires recursion; therefore any adequate account of linguistic competence will require some kind of recursive device. The first part of this paper argues that this argument is not convincing. The second part argues that it was not the original reason recursive devices were introduced into generative linguistics. The real basis for the use of recursive devices stems from a deeper philosophical concern; a grammar that functions merely as a metalanguage would not be explanatorily adequate as it would merely push the problem of explaining linguistic competence back to another level. The paper traces this concern from Zellig Harris and Chomsky’s early work in generative linguistics and presents some implications.

The method of linguistic palaeontology has a controversial status within archaeology. According to its defenders, it promises the ability to see into the social and material cultures of prehistoric societies and uncover facts about peoples beyond the reach of archaeology. Its critics see it as essentially flawed and unscientific. Using a particular case-study, the Indo-European homeland problem, this paper attempts to discern the kinds of inference which proponents of linguistic palaeontology make and whether they can be warranted. I conclude that, while the case for linguistic palaeontology has often been overstated, so has the case against it.

ABSTRACT This paper is concerned with the ontological status of linguistic types. According to a widely held view, linguistic types are abstract objects that are instantiated or represented by tokens. The same types might be tokened by both speech, signing and text. This view has implications for how we consider what it is to know a language since knowledge of language is typically taken to be knowledge of linguistic types. We argue below that linguistic types are not abstract objects but action-types and that these action-types are individuated by the capacities that they manifest. The picture that results from this view is that to know a language is to have the capacities required to perform these action-types.

The traditional Turing test appeals to an interrogator's judgement to determine whether or not their interlocutor is an intelligent agent. This paper argues that this kind of asymmetric experimental set-up is inappropriate for tracking a property such as intelligence because intelligence is grounded in part by symmetric relations of recognition between agents. In place, it proposes a reciprocal test which takes into account the judgments of both interrogators and competitors to determine if an agent is intelligent. This form of social interaction better tracks both the evolution of natural intelligence and how the concept of intelligence is actually used within our society. This new test is defended against the criticisms that a proof of intelligence requires a demonstration of self-consciousness and that semantic externalism entails that a non-embodied Turing test is inadequate.